Health Data Aggregator · India

India Consumer Health Data Aggregator — Product Research Report and Playbook

Version 1.1 (research edition of Playbook v1.0) · 21 October 2026

Document control and verification status

Item Detail
Purpose Working reference for product, engineering, clinical and compliance teams building a consent-based consumer health data aggregator (“Individual Health Data Aggregator”) in India
Builds on India Consumer Health Data Aggregator Playbook v1.0 (21 Oct 2026). This edition restructures it into the requested 15-section report, adds the source-by-source data map, graph model, journeys and insight engine, and corrects several technical points (see box below).
Audience Founders, PMs, solution architects, ABDM integration engineers, clinical safety officer, DPO/compliance counsel
Citation convention [S#] refers to the Source Register in Appendix A.

Verification status — read before relying on this document. Live web access was not available when this edition was prepared: search, page fetch and general network calls all failed. The one exception was GitHub, which let the team read NHA’s own public repositories (NHA-ABDM/NHCX and NHA-ABDM/ABDM-wrapper). Facts confirmed from those repositories are labelled V (verified in-session from an official NHA source). Other sources are labelled: A = URL supplied in the v1.0 playbook (author-supplied, not re-checked); K = official source named from the research team’s knowledge (URL or document title to be confirmed). Claims that are time-sensitive or uncertain are marked ⚠ Validate in the text. Appendix B lists them as a verification checklist. Before any external, investor, regulatory or compliance use, the checklist must be cleared against the primary source, and legal advice must be obtained on the regulatory items. All [web:N] placeholders from v1.0 have been removed. They were not resolvable citations.

Material corrections to Playbook v1.0 made in this edition

  1. Who plays the AA role. In the Account Aggregator analogy, ABDM’s Consent Manager (HIE-CM) plays the AA role: a consent conduit. The consumer aggregator plays the FIU / personal-finance-app role. It stores and analyses data, so it is a full Data Fiduciary under DPDP, not a “data-blind” conduit (§5.1).
  2. NHCX is a provider↔︎payer exchange. Its participants are hospitals, insurers, TPAs and schemes. NHA’s NHCX participant APIs list only PAYER, PROVIDER and TPA roles [S59]. A consumer app cannot simply “integrate with NHCX” to pull eligibility or claim status. Consumer claims data in the MVP must come from document ingestion and insurer/TPA/employer partnerships. NHCX becomes relevant through a provider-SaaS line, payer partnerships, or the NHCX ABHA↔︎policy linking capability (§3.4, §4.2).
  3. The consumer app should integrate as an ABDM PHR app as well as an HIU. Users discover, link and pull their own records through the PHR / consent-manager-facing flows (purpose self-requested, PATRQT). The HIU consent-request flow is used for other purposes, such as a clinician view (§5.3) ⚠ Validate.
  4. Use ABDM purpose codes and health-information types. Free-text purposes such as chronic_care_management are replaced with ABDM purpose codes (e.g. CAREMGT, PATRQT) and HI types (e.g. DiagnosticReport, Prescription) (§5.4).
  5. Consent approval is asynchronous. The v1.0 pseudocode fetched the consent artifact immediately after the request. In ABDM, approval arrives later by callback. The flow in §10.2–10.3 is corrected.
  6. FHIR coding errors fixed (Appendix D). The discharge summary used LOINC 34117-2 (“History and physical note”); a discharge summary is 18842-5. The lab DiagnosticReport code (comprehensive metabolic panel) did not match its HbA1c result.
  7. Terminology. “PHI” is a US HIPAA term. This edition uses personal health data and health data of a Data Principal in line with DPDP.
  8. Licensing trigger. Policy-renewal or product recommendations may trigger IRDAI intermediary licensing (§9.6, §12, §13).
  9. Provider cohort analytics moved to Phase 2. Cohort analytics need a separate consent purpose and a provider-side legal basis. The MVP offers consumer-initiated, time-bound “share with my doctor” instead (§11).

1. Executive Summary

The problem. An Indian family’s health data is scattered. There are paper and PDF lab reports on WhatsApp, prescriptions photographed on phones, discharge summaries in email, insurance policy schedules and e-cards from employers and retail insurers, TPA claim letters, pharmacy invoices from e-pharmacies, and step and sleep data locked in wearables. Out-of-pocket spending remains a large share of health expenditure: roughly two-fifths in recent National Health Accounts estimates [S41] ⚠ Validate latest figure. The burden of diabetes and hypertension is very large: ICMR-INDIAB estimated about 101 million people with diabetes and about 315 million with hypertension [S40]. People pay for, carry and re-explain their health history at every encounter.

The opportunity. India now has public digital rails that make a consent-first aggregator buildable:

What to build. A consumer-first “Health + Insurance Wallet” that:

  1. joins ABDM as a PHR app + HIU (and optionally an HIP for re-sharing user-uploaded records);
  2. ingests the documents Indians actually hold (lab PDFs, prescriptions, discharge summaries, policy schedules, claim letters, pharmacy invoices) through upload, camera, email-forward and WhatsApp-forward channels with document AI and human-in-the-loop confirmation;
  3. reads wearable data on-device through Google Health Connect and Apple HealthKit, rather than integrating dozens of vendor clouds;
  4. normalizes everything into a longitudinal Individual Health Graph, where every fact carries provenance, consent ID and a trust tier (verified clinical / document-derived / device-measured / user-entered / inferred / reference knowledge);
  5. produces a Health Financial + Clinical Snapshot: timeline, active problems and medicines, care gaps, trends, coverage and exclusions, waiting periods, claims, and net out-of-pocket spend. Insights are framed as decision support, never diagnosis.

What not to build. No scraping of hospital, insurer, pharmacy or app portals. No credential-sharing “screen-scraper” integrations. No unconsented secondary use. No individual-level data sales. No autonomous diagnosis or prescribing.

Top recommendations

# Recommendation Why
1 Lead with document intelligence + ABDM PHR/HIU, not ABDM alone Many Indian records are still paper/PDF, and structured FHIR coverage from HIPs is uneven ⚠ Validate. Documents deliver value on day one.
2 Make insurance clarity the paid wedge (coverage, waiting periods, exclusions, claim tracking, OOP spend) The pain point is acute and money-linked. The IRDAI Customer Information Sheet gives a parseable structure [S18] ⚠ Validate. It is differentiated from pure PHR apps.
3 Design for the family from day one (caregiver/proxy, minors, elderly parents) Indian health decisions are family-led. DPDP has specific rules for children and for persons with disability [S11][S12].
4 Treat consent as a product surface, not a legal page Trust is the moat. DPDP consent must be free, specific, informed, unconditional and unambiguous, with easy withdrawal [S11].
5 Build a clinical safety case before launching any risk flag Telemedicine guidelines bar AI from counselling or prescribing [S21]. Software that drives clinical decisions may be a medical device [S22] ⚠ Validate.
6 Get regulatory opinions early on four points: IRDAI intermediary status, CDSCO SaMD classification, DPDP Significant Data Fiduciary exposure, ABDM HDM Policy retention These decide product scope, not only paperwork.

DPDP timing note. The DPDP Rules 2025 were notified in November 2025 with phased commencement. As understood by the research team, Consent Manager provisions commence about 12 months after notification and most Data Fiduciary obligations about 18 months after (around May 2027) [S12] ⚠ Validate. Build to the full standard now. The MVP will be in market as the obligations take effect.

2. Market Opportunity and Product Thesis

2.1 Demand drivers (India-specific)

Driver Evidence (cite and validate) Product implication
High out-of-pocket (OOP) spending National Health Accounts estimates show OOP at roughly 39–40% of total health expenditure for 2021-22 [S41] ⚠ Validate latest NHA release Spend tracking and “net OOP after reimbursement” are meaningful features, not vanity metrics
Heavy NCD burden ICMR-INDIAB (Lancet Diabetes & Endocrinology, 2023): about 101 M diabetes, about 136 M prediabetes, about 315 M hypertension [S40] Chronic-care tracking (HbA1c, BP, lipids, weight) is the clinical anchor use case
Fragmented, paper-heavy records ABDM exists because records sit with each provider; many reports are still issued as PDF/paper (industry observation) ⚠ Validate with pilot data Document AI is a core capability, not a fallback
Insurance complexity Waiting periods, sub-limits, room-rent caps, co-pays, exclusions; IRDAI 2024 reforms on cashless timelines, PED waiting periods and the Customer Information Sheet [S18][S19] ⚠ Validate A “Am I covered?” explainer and claim tracker address real anxiety
Large public insurance base PM-JAY covers about 12 crore families (about 55 crore people); extended in 2024 to all citizens aged 70+ (Ayushman Vay Vandana) [S43] ⚠ Validate Support PM-JAY entitlement awareness. Do not overpromise data access.
Digital public infrastructure ABHA counts in the high tens of crores and tens of crores of linked records reported by PIB/ABDM dashboards [S14] ⚠ Validate current dashboard figures ABHA linking is a realistic onboarding step for many users. Linked-record depth varies widely.
Smartphone-first, multilingual users 22 scheduled languages; most new internet users prefer Indian languages (industry reports) ⚠ Validate Vernacular UI, voice and read-aloud summaries; Bhashini as a partner option [S58]
Family-led decisions Caregivers (often adult children) manage elderly parents’ care and insurance Proxy/caregiver model with separate consent per family member

2.2 Target users and jobs-to-be-done

Segment Primary job Wedge feature Willingness to pay (hypothesis)
Urban salaried individual (25–45) “Keep my records and my company + personal insurance in one place” Insurance wallet + document vault Low–medium; prefers free with premium add-ons
Family caregiver (managing parents/children) “Know my parents’ medicines, reports, policy and what is covered before an admission” Family profiles, medicine list, pre-admission checklist Highest; buys a family plan
Chronic-condition patient (diabetes, hypertension, thyroid, cardiac) “See whether I’m improving and never miss a test” Trends, care-gap reminders, doctor-share Medium; high retention
Employer HR / benefits platform “Reduce claim friction and support preventive care” Group policy explainer, claims assistance B2B contract
Clinician (consented view) “See the history in 30 seconds” Time-bound patient-shared summary Indirect (drives consumer value)
Insurer / TPA “Get complete documents first time; engage members” Pre-submission document completeness B2B, later phase

2.3 Competitive landscape (positioning, not a market census)

Categories below are described from public positioning known to the research team. Every player’s ABDM integration status, features and traction must be validated against the ABDM integrator list and company sources [S14] ⚠ Validate.

Category Examples What they do well Gap this product fills
Government PHR ABHA app (NHA), Aarogya Setu PHR features ABHA creation, linking, consent approval Limited insights, insurance and family-finance view
Private ABDM PHR apps e.g. Eka Care, DRiefcase, Bajaj Finserv Health (reported ABDM integrations) Record vault, ABHA flows, some vitals Insurance clarity, OOP spend, cross-domain graph
E-health marketplaces Practo, Tata 1mg, PharmEasy, Apollo 24/7 Consults, pharmacy, diagnostics, own-service records Data siloed to own services; commercial bias
Wellness/chronic-care HealthifyMe, Ultrahuman, diabetes programmes (e.g. Sugar.fit, Fitterfly, BeatO) Engagement, coaching, devices Clinical records and insurance absent or thin
Insurance distribution/advice Policybazaar, Ditto, Beshak and others Policy comparison and advisory Not a health record; regulated distribution model
Employer benefits/TPA apps Plum, Loop, Onsurity; TPA apps such as Medi Assist Group e-cards, claim intimation Only their own policy; no clinical view

Positioning statement: “The neutral, consent-first wallet that joins your medical records and your health money, for your whole family.” Neutrality is a design constraint (§2.5, §9.6). Marketplace revenue must never shape clinical or coverage insights.

2.4 Product thesis and core user promise

“One secure place for my family’s health: records, medicines, tests, fitness, diet, insurance, claims and out-of-pocket spend — under my control.”

  1. See one timeline across providers and sources, each item labelled with where it came from.
  2. Understand active conditions, medicines, allergies, vaccinations and due tests (care gaps).
  3. Track vitals, activity, sleep, nutrition and chronic-condition trends.
  4. Know what is covered, excluded, waiting-period restricted, or likely to need pre-authorisation, with the policy clause cited.
  5. Track claims, reimbursements, rejections and net out-of-pocket spend.
  6. Control who sees what, for what purpose and for how long, and withdraw at any time.

2.5 Monetization (carried forward from v1.0, with guardrails)

Segment Primary model Secondary model Guardrail
Consumers Freemium: free vault + ABHA; Premium (AI summaries, insurance intelligence, trends) Family plan (likely the strongest converter, to be validated) No paywall on consent withdrawal, data export or deletion; no dark patterns in upgrade flows [S45]
Employers Annual benefits-navigation contract, per-employee pricing Wellness/preventive modules Employer sees only aggregated, de-identified metrics; never individual records
Insurers / TPAs Document-completeness and member-engagement platform fee Per-claim workflow pricing Member data only with member’s specific consent; no underwriting use without separate explicit consent and legal review
Providers (Phase 2+) SaaS for patient-shared summaries and care-gap lists Usage-based document AI Consent-scoped; clinician remains decision-maker
Diagnostics / pharmacy Referral or fulfilment fee Co-branded preventive packages Disclose conflicts; recommendations ranked by clinical need, never by commission
Data products Aggregated, de-identified, consented analytics only — No individual-level data sales, ever. Research use requires separate consent and ethics review [S44]

Insurance-specific warning. Showing a user’s own coverage is information service. Recommending, comparing or facilitating purchase or renewal of insurance products may require registration as an IRDAI-regulated intermediary (e.g. insurance web aggregator, corporate agent or broker) [S47] ⚠ Validate with insurance counsel before launching any “renew/upgrade” CTA.

3. India Health Data Landscape

3.1 Ecosystem map (corrected)

                          Individual (Data Principal) + family/caregivers
                                         |
                          ABHA number / ABHA address (identity & addressability)
                                         |
      +----------------------------------+-----------------------------------+
      |                                  |                                   |
 ABDM HIE-CM + Gateway             Direct, user-consented              Claims & care rails
 (consent artifacts; routing;      channels (outside ABDM)             (participant-to-participant)
  no central record store)               |                                   |
      |                       +----------+-----------+            +----------+----------+
 HIPs: hospitals, clinics,    | On-device: Health    |            | NHCX: provider <-> |
 labs, PHR apps (FHIR R4)     | Connect / HealthKit  |            | insurer/TPA/scheme |
      |                       | OAuth: Oura, Fitbit, |            | (consumer app is   |
      |                       |  Withings, Garmin*   |            |  NOT a participant)|
      |                       | Documents: upload,   |            | UHI: discovery,    |
      |                       |  camera, email/WA fwd|            |  booking (HSPA)    |
      |                       | Manual entry         |            +----------+---------+
      |                       +----------+-----------+                       |
      v                                  v                                   v
 +--------------------------------------------------------------------------------+
 |   Consumer Health Aggregator = ABDM PHR app + HIU (+ optional HIP)              |
 |   Data Fiduciary under DPDP: stores, normalises, analyses with consent          |
 +--------------------------------------------------------------------------------+
                 |  Individual Health Graph + Insight Engine + Consent Ledger
                 v
   Consumer app (family)  ·  time-bound "share with doctor"  ·  (Phase 2+) provider / insurer views
   * Garmin and Samsung cloud APIs need partner approval

3.2 Component roles

Component Role Relevance to the aggregator Source
ABHA number 14-digit health ID, created via Aadhaar or other supported verification (e.g. mobile/driving licence) ⚠ Validate current methods Identity anchor for ABDM linking; not a record store [S1][S2][S14]
ABHA address Human-readable handle (e.g. name@abdm) used for consent and sharing Primary key for ABDM consent; a user may hold more than one [S1][S2]
HIP Health Information Provider: hospital, clinic, lab, PHR app holding records and publishing care contexts Source of verified clinical records [S1]
HIU Health Information User: requests records with a valid consent artifact The aggregator, for consent-based pulls [S1][S4]
PHR app Patient-facing app: ABHA creation, care-context discovery and linking, consent approval, self-requested record access The aggregator’s consumer integration mode [S2][S5] ⚠ Validate PHR-app API scope
HIE-CM Health Information Exchange–Consent Manager: creates, stores and enforces consent artifacts Plays the AA-equivalent role; operated within ABDM today ⚠ Validate whether third-party CMs are permitted [S2][S5]
Gateway Routes discovery, consent and data-flow messages between CM, HIP and HIU All ABDM calls go through it [S2][S5]
HFR / HPR Health Facility Registry / Healthcare Professionals Registry Verify the facility and clinician on a record; build the provider directory [S2]
NHCX National Health Claims Exchange for eligibility, pre-auth, claims and communications between providers and payers (roles PAYER/PROVIDER/TPA) Not directly accessible to a consumer app; see §3.4 [S7][S8][S59]
UHI Open protocol for discovery, booking, fulfilment and follow-up (EUA ↔︎ HSPA) Later-phase booking for care gaps [S13]
NRCES National Resource Centre for EHR Standards: publishes the ABDM FHIR R4 IG; SNOMED CT National Release Centre for India Normalization target; terminology licence [S6][S17][S50]

3.3 What a consumer aggregator can and cannot access

Data Can access? Condition / mechanism
Records at ABDM-integrated HIPs linked to the user’s ABHA Yes User discovers/links care contexts and approves consent; data pushed encrypted to the aggregator’s endpoint [S1][S4]
Records at providers not integrated with ABDM Only via the user User uploads, photographs or forwards documents; the provider can be invited to onboard as an HIP
Hospital/lab/insurer/pharmacy portals via the user’s password (“screen-scraping”) No Prohibited by this design. It violates platform terms and risks unauthorized access under the IT Act; it also undermines consent and auditability.
ABDM “central database” Does not exist ABDM is federated. Records stay with HIPs [S1][S2].
Insurance claim status via NHCX Not directly NHCX participants are providers and payers; consumer access needs a payer/TPA partnership or a future consumer rail ⚠ Validate
Insurance policy data via the RBI Account Aggregator Possibly Insurance is an FIP category in the AA framework, but live insurance data flows are limited ⚠ Validate. The aggregator would need to be an FIU [S9][S10][S46].
Wearable data Yes On-device (Health Connect/HealthKit) with OS permission, or vendor OAuth where available
Pharmacy orders/invoices Via the user No consumer APIs from major e-pharmacies (assumed; ⚠ Validate). Use invoice upload or email-forward.
Mental-health notes from therapists Only with explicit, separate consent Higher confidentiality expectation under the Mental Healthcare Act 2017 [S23]

3.4 NHCX: what it is and is not for this product

3.5 Integration pathways summary

Capability MVP pathway Later pathway Standard
Clinical history, labs, prescriptions, discharge summaries ABDM PHR + HIU; document upload More HIPs onboarded; provider partnerships NRCES FHIR R4 [S6]
Linking across providers ABHA care-context discovery and linking — ABDM APIs [S5]
Insurance policy terms Policy schedule / CIS upload + document AI Insurer/TPA/employer partnership APIs; possibly AA Internal model mapped to FHIR Coverage/InsurancePlan
Claims Claim-document upload + manual status Payer/TPA event push; NHCX via provider SaaS FHIR Claim/ClaimResponse
Wearables/vitals Health Connect, HealthKit (on-device) Selected vendor clouds (Oura, Withings, Fitbit) Internal metric model, LOINC/UCUM
Diet In-app logging over Indian food DB; photo logging Partnerships with nutrition apps IFCT-based nutrient model [S34]
Pharmacy Invoice/prescription upload; email-forward e-pharmacy partnerships (order history with consent) Internal medication model; SNOMED CT
Booking care-gap services Deep links to user’s chosen provider UHI EUA integration UHI/Beckn [S13]
Clinician/facility verification HFR/HPR lookup — ABDM registry APIs [S2]

4. Source-by-Source Data Map

Feasibility key:

Consent key:

4.1 Medical records

Source Data available Access mechanism Format → normalization Consent / privacy Feas. Regulatory / commercial constraint MVP approach
ABHA (number/address) Demographics, ABHA address(es), mobile; not records ABDM M1 APIs (create/verify/link ABHA) via sandbox → production [S5][S15] JSON → Person, Identifier DPDP-C; Aadhaar-based flows follow Aadhaar rules (no storage of Aadhaar number beyond permitted use) ⚠ Validate High ABDM integration approval, security audit, sandbox exit [S4] ABHA create/login in onboarding (skippable)
ABDM-integrated hospitals/clinics (HIPs) Seven HI types in NHA’s reference implementation: OPConsultation, Prescription, DischargeSummary, DiagnosticReport, ImmunizationRecord, HealthDocumentRecord, WellnessRecord [S60] (an Invoice type appears in later NRCES material ⚠ Validate) ABDM PHR discovery/link + data pull (self-requested) or HIU consent request [S1][S4] Encrypted FHIR R4 bundles (NRCES profiles) → canonical graph; keep the raw bundle ABDM-CA + DPDP-C; honour dataEraseAt and access mode Medium Coverage varies by HIP; many bundles carry PDFs inside DocumentReference/Binary rather than coded data ⚠ Validate Core MVP source; measure structured vs PDF share
ABDM-integrated labs (HIPs) Lab reports (DiagnosticReport + Observations, often PDF) Same as above FHIR → LabResult; map local test names to LOINC ABDM-CA + DPDP-C Medium Lab chains’ ABDM participation varies ⚠ Validate per chain Core; fall back to PDF upload
Other PHR apps (as HIPs) Records users stored in other apps ABDM: if the other PHR acts as HIP, the user can link and pull FHIR (often HealthDocumentRecord) ABDM-CA Medium Depends on the other app’s HIP integration Supported automatically through the ABDM flow
Non-ABDM providers Same clinical content on paper/PDF User upload, camera scan, email-forward, WhatsApp-forward PDF/image → OCR + clinical NER → canonical with confidence + page reference DPDP-C; user is the source; label “document-derived” High OCR accuracy on handwriting is low (manual review queue) Core MVP
Handwritten prescriptions Drug names, doses, frequency, advice Camera scan Handwriting OCR (low confidence) → user confirmation DPDP-C Low–Med Never auto-confirm; misread drug names are a safety risk Capture image, assisted transcription, user confirms each line
Discharge summaries Diagnoses, procedures, meds on discharge, follow-up ABDM (DischargeSummary) or upload FHIR Composition or PDF → Encounter, Condition, Procedure, MedicationRequest, CarePlan ABDM-CA / DPDP-C Med Free-text heavy; clinical NER needed Core; human-review queue for low confidence
Imaging reports (X-ray, USG, CT, MRI) Report text, impression; images rarely ABDM DiagnosticReport (imaging) or upload PDF → DiagnosticReport(imaging); store impression text, no auto-interpretation DPDP-C Med DICOM images are large; out of MVP scope Store report + impression; no image viewer in MVP
Vaccination records Child UIP vaccines, adult vaccines (incl. CoWIN certificates) ABDM ImmunizationRecord, upload of cards/certificates → Immunization with vaccine code + dose DPDP-C; parental consent for minors [S12] Med Paper immunization cards dominate for children Upload + guided manual entry against the UIP schedule [S38]
Manual entry (history, allergies, family history) Conditions, allergies, surgeries, family history In-app forms, voice entry (vernacular) → canonical with trust=user_entered DPDP-C High Accuracy depends on the user Core; structured pick-lists (SNOMED-backed)

4.2 Health insurance

Source Data available Access mechanism Format → normalization Consent / privacy Feas. Regulatory / commercial constraint MVP approach
Retail policy schedule + policy wording Insurer, plan, policy no., insured members, sum insured, deductibles, co-pay, room-rent limits, sub-limits, waiting periods, exclusions, riders, renewal date, premium User upload (PDF from insurer email/app), email-forward PDF → policy parser → InsurancePolicy, Coverage, Exclusion, WaitingPeriod (FHIR Coverage/InsurancePlan-aligned) DPDP-C High Wordings vary by insurer and product; clause-level citation essential Core MVP: parse schedule + CIS; link clauses to the original page
Customer Information Sheet (CIS) Standardised summary of key features, exclusions, waiting periods Usually attached with the policy Semi-structured → high-precision parser DPDP-C High IRDAI mandates CIS in a prescribed format [S18] ⚠ Validate current format Prioritise the CIS parser; it is the best structure available
Employer group policy (GMC) E-card, member ID, TPA, sum insured (often family floater), maternity/parents cover, room-rent caps E-card/benefit booklet upload; HR portal export; employer partnership PDF/image → InsurancePolicy(group) DPDP-C; employer must not see individual data High (upload) / Med (partnership) Policy wording often not given to employees; benefit summary only Upload e-card + benefit summary; B2B employer channel in Phase 2
Insurer apps/portals Policy docs, claim status, network hospitals No public consumer APIs (assumed); partnership — DPDP-C + partner data-sharing agreement Low (now) Commercial negotiation; IRDAI rules on outsourcing/data sharing ⚠ Validate Do not scrape; user downloads PDFs and uploads them
TPA portals (e.g. Medi Assist, MDIndia, FHPL, Paramount) Claim intimation, pre-auth status, query letters, settlement letters Partnership; user upload of letters/SMS/email PDF/email → Claim, ClaimEvent DPDP-C Low–Med TPA contracts are with insurers; data-sharing terms needed Email-forward of TPA mails; claim letter parser
Claim documents Pre-auth approval/denial, final bill, discharge bill, settlement letter with deductions, rejection reason Upload, email-forward PDF → Claim, line-item Deduction, DenialReason (taxonomy) DPDP-C High Reason text is unstructured; map to a denial taxonomy Core MVP: claim tracker + deduction explainer
Cashless network hospital lists Hospitals in the insurer/TPA network Insurer/TPA websites publish lists (use only where terms permit), partnership CSV/PDF → NetworkListing linked to HFR IDs Public business data, no personal data Med Website terms may restrict automated reuse; prefer licensed feeds ⚠ Validate Phase 2 via partnership; MVP shows “check with insurer” + link
NHCX Eligibility, pre-auth, claims, communications, insurance-plan exchange between provider and payer; ABHA↔︎policy linking Participant onboarding, roles PAYER/PROVIDER/TPA only [S59] NRCES/NHCX FHIR bundles, JWE-encrypted Participant agreements Low for consumer Consumer app is not a participant (V) Phase 3 via provider SaaS or payer partner; track ABHA-policy linking for consumer discovery
Account Aggregator (insurance FIP) Policy holdings/details where insurers/repositories act as FIPs Register as FIU; consent via an AA [S9][S10][S46] ReBIT FI schemas → InsurancePolicy AA consent artefact + DPDP-C Low–Med Coverage of health-insurance FIPs live on AA is uncertain ⚠ Validate Phase 2 discovery: assess FIP availability
e-Insurance Account (insurance repositories) Dematerialised policies held in eIA Consumer download; possibly via AA PDF/statement DPDP-C Low–Med Low adoption for health policies (assumed) ⚠ Validate Accept uploads
PM-JAY / state schemes Entitlement (Ayushman card), empanelled hospitals; claims are handled provider↔︎scheme Beneficiary portals/apps; ABHA linkage ⚠ Validate — DPDP-C Low No consumer claim API known Show eligibility info + card upload; signpost to official helpline
Premium payments Premium amount, date, payer, tax-deduction eligibility (Sec 80D) Upload premium receipt; bank SMS (not recommended) → Premium DPDP-C High Tax advice boundaries; show information only Upload premium receipts; renewal reminder

4.3 Fitness and vitals

Source Data available Access mechanism Format → normalization Consent / privacy Feas. Constraint MVP approach
Google Health Connect (Android) Steps, distance, exercise, heart rate, HRV, sleep sessions/stages, weight, body fat, BP, blood glucose, SpO2, nutrition, hydration, mindfulness, etc. On-device Android API; per-data-type runtime permission; built into Android 14+ (app install on older versions) [S26] ⚠ Validate Records → WearableMetric/Vital/Sleep/Activity (LOINC/UCUM mapping) OS-P + DPDP-C; Google Play Health Apps declaration; background read and history read (>30 days before grant) need extra permissions [S26] ⚠ Validate High Play policy restricts use (no ads/data brokering) Core MVP (Android-heavy India)
Apple HealthKit (iOS) Similar set + State of Mind (iOS 17+) On-device only; no Apple cloud API [S24] HKSample → canonical OS-P; App Review 5.1.3: no advertising or data-mining use, no personal health info in iCloud [S25] ⚠ Validate wording High Apple’s Health Records (FHIR) feature is not available in India (assumed) ⚠ Validate Core MVP (iOS)
Google Fit APIs Legacy fitness data Deprecated; migrate to Health Connect [S27] ⚠ Validate shutdown date (2026) — — Low Being shut down Do not build
Fitbit Activity, sleep, HR, SpO2, weight Fitbit Web API (OAuth 2.0) or via Health Connect [S28] JSON → canonical OAuth + DPDP-C Med Google is moving Fitbit APIs to a new Google Health API ⚠ Validate status Prefer Health Connect; direct API later
Garmin Activity, sleep, stress, Body Battery, HR Garmin Connect Developer Program (Health/Activity API), approval required [S29] Push/pull JSON → canonical OAuth + DPDP-C Med Business approval; usage terms Phase 2
Samsung Health Steps, sleep, HR, BP (Galaxy Watch), body composition Samsung Health Data SDK (partner approval) or Samsung → Health Connect sync [S30] ⚠ Validate → canonical OS-P/partner Med Partner programme Via Health Connect in MVP
Oura Sleep, readiness, HRV, temperature trend, activity Oura Cloud API v2 (OAuth 2.0) [S31] ⚠ Validate PAT deprecation JSON → canonical OAuth + DPDP-C High Small India base Phase 2
Withings (scales, BP) Weight, body composition, BP, ECG summary Withings Public API (OAuth) [S32] JSON → canonical OAuth High Small India base Phase 2
Indian wearables (Noise, boAt, Fire-Boltt, Amazfit/Zepp, Ultrahuman) Steps, HR, sleep, SpO2 (accuracy varies) Mostly through Health Connect/HealthKit sync from the brand app; direct APIs uncommon ⚠ Validate per brand → canonical with device_class=consumer_wearable OS-P Med Data quality varies; consumer wearables are not diagnostic devices Via Health Connect/HealthKit; label as “device-measured, non-clinical”
BP monitors (Omron, Dr Trust, others) Systolic/diastolic, pulse Brand app → Health Connect/HealthKit where supported; else manual/photo of display → Vital(BP) LOINC 85354-9 panel / 8480-6 systolic / 8462-4 diastolic OS-P Med Validate device sync per brand Health Connect + manual entry with photo
Glucometers (Accu-Chek, OneTouch, Dr Morepen, BeatO) SMBG readings Brand app sync/Health Connect where supported; manual entry → Vital(glucose) with meal context OS-P Med Many Indian meters have no export Manual entry with fasting/post-meal tags
CGM (FreeStyle Libre, Ultrahuman M1) Continuous glucose No public consumer API known for LibreView ⚠ Validate; CSV export from vendor apps; Health Connect where vendor writes CSV → time series DPDP-C Low–Med Vendor terms; clinical interpretation risk Phase 2: CSV import; show trends only
Smart scales Weight, BMI, body-fat estimates Health Connect/HealthKit or vendor OAuth → Vital(weight) OS-P High Body-fat estimates are approximate Via Health Connect

4.4 Diet and nutrition

Source Data available Access mechanism Format → normalization Consent / privacy Feas. Constraint MVP approach
IFCT 2017 (ICMR-NIN) Nutrient composition of Indian foods (raw foods) Published by ICMR-NIN (book/digital); licensing terms for commercial digital reuse ⚠ Validate [S34] Reference table → FoodItem nutrients Reference data Med Covers raw ingredients, not cooked regional dishes License/obtain permission; base nutrient layer
Indian recipe/dish databases (e.g. Indian Nutrient Databank and similar academic efforts) Cooked-dish nutrients for regional dishes Academic/open datasets ⚠ Validate licence and provenance Recipe → nutrients via IFCT Reference data Med Portion sizes (katori, roti) need standardisation Build a curated dish DB with dietitian review
ICMR-NIN Dietary Guidelines for Indians (2024) and Nutrient Requirements (2020) RDAs/EARs, dietary guidance Public documents [S34] ⚠ Validate Rules → NutritionTarget Reference High Guidance must be presented as general, not prescriptive Default targets; clinician can override
Indian nutrition apps (HealthifyMe etc.) Food logs, calories, macros No public API known ⚠ Validate; may write nutrition to Health Connect/HealthKit → NutritionLog OS-P Low–Med Commercial partnership needed for direct data Read nutrition records via Health Connect/HealthKit where written
In-app food logging Meals, portions, water, supplements Native, vernacular dish search, voice, photo → NutritionLog with trust=user_entered DPDP-C High Food photo AI accuracy is moderate Simple logging for chronic-care users; photo logging later
Food delivery/grocery receipts (Swiggy, Zomato, Blinkit, Zepto, BigBasket) Items, restaurant, time, cost No consumer APIs; email-forward/upload of invoices Email/PDF → item list → approximate nutrition DPDP-C; highly sensitive inference (eating patterns) — opt-in only Low–Med Gmail API inbox access is a restricted scope: Google verification plus annual third-party security assessment (CASA) [S33] ⚠ Validate. Prefer user-initiated forwarding to a unique inbound address. Phase 3, opt-in only; never infer from bank SMS
Supplements/Ayurveda Supplement names, dose Manual entry, label photo → Medication(type=supplement) DPDP-C High Interaction data for herbal products is sparse Capture in the medicine list; flag “interaction data limited”
Packaged food labels Nutrition facts Label photo OCR → nutrients per serving DPDP-C Med FSSAI labelling formats; front-of-pack rules still evolving ⚠ Validate Phase 3

4.5 Pharmacy and medication

Source Data available Access mechanism Format → normalization Consent / privacy Feas. Constraint MVP approach
E-pharmacy invoices (Tata 1mg, PharmEasy, Apollo 24/7, Netmeds, Truemeds) Brand, strength, quantity, date, price, prescriber (sometimes) Upload, email-forward; no consumer APIs known ⚠ Validate; partnership later PDF → MedicationDispense → brand→molecule mapping DPDP-C High (upload) E-pharmacy rules still in draft form (Draft Drugs and Cosmetics Amendment Rules, 2018) with litigation ⚠ Validate current status [S52] Core MVP: invoice parser feeds the medicine list and spend
Offline pharmacy bills Same, often thermal-printed or handwritten Camera scan OCR (low quality) → user confirm DPDP-C Med Faded thermal paper Capture + confirm
Prescriptions (ABDM or paper) Drug, dose, frequency, duration, advice ABDM Prescription HI type; upload FHIR MedicationRequest / OCR ABDM-CA / DPDP-C Med Handwriting Core
Brand→molecule mapping Molecule(s), strength, form, schedule (H/H1/X) Licensed commercial Indian drug database (e.g. CIMS/MIMS India) or curated build from public lists (CDSCO approvals, NLEM, Jan Aushadhi product list) [S35][S36] → internal DrugProduct → molecule → ATC/SNOMED CT Reference Med No complete free Indian brand dictionary (assumed); licensing cost Partner/license for MVP; supplement with curation
Drug-interaction knowledge DDI severity, duplicate therapy, contraindications Licensed clinical knowledge base (e.g. First Databank, Medi-Span, Lexicomp, DrugBank commercial) Molecule-level rules Reference Med Licence cost; must map Indian brands → molecules; clinical governance of alert thresholds Phase 2 with clinician-reviewed severity filter; MVP shows duplicate-molecule detection only
Terminology SNOMED CT (incl. India drug extension) SNOMED CT free for use in India via NRCES as National Release Centre [S50] ⚠ Validate terms and drug-extension scope Codes — High Affiliate licence registration Use SNOMED CT + ATC
Medicine reminders/adherence Taken/missed doses In-app reminders; caregiver nudges; WhatsApp reminders via WhatsApp Business API (opt-in) → AdherenceEvent DPDP-C; WhatsApp template-message and opt-in rules High Notification fatigue MVP: reminders + refill-due estimate

4.6 Diagnostics

Source Data available Access mechanism Format → normalization Consent / privacy Feas. Constraint MVP approach
Lab chains (Dr Lal PathLabs, Metropolis, Thyrocare, Agilus/SRL, Apollo Diagnostics, Redcliffe, Tata 1mg Labs) Test results, reference ranges, flags, PDFs ABDM HIP where integrated ⚠ Validate each; otherwise PDF via email/WhatsApp/download FHIR or PDF → LabResult (LOINC mapping, UCUM units, lab-specific reference range retained) ABDM-CA / DPDP-C High (PDF) / Med (ABDM) Template diversity is huge; report layout drift Core MVP: lab PDF parser with per-lab templates + generic table extraction; partner with 2–3 chains
Preventive health check packages Bundled panels (CBC, lipid, LFT, KFT, HbA1c, TSH, vitamin D/B12, urine) Same Same Same High Same Core; “health check summary” view
At-home tests (glucose, pregnancy, rapid antigen) Self-reported results Manual entry/photo → LabResult(trust=user_entered) DPDP-C High Not lab-verified Allow; label clearly
Hospital in-house labs Same ABDM where hospital is an HIP; else PDF Same Same Med — Via ABDM/upload
Reference ranges & units Lab-specific ranges From the report Store the lab’s range; do not substitute a generic range for flags — High Units differ (mg/dL vs mmol/L) Unit normalisation with UCUM; keep the original

4.7 Mental wellbeing (explicit, separate opt-in)

Source Data available Access mechanism Format → normalization Consent / privacy Feas. Constraint MVP approach
Self-reported mood/stress check-ins Mood scale, tags, notes In-app → MoodLog Separate opt-in purpose; excluded from family/caregiver view and doctor-share by default; never used for marketing High Mental Healthcare Act 2017 confidentiality principles [S23] Phase 2; local-first option
Validated screeners (e.g. PHQ-9, GAD-7) Scores In-app questionnaire → Observation(screener) Separate opt-in; crisis routing when thresholds/self-harm item triggered High Licensing of instruments ⚠ Validate; clinical escalation pathway required Phase 2 with clinical safety sign-off
Sleep, HRV, mindfulness minutes From wearables Health Connect/HealthKit (incl. iOS State of Mind) → Sleep, WearableMetric OS-P; treated as wellbeing context High Not diagnostic MVP: show sleep; stress inference not in MVP
Therapy notes/journals Free text User-entered only; no therapist-system integration in MVP Encrypted, optionally on-device only Highest sensitivity; excluded from AI by default Med Professional confidentiality Phase 3, on-device encrypted journal
Crisis resources Tele-MANAS 14416 / 1-800-891-4416; emergency 112 [S39] ⚠ Validate numbers before release Static content — — High Must be correct and localised MVP: always available from Help and from any crisis trigger

4.8 Family and context

Source Data available Access mechanism Format → normalization Consent / privacy Feas. Constraint MVP approach
Dependents (children <18) Child profile, vaccinations, growth, records Parent creates; child ABHA where supported ⚠ Validate → Person(dependent) + Relationship Verifiable parental consent under DPDP s.9 and Rules [S11][S12]; no tracking or targeted ads for children High Age-gating; transition at 18 (hand over control) MVP
Adult dependents (elderly parents, spouse) Same as individual Invite flow: the adult consents in their own right (OTP on their phone) or appoints the caregiver → CaregiverGrant The adult’s own consent; for persons unable to consent, lawful guardian rules [S12] ⚠ Validate High Never let one adult silently manage another adult’s data MVP: invite + assisted-consent mode
Family history / hereditary risk Conditions in relatives Manual entry → FamilyHistory (SNOMED) DPDP-C; it is data about relatives, so collect minimally High Third-party data MVP: simple first-degree relatives form
Lifestyle (tobacco, alcohol, activity) Self-report Questionnaire → RiskFactor DPDP-C High Social-desirability bias MVP
Occupation, travel, exposure Shift work, dust/chemical exposure, travel to endemic areas Questionnaire → RiskFactor DPDP-C; optional High Minimise Phase 2
Environmental (air quality) Local AQI Public AQI feeds (e.g. CPCB) by coarse location ⚠ Validate terms → EnvironmentContext Coarse location only, opt-in Med Location sensitivity Phase 3 (asthma/COPD use case)

5. ABDM and Consent Architecture

5.1 Financial AA vs ABDM: map the analogy correctly

The most common design error is to treat the consumer app as the “health Account Aggregator”. In the RBI framework the AA is a data-blind consent conduit: it cannot see or store the financial data it moves [S10][S46]. In ABDM, the HIE-CM plays that conduit role [S1][S2]. A consumer health aggregator that stores, normalises and analyses records is the equivalent of an FIU / personal-finance app. It is therefore a Data Fiduciary with full DPDP obligations [S11].

Dimension Financial Account Aggregator ABDM health exchange Where the consumer health aggregator sits
Regulator / owner RBI; AAs are NBFC-AAs under the 2016 Master Direction [S10] National Health Authority (ABDM) [S1][S14] DPDP (MeitY/DPB), ABDM policies, plus sector rules (IRDAI, CDSCO, NMC) depending on features
Consent conduit AA (data-blind) HIE-CM + Gateway [S2] Not the conduit; it is a consent requester and consent UI (PHR)
Data holder FIP (bank, insurer, MF, GSTN) [S9] HIP (hospital, clinic, lab, PHR) [S1] Can also be an HIP when re-sharing user-uploaded documents
Data requester FIU (lender, wealth app) HIU [S1] HIU (and PHR app)
Consent standard ReBIT consent artefact [S46] ABDM consent artifact: purpose, HI types, care contexts, permission{accessMode, dateRange, dataEraseAt, frequency} [S60] Must honour the artifact and run its own DPDP consent ledger for non-ABDM sources
Data format ReBIT FI schemas NRCES FHIR R4 [S6] FHIR as the canonical exchange format; internal graph for analytics
Encryption in transit Per-session keys ECDH key agreement on Curve25519, HKDF-derived key, AES-GCM; bundles pushed to the HIU’s data-push URL [S60] Generate fresh key pairs per data request; never reuse
Retention AA cannot store data Federated; no central record store [S1][S2]; artifact carries dataEraseAt [S60] Must purge or crypto-shred ABDM-sourced data at dataEraseAt or on revocation
Adjacent rails — NHCX (claims) [S59], UHI (discovery/booking) [S13] Partnerships and provider SaaS give access to these rails
Product analogy “My money across banks, investments, insurance” “My records across hospitals and labs” “My health and my health money across records, devices, documents and insurance”

Implication. Health data in India is more fragmented and more paper-based than financial data. ABDM covers only part of the record landscape. The product must combine ABDM with documents, devices and partnerships. The DPDP posture should be a custodian with minimal retention, not a data lake.

5.2 ABDM building blocks explained for the product team

Concept Explanation Product design consequence
ABHA number A 14-digit identifier created with Aadhaar or driving-licence-based verification (per NHA’s NHCX guide, which lists “Create ABHA number using Aadhaar / Driving License”) [S59] Offer creation in-app (M1). Never make ABHA mandatory for basic use.
ABHA address A user-chosen handle (e.g. name@abdm in production; name@sbx in sandbox) [S60]. Consent and linking are addressed to it. Store all addresses a user holds. Show which address each record came through.
Care context A provider-side grouping of records (e.g. a visit or an admission) that the HIP links to the user’s ABHA address. In v3 the HI type is declared on the care context [S60]. Discovery returns care contexts. The UI should show them as “visits”.
Discovery and linking User-initiated: the PHR asks an HIP to find the user’s records (demographic match + OTP); HIP-initiated: the facility links at the point of care; Scan & Share: QR-based registration at a facility [S60] Build a “Find my records” flow: choose a hospital or lab → OTP → link. Show progress and failures honestly.
Consent request An HIU asks the CM for access with purpose, HI types, date range, expiry/erase date and frequency. The user approves in a PHR/ABHA app. The CM issues consent artifacts per HIP [S1][S4][S60]. Requests should be narrow by default (purpose-specific, bounded date range, explicit erase date).
Purpose codes Coded purposes, e.g. CAREMGT (care management, seen in NHA’s reference implementation [S60]); others in ABDM specifications include BTG (break-the-glass), PUBHLTH, HPAYMT (healthcare payment), DSRCH (disease-specific research), PATRQT (self-requested) ⚠ Validate full list in current spec Map each product feature to one purpose. Self-access → PATRQT; doctor-share → CAREMGT; never request DSRCH without a separate research consent and ethics approval.
Access mode VIEW, STORE, QUERY, STREAM [S60] Request STORE only where the feature needs persistence (personal vault). The clinician view should use VIEW.
Health-information request and data push The HIU requests data against a consent artifact with its public key; HIPs encrypt FHIR bundles and push them to the HIU’s data-push URL; HIU acknowledges per care context [S1][S4][S60] Idempotent, retry-safe ingestion with per-transaction keys; partial-delivery reconciliation.
Revocation and expiry Users can revoke consents from the PHR/ABHA app; artifacts expire Subscribe to consent notifications; on revoke or expiry, stop fetches and run the erasure workflow.
HFR / HPR Registries of facilities and professionals [S2]; HFR registration is a prerequisite for NHCX participation [S59] Verify facilities and clinicians on records (“Verified facility” badge).
Milestones M1: ABHA creation, verification, link token. M2: linking and exporting health data (HIP). M3: consent request and importing data (HIU) [S60] The aggregator needs M1 + M3 (and M2 if it acts as an HIP for user-uploaded records), plus PHR-app flows ⚠ Validate PHR-app certification path.
API generation NHA’s reference implementation moved from V1 to V3 gateway APIs (/api/hiecm/gateway/v3/…, X-CM-ID: sbx|abdm) in 2024 [S60] Build directly on V3. Note that some NHCX docs still reference a v0.5 sessions endpoint [S59]; confirm with NHA integration support.
Certification Functional testing, WASA security testing, sandbox exit, production credentials [S59][S4] Plan security audit and demo as gating items on the critical path.

5.3 How the aggregator participates: integration modes

Mode What it enables Required milestones MVP?
PHR app (consumer-facing) ABHA create/login, discover and link care contexts, approve/deny/revoke consents, self-requested data pull M1 + PHR-app integration with the CM ⚠ Validate PHR certification requirements Yes
HIU Consent requests for defined purposes (e.g. a clinician the user authorises; family caregiver flows), receipt of encrypted bundles M3 Yes
HIP Re-share user-uploaded documents (as HealthDocumentRecord) and app-generated wellness data (WellnessRecord) to other HIUs the user chooses M2 Phase 2
NHCX participant Eligibility/pre-auth/claim flows Only as provider/payer/TPA or their tech partner [S59] Phase 3 (provider SaaS)
UHI EUA Book diagnostics or consultations to close care gaps UHI onboarding [S13] Phase 2–3

The product needs two consent layers recorded in one immutable consent ledger:

  1. ABDM consent artifacts. These govern records pulled from HIPs. They are issued by the HIE-CM and honoured as-is.
  2. DPDP platform consents. These govern every other source (documents, devices, manual entry) and every processing purpose (AI summaries, insurance analysis, reminders, research, product improvement). Each must be free, specific, informed, unconditional and unambiguous, with withdrawal as easy as giving consent [S11]. Notices must meet the content requirements of the DPDP Rules [S12] ⚠ Validate rule-level notice content.

Purpose catalogue (each purpose is a separate toggle; default in brackets)

Purpose ID Description Lawful basis Default Notes
P1 vault_storage Store my records and documents Consent Required for the core service Explain storage location (India) and retention
P2 abdm_self_access Fetch my ABDM records (PATRQT/STORE) ABDM artifact + consent Opt-in per HIP Narrow date range
P3 clinical_summary Generate timeline, summaries, trends, care gaps Consent Opt-in (recommended) Includes AI processing disclosure
P4 insurance_analysis Parse policies, explain coverage, track claims, compute OOP Consent Opt-in
P5 device_sync Read selected Health Connect/HealthKit types OS permission + consent Opt-in per data type
P6 reminders Medicine, test and renewal reminders (push, SMS, WhatsApp) Consent Opt-in per channel WhatsApp opt-in recorded
P7 doctor_share Time-bound share with a named clinician (CAREMGT, VIEW) Consent Per share Expiry default 72h (product choice)
P8 caregiver_access A named family member can view/manage Consent of the person, or verifiable parental consent Per grant Scope: view / manage / emergency-only
P9 mental_wellbeing Mood and screener data Consent Off Excluded from P7/P8 unless explicitly included
P10 product_improvement Use de-identified usage and extraction-error data to improve OCR/parsers Consent Off Never raw clinical content without explicit consent
P11 research Contribute de-identified data to approved studies Consent + ethics approval Off Per-study consent [S44]
P12 offers Partner offers (diagnostics, pharmacy) Consent Off Never driven by clinical data without explicit consent; never for minors

Consent record (internal, DPDP-aligned; corrects the v1.0 object)

{
  "consent_id": "cn-2026-001234",
  "data_principal_ref": "person:7f3c…",
  "granted_by": {"type": "self", "person_ref": "person:7f3c…"},
  "purpose_id": "P2_abdm_self_access",
  "abdm": {
    "consent_request_id": "…",
    "consent_artefact_ids": ["…"],
    "purpose_code": "PATRQT",
    "hi_types": ["DiagnosticReport", "Prescription", "DischargeSummary"],
    "access_mode": "STORE",
    "date_range": {"from": "2023-01-01", "to": "2026-10-21"},
    "data_erase_at": "2027-10-21T00:00:00+05:30",
    "frequency": {"unit": "MONTH", "value": 1, "repeats": 0}
  },
  "sources": ["HIP:hfr-id-…"],
  "processing_allowed": ["store", "summarise", "trend", "care_gap"],
  "processing_prohibited": ["advertising", "model_training", "sale", "underwriting"],
  "notice_version": "notice-v3.2-hi-IN",
  "language": "hi-IN",
  "channel": "android_app",
  "status": "granted",
  "created_at": "2026-10-21T19:10:00+05:30",
  "withdrawn_at": null,
  "evidence_hash": "sha256:…"
}

The ABDM purpose_code and access_mode values follow NHA’s reference implementation [S60]. PATRQT and the frequency unit values must be validated against the current ABDM V3 specification ⚠ Validate.

5.6 Onboarding as an ABDM participant (organisational checklist)

Step Activity Owner Evidence
1 Incorporate in India; appoint a grievance officer and DPO contact; adopt an ABDM-aligned privacy policy Legal [S11][S16]
2 Apply for ABDM sandbox (V3) registration; obtain client ID/secret Engineering [S59][S15]
3 Build M1 (ABHA), PHR-app flows, M3 (HIU); register bridge/callback URL Engineering [S60]
4 Pass functional test cases; FHIR bundle validation QA / NRCES [S59]
5 WASA security audit (CERT-In empanelled auditor recommended); fix findings Security [S59][S20] ⚠ Validate auditor requirement
6 Sandbox exit; NHA demo; production credentials (X-CM-ID: abdm) Programme [S59][S60]
7 Production go-live with monitoring of HIP response rates and failure reasons SRE / Product —

6. Unified Individual Health Graph

6.1 Design principles

  1. FHIR at the edges, graph at the core. Ingest and export FHIR R4 (NRCES profiles). Internally, keep a typed property graph (or relational + edge tables) tuned for timeline, trend and coverage queries.
  2. Every fact is an assertion with provenance. A lab value is stored as “Source X asserted HbA1c = 7.1% on date D, extracted with confidence c, verified by V”. It is never stored as bare truth.
  3. Trust tiers are first-class and are shown in the UI:
Tier Label in UI Definition Example
T1 Verified clinical Received from an identified HIP via ABDM (or a signed partner feed), unchanged HbA1c in a FHIR DiagnosticReport from a lab HIP
T2 From your document Extracted from a user-supplied provider document; shows confidence and user confirmation status HbA1c parsed from a lab PDF on WhatsApp
T3 Measured by your device From a consumer device via Health Connect/HealthKit/OAuth Resting heart rate from a smartwatch
T4 Entered by you / your family Manually entered or self-reported “Mother has diabetes”
T5 Insight (computed) Derived by rules or models from T1–T4, with method and inputs listed “HbA1c has risen across the last two tests”
T6 Reference knowledge External guidelines, drug knowledge, policy clauses NP-NCD screening guidance; IRDAI CIS clause
  1. Never silently overwrite. Conflicts are preserved and surfaced.
  2. Consent-scoped by construction. Every node and edge carries consent_ids[]. Queries filter by active consent and purpose.
  3. Erasable by design. Per-person, per-source data-encryption keys allow crypto-shredding on withdrawal or dataEraseAt.

6.2 Common provenance envelope (on every node)

Field Type Description
id UUID Internal ID
person_id UUID Subject (Data Principal or dependent)
trust_tier enum T1–T6 See table above
source_type enum abdm_hip, document, device, manual, partner_api, derived, reference
source_ref string HIP/HFR ID, document ID + page + bbox, device/app ID, partner ID
raw_ref URI Pointer to the encrypted raw FHIR bundle or document blob
consent_ids UUID[] Consents that authorise holding this node
abdm_txn_id / care_context_ref string For ABDM-sourced data
ingested_at, effective_at timestamp Ingestion time vs clinical/observation time
extraction object {method: ocr+ner|fhir|api, model_version, confidence: 0–1, field_confidences}
verification enum unverified, user_confirmed, user_corrected, clinician_confirmed, source_verified
corrections[] array Who changed what, when, old/new values
supersedes / conflicts_with UUID[] Versioning and conflict links
retention object {policy_id, erase_at, legal_hold: bool}
language BCP-47 Original document language

6.3 Entity catalogue

Update-frequency codes: E = event-driven (on ingestion), D = daily sync, C = continuous/near-real-time device sync, M = manual.

Entity Key fields Source systems Update Data-quality rules Linkage keys
Person name (+ local-script name), DOB/age, sex/gender, mobile, email, preferred language, city/PIN (coarse), blood group Onboarding, ABHA profile, documents M / E ABHA demographics take precedence for name/DOB; mismatch with documents triggers identity check (avoid mixing family members’ records) person_id; ABHA number (hashed); mobile (hashed)
Dependent / Relationship relationship type, guardian flag, minor flag, date of majority, access scope Onboarding M Minor flag auto-computed from DOB; at age 18, trigger hand-over of control person_id ↔︎ related_person_id
ABHA Identity ABHA number (encrypted), ABHA address(es), KYC status, linked HIPs ABDM M1/PHR E One ABHA number per person; many addresses allowed; never store Aadhaar number ABHA number (hashed), ABHA address
Consent Record see §5.4 Consent service, HIE-CM E Immutable append-only; status transitions validated consent_id, ABDM artifact IDs
Caregiver Grant grantor, grantee, scope (view/manage/emergency), categories excluded (e.g. mental), expiry Consent service E Requires grantor authentication (or verifiable parental consent) grant ID

B. Care delivery

Entity Key fields Source systems Update Data-quality rules Linkage keys
Facility name, HFR ID, type, address, city, network-hospital flags HFR, documents E Fuzzy-match document facility names to HFR; mark unmatched as “unverified facility” HFR ID; normalised name + PIN
Clinician name, HPR ID, registration no./council, specialty HPR, documents E Match registration numbers on prescriptions to HPR where possible HPR ID; registration no.
Encounter type (OP/IP/ER/tele/lab visit), start/end, facility, clinician, reason, care-context ref ABDM (OPConsultation, DischargeSummary), documents E Duplicate rule: same person + facility + encounter class + date (±1 day for IP) → merge candidate care-context ref; facility + date
Diagnosis / Condition code (ICD-10 / SNOMED CT), text, clinical status, verification status, onset, recorded date, asserter ABDM, discharge summaries, prescriptions, user E / M User-entered conditions stay T4 until corroborated; “rule out”/“?” phrases parsed as provisional, never as confirmed SNOMED/ICD code + person
Symptom description, onset, severity, duration User (Q&A), documents M Never auto-promote to diagnosis; red-flag terms trigger safety guidance (§9.4) —
Procedure code, date, facility, outcome Discharge summaries, claims E Link to claim line items code + date
Allergy / Intolerance substance, reaction, severity, criticality ABDM, documents, user E / M Always shown with source; “no known allergies” stored distinct from “unknown” substance code
Immunization vaccine, dose number, date, batch, administering facility ABDM ImmunizationRecord, certificates, cards E / M Validate against schedule spacing; flag implausible dates vaccine code + dose

C. Medications

Entity Key fields Source systems Update Data-quality rules Linkage keys
Medication (statement/request) brand, molecule(s), strength, form, dose, route, frequency, duration, start/stop, prescriber, indication ABDM Prescription, prescription images, discharge summaries E Brand→molecule mapping required; unknown brand → “needs confirmation”; duplicate molecule from different brands flagged molecule code (SNOMED/ATC) + person
Medication Dispense brand, qty, date, pharmacy, price, prescription link Pharmacy invoices E Infer days-supply = qty / daily dose (only when dose known) invoice ID; molecule
Adherence Event scheduled time, taken/missed/snoozed App reminders C / M Self-reported; label as such medication ID

D. Diagnostics, vitals and devices

Entity Key fields Source systems Update Data-quality rules Linkage keys
Lab Result (Observation) test name (original), LOINC code, value, unit (original + UCUM), lab reference range, flag, specimen, collected/reported time, lab ABDM DiagnosticReport, lab PDFs E Use the lab’s own reference range for flags. Unit conversion only with a known factor. Values outside physiological plausibility → hold for review. LOINC + collected date + lab
Diagnostic Report panel, report date, performer, result refs, PDF ref, imaging impression (text) Same E Report-level dedup by lab + accession no. + date accession no.
Vital (clinical) BP, pulse, weight, height, BMI, temperature, SpO2, glucose (with meal context) ABDM WellnessRecord, documents, devices, manual E / C / M Context tags (fasting/post-prandial; seated); device class recorded LOINC + timestamp
Wearable Metric metric type, value, unit, interval, device/app, aggregation Health Connect, HealthKit, vendor APIs C / D Daily aggregates stored by default (minimisation); raw high-frequency only if a feature needs it; dedupe overlapping sources by priority order metric + interval + source
Activity steps, active minutes, workouts (type, duration, distance, energy) Same D Prefer the device the user marks as primary interval
Sleep start/end, duration, stages, interruptions Same D Merge overlapping sessions; flag implausible (>16h) session interval

E. Lifestyle and goals

Entity Key fields Source systems Update Data-quality rules Linkage keys
Nutrition Log meal time, dish (Indian dish DB ID), portion (katori/piece/grams), energy, macros, sodium, sugar, water, supplements In-app, Health Connect/HealthKit, receipts (opt-in) M / D Portion conversion table; dietitian-curated dish DB; receipts give “purchased”, not “consumed” dish ID + time
Wellness Goal goal type (steps, weight, HbA1c target set by clinician, sleep), target, start, review date, set-by User, clinician (via share) M Clinical targets only when entered by or confirmed with a clinician goal ID
Risk Factor tobacco, alcohol, activity level, family history, occupation exposure, travel User questionnaire M Time-stamped; re-ask periodically type
Mood Log (opt-in) score, tags, note (encrypted), screener scores User M Separate key; excluded from AI and sharing by default —

F. Insurance and money

Entity Key fields Source systems Update Data-quality rules Linkage keys
Insurance Policy insurer, product name, UIN, policy no., type (retail/group/government), policy period, policyholder, insured members, TPA, sum insured, floater flag, renewal date, document refs Policy schedule, CIS, e-card, (future) payer API / ABHA-policy link E Policy no. + insurer unique; group vs retail detection; renewal date sanity insurer + policy no.; member ID
Coverage (benefit) benefit type (IP, day-care, OPD, maternity, AYUSH, pre/post hospitalisation, ambulance, domiciliary), limit, sub-limit, co-pay %, deductible, room-rent cap, clause ref Policy wording, CIS E Every coverage item must hold a clause_ref (doc + page + text span) or it is not shown as fact policy ID + benefit code
Exclusion type (permanent, specific, PED), description, code mapping (ICD/procedure where possible), clause ref Wording, CIS E Same clause-ref rule policy ID + exclusion ID
Waiting Period type (initial, PED, specific disease/procedure, maternity), duration, start date (incl. portability credit), end date Wording, CIS, user (continuity of cover) E / M Compute end dates from the inception date of continuous cover; ask the user to confirm portability history policy ID + type
Premium amount, due date, paid date, payer, mode, tax-deduction eligibility (information only) Premium receipts, schedule E — policy ID + period
Claim claim no., type (cashless/reimbursement), member, encounter link, admission/discharge dates, diagnosis, hospital, amount claimed/approved/paid, deductions[], status history, denial reason (text + taxonomy code), TAT Claim letters, TPA emails, user updates; (future) payer feed E / M Status transitions validated; deductions sum check; link to encounter and bills insurer + claim no.
Expense category (consult, diagnostics, pharmacy, hospitalisation, device, therapy), amount, date, payer (self/insurer/employer), receipt ref Invoices, bills, claims E Avoid double counting: reimbursed amounts offset, not duplicated receipt ID

G. Care management

Entity Key fields Source systems Update Data-quality rules Linkage keys
Care Gap rule ID, guideline source, due date, status (due/overdue/done/dismissed), evidence refs, coverage status Insight engine (T5) + guidelines (T6) D Each gap cites rule + guideline + input facts; user can dismiss with reason rule ID + person
Alert / Flag type (trend, interaction, duplicate therapy, red-flag), severity, message, evidence, acknowledgement Insight engine E Severity thresholds signed off by the clinical safety officer —

6.4 Relationships (edge list)

Person -[HAS_IDENTITY]-> ABHAIdentity
Person -[RELATED_TO {type, guardian, scope}]-> Person (Dependent)
Person -[GRANTED]-> ConsentRecord -[COVERS]-> (any node)
Person -[HAD]-> Encounter -[AT]-> Facility ; Encounter -[WITH]-> Clinician
Encounter -[RECORDED]-> Condition | Procedure | Observation | MedicationRequest | Immunization
DiagnosticReport -[CONTAINS]-> Observation(LabResult)
MedicationRequest -[FULFILLED_BY]-> MedicationDispense -[EVIDENCED_BY]-> Expense
Person -[HAS_DEVICE_DATA]-> WearableMetric | Activity | Sleep | Vital
Person -[LOGGED]-> NutritionLog | MoodLog
Person -[INSURED_UNDER {member_id, role}]-> InsurancePolicy
InsurancePolicy -[PROVIDES]-> Coverage ; -[EXCLUDES]-> Exclusion ; -[IMPOSES]-> WaitingPeriod ; -[BILLED_AS]-> Premium
Claim -[UNDER]-> InsurancePolicy ; Claim -[FOR]-> Encounter ; Claim -[SETTLES]-> Expense
CareGap -[DERIVED_FROM]-> (Condition | Observation | RiskFactor | Person.age/sex)
CareGap -[COVERED_BY?]-> Coverage | Exclusion | WaitingPeriod   (Coverage-to-Care link)
Alert -[EVIDENCE]-> (any clinical node)
Any node -[SOURCED_FROM]-> SourceArtifact (FHIR bundle | Document page | Device | Manual)

6.5 Data-quality rules

Situation Rule User experience
Duplicate records (same report via ABDM and PDF) Match on person + test/LOINC + collected time (±24h) + value (exact) + lab; keep T1, link T2 as duplicate_of Single entry with “also in your uploaded report”
Duplicate encounters Person + facility (HFR or fuzzy name) + class + date window Merge suggestion; auto-merge only for T1 + T1 with same care context
Conflicting values (same test, same time, different values) Preserve both; mark conflicts_with; exclude from trend computation until resolved “Two different values found — which is correct?” with both source images
Unit mismatch Convert using a UCUM-validated factor; keep original; if the unit is missing, infer only when the value range is unambiguous, else hold Shows original and converted
Reference-range mismatch across labs Store per-result range; trend charts show range bands per lab Note: “Different labs use different normal ranges”
Missing data Distinguish unknown vs negative vs not applicable; care gaps fire only when evidence is absent and the source coverage is adequate “We don’t have a record of this test — have you done it elsewhere?”
OCR/document uncertainty Field confidence < threshold (e.g. 0.90 for numeric lab values, 0.98 for drug names/doses — thresholds set by clinical safety review) → requires user confirmation; low-confidence data excluded from insights Side-by-side crop of the source with an editable value
Handwritten prescriptions Never auto-confirm; drug names checked against the brand dictionary; dose/frequency confirmation required Guided confirmation; option to ask the pharmacist/doctor
Identity mismatch (document name ≠ person) Hold the document; ask “Whose report is this?” (common in family phones) Assign to the right family member
Partial ABDM bundles Track per-care-context delivery status; retry within consent validity; reconcile “3 of 4 records received from Hospital X”
Revoked/expired consent Stop processing; remove from views; crypto-shred or delete per retention policy; recompute derived insights without the data Confirmation with what was deleted
Stale data Last-updated per domain; insights note data age “Based on reports up to March 2026”

7. User Journeys and Experience Design

7.1 Design principles for India

Principle Practical design
Vernacular-first UI in English + Hindi at launch, then Tamil, Telugu, Marathi, Bengali, Kannada, Gujarati, Malayalam (prioritise by pilot geography). Medical terms are shown with a plain-language gloss (“HbA1c — 3-month average sugar”). Use speech input and read-aloud; consider Bhashini for ASR/TTS/translation [S58] ⚠ Validate API terms.
WhatsApp-native ingestion “Forward reports to our WhatsApp number” (WhatsApp Business Platform, user-initiated, opt-in recorded); unique email inbox per user (ravi.k7@inbox.<brand>.in); Android share sheet target.
Paper-to-digital Camera capture with edge detection, multi-page stitching, glare warnings and offline queueing for poor connectivity.
Family-centred Profile switcher on every screen; colour + photo per family member; “whose report is this?” prompt.
Low digital literacy Large tap targets, icon + text, minimal typing, assisted mode, voice guidance; never more than one decision per screen in consent flows.
Low bandwidth / low-end Android APK size budget; lazy-load charts; on-device caching (encrypted); SMS fallback for OTP and critical reminders.
Trust signals Source badges on every item (Verified clinical / From your document / Device / You entered); “Why am I seeing this?” on every insight.
Clinical safety Emergency banner and helplines available from every screen (§9.4); no diagnosis language.

7.2 Journey 1: Onboarding and ABHA linking

Step User sees / does System Consent / safeguard Failure handling
1 Chooses language; sees a 3-card value tour (“records, insurance, family”) Locale set — —
2 Mobile number + OTP Account created Short notice + link to full notice; consent P1 (vault) OTP retry, SMS fallback, IVR fallback
3 “Do you have an ABHA?” → Yes: log in with ABHA address/number + OTP · No: create ABHA (Aadhaar or other supported method) · Skip ABDM M1 / PHR APIs [S59][S60] Explain what ABHA is and is not (“ABHA links records; it does not store them centrally”) [S1] Aadhaar OTP failure → alternate method or skip; never block the app
4 “Find my records”: pick hospitals/labs (search with HFR data; “near me”; recently visited) Discovery requests to HIPs Per-HIP opt-in; date range selector (default: last 3 years) “No records found” → explain likely reasons (name mismatch, facility not ABDM-enabled) + upload option
5 Approve consent in-app (PHR) Consent artifact issued; data request; encrypted push P2 + ABDM artifact (PATRQT, STORE, explicit erase date) Partial delivery tracking
6 “Add what you have”: upload/forward reports, policy, prescriptions Document pipeline P1, P3, P4 shown as separate toggles Offline queue
7 Add family members (optional) Dependent profiles / invites §7.8 —
8 First Snapshot appears (even if partial) Insight engine Labels on every item “We’re still receiving records from 2 hospitals”

Success metrics: onboarding completion; ABHA link rate; HIP discovery success rate; time-to-first-value (first parsed record).

7.3 Journey 2: Connecting wearables and health apps

  1. Settings → Connect devices. The list shows “Phone health data (Health Connect / Apple Health)” first, then “Other apps (Oura, Withings, Fitbit…)” in later phases.
  2. Data-type picker. Steps, sleep, heart rate, weight and BP are pre-selected only if the user’s goals need them. Each type has a one-line “why we need this”. Minimisation applies throughout.
  3. The OS permission dialog is shown (Health Connect / HealthKit) [S24][S26]. If denied, the app explains how to re-enable it, with no nagging [S45].
  4. Optional history import (Health Connect requires a separate permission for data older than 30 days before the grant) [S26] ⚠ Validate.
  5. Background sync schedule. Daily aggregates by default; battery-friendly.
  6. Source priority. If both watch and phone count steps, the user picks a primary source. Dedup rules apply.
  7. Device data is labelled “Measured by your device — not a medical test”.

7.4 Journey 3: Uploading an insurance policy and claims

Step Detail
1. Add policy Upload PDF / forward insurer email / photograph the e-card. Select “Personal” or “Employer (group)”.
2. Parse Policy parser extracts schedule fields + CIS + benefit table + exclusions + waiting periods, each with clause references.
3. Confirm User confirms insured members (maps to family profiles), inception date of continuous cover (for waiting periods/portability) and TPA.
4. Coverage card Sum insured · balance (if claims known) · room-rent cap · co-pay · key sub-limits · waiting periods with end dates · top exclusions · renewal date. Every line has “See clause” (opens the PDF at the highlighted span).
5. Add claim Upload pre-auth letter, discharge bill, settlement or rejection letter; or forward TPA emails.
6. Claim tracker Timeline: intimation → pre-auth → approval/query → discharge → final approval → settlement; deductions explained line by line (“Non-payable consumables ₹1,840 — Clause 4.2, List I”); denial reason mapped to a taxonomy with suggested next steps (e.g. “insurer’s grievance cell → Insurance Ombudsman”, information only) [S18] ⚠ Validate grievance path wording.
7. Spend Claims feed into the OOP spend view (claimed vs paid vs self-paid).

Guardrail: coverage answers are informational interpretations of the policy document, not a guarantee of claim admissibility. Show this label on every answer. Never recommend buying, switching or upgrading a policy unless the IRDAI licensing question is resolved (§9.6).

7.5 Journey 4: Uploading lab reports and prescriptions

  1. User shares a PDF/photo (in-app, WhatsApp-forward, email-forward, share sheet).
  2. Classification. Document type (lab report, prescription, discharge summary, bill, policy, other), person match (name/age/sex vs family profiles) and language.
  3. Extraction.
    • Lab report: test rows → name, value, unit, range, flag → LOINC mapping.
    • Prescription: drug lines → brand → molecule, strength, dose, frequency, duration; doctor name/registration → HPR lookup.
  4. Confidence routing.
    • High confidence: auto-accepted as T2 “unconfirmed”.
    • Below threshold: “Please check these 3 values”, with crops of the original.
    • Handwritten drug lines always need confirmation.
  5. Results.
    • The timeline is updated.
    • Trend charts are updated (only confirmed or high-confidence values).
    • The medicine list is updated and duplicate-molecule check runs.
    • A refill estimate is set if quantity is known.
  6. Explain. Plain-language explanation of each test (reference knowledge, T6). Values outside the lab’s range are marked “outside the lab’s reference range” with “Discuss with your doctor”. There is no interpretation of what the abnormality means for the individual.

7.6 Journey 5: Generating a “Health Financial + Clinical Snapshot”

A single shareable page (in-app, PDF export, and a time-bound doctor-share link) with these sections:

Section Content Tier labels
Header Name, age, sex, blood group, ABHA address (masked), emergency contact, last updated T4/T1
Active conditions Each condition with first-recorded date and source T1/T2/T4
Current medicines Molecule + brand, dose, frequency, prescriber, last dispensed; duplicate-molecule flags T1/T2/T4
Allergies Substance + reaction + source; “unknown” vs “none” T1/T2/T4
Recent key results Last 3 values for condition-relevant tests (e.g. HbA1c, fasting glucose, lipids, creatinine/eGFR, TSH, Hb) with lab ranges and trend arrows T1/T2
Vitals & lifestyle 30/90-day BP average (if home readings), weight trend, steps, sleep T3/T4
Vaccinations & care gaps Due/overdue items with guideline source T5/T6
Hospitalisations & procedures Dates, facility, discharge diagnosis T1/T2
Insurance Policies, sum insured, balance, waiting periods ending soon, key exclusions, renewal dates T2
Spend (12 months) Consultations, diagnostics, pharmacy, hospitalisation; reimbursed vs out-of-pocket T2/T5
Data coverage “Sources included: 2 hospitals (ABDM), 14 uploaded documents, Health Connect. Not included: …” —
Disclaimer “Summary of your records for information. Not a diagnosis. Verify with your clinician.” —

7.7 Journey 6: Asking questions (conversational assistant)

The assistant uses retrieval over the user’s consented graph only. It answers with citations to source items, labels trust tiers, and refuses or redirects out-of-scope clinical questions. Templates for the five reference questions:

Q1. “What are my recurring health costs?”

In the last 12 months you spent ₹38,420 on health. ₹24,100 went to medicines (mostly 3 monthly refills: metformin + glimepiride, atorvastatin, telmisartan) and ₹6,800 to diagnostics (4 lab visits). Your insurer reimbursed ₹0 of this, because your policy does not cover OPD expenses (Policy clause 3.1, see clause). Recurring monthly average: ≈ ₹2,000 for medicines. Sources: 11 pharmacy invoices, 4 lab bills, policy wording. Amounts from documents — please check any marked “unconfirmed”.

Q2. “Am I covered for this procedure?” (e.g. cataract surgery)

Based on your policy document (Insurer X, “Plan Y”, policy no. …1234): • Cataract surgery is listed under day-care procedures (Annexure II, see clause). • There is a specific waiting period of 24 months for cataract (Clause 4.2.b). Your continuous cover started 15 Mar 2025, so this waiting period ends on 15 Mar 2027 (computed). • A sub-limit of ₹40,000 per eye applies (Schedule, see clause). • Planned surgery usually needs pre-authorisation from your TPA (Medi Assist) at a network hospital for cashless treatment. Expected out-of-pocket (estimate): any cost above the sub-limit + non-payable items. This is an interpretation of your policy document, not a claim decision. Confirm with your insurer/TPA before admission.

Q3. “What preventive tests are due?”

For a 52-year-old woman with type 2 diabetes recorded: • HbA1c — last done 10 Sep 2026; commonly repeated every 3–6 months as advised by your doctor → next due around Dec 2026–Mar 2027 (guideline, T6). • Kidney function and urine albumin — no record found in your data. • Eye (retina) check — no record found. • Cervical and breast cancer screening — population screening for women 30+ is offered under the national NCD programme (NP-NCD) [S37]; no record found. • Blood pressure — home readings in the last 30 days: average 136/86 (device). Coverage note: preventive health check up to ₹5,000 every 2 policy years under your policy (see clause). These are reminders based on general guidelines and your records. Your doctor may advise differently. [Book with your doctor] [Mark as done elsewhere]

Q4. “How is my diabetes trend changing?”

HbA1c: 6.4% (12 Jan 2026, Lab A, verified clinical) → 6.9% (20 May 2026, Lab B, from your document) → 7.1% (10 Sep 2026, Lab A, verified clinical). The values have risen across your last three tests. The lab’s reference range is shown on the chart. Fasting glucose (home meter): 30-day average 138 mg/dL vs 124 mg/dL three months ago (device/self-entered). Context you logged: average steps fell from ~7,500 to ~4,200/day over the same period (device). Discuss these trends with your doctor. This is not a diagnosis or a treatment recommendation. [Share summary with my doctor]

Q5. “Which claims were rejected and why?”

1 of 3 claims in the last 2 years was rejected: Claim …778 (reimbursement, Oct 2025, ₹18,600). Reason stated in the TPA letter: “Treatment for condition within the 2-year specific-disease waiting period” → taxonomy: Waiting period. 1 claim was partially approved: ₹6,240 deducted (₹3,900 room-rent proportionate deduction; ₹2,340 non-payable consumables). What you can do (information): request the detailed reasoning from the insurer; use the insurer’s grievance process; if unresolved, the Insurance Ombudsman route is available [S18] ⚠ Validate. Sources: 2 TPA letters, 1 settlement advice.

Assistant refusal and redirect patterns

User asks Response pattern
“Do I have diabetes?” Shows recorded diagnoses and relevant results with sources; “Only a doctor can diagnose. Here’s a summary you can share.”
“Should I stop my BP tablet?” “Please don’t change medicines without your doctor’s advice.” Offers doctor-share; no dosing advice.
“I have chest pain now” Emergency interrupt (§9.4): call 112 / nearest emergency; no further chat until acknowledged.
“Which policy should I buy?” Information on own-policy gaps only; no product recommendation (licensing).
Scenario Flow Consent rule
Child (<18) Parent creates profile → verifiable parental consent captured [S11][S12] → child’s records, vaccinations, growth No behavioural tracking or targeted ads for children [S11]. At 18: notify and transfer control; the parent loses access unless the new adult re-grants.
Elderly parent with own phone Caregiver sends invite → parent opens link, chooses language, consents with OTP → selects scope for the caregiver: view, manage (upload, reminders), emergency-only Parent can revoke at any time; the activity log is visible to the parent.
Elderly parent without a smartphone Assisted consent: caregiver initiates; parent’s OTP on a basic phone or in-person confirmation; IVR read-out of notice in local language assisted_by recorded. Periodic re-confirmation. ⚠ Validate legal sufficiency with counsel.
Adult lacking capacity Lawful guardian flow per DPDP Rules for persons with disability [S12] Guardian documentation verified ⚠ Validate
Spouse Each adult has their own account; mutual sharing only by explicit grant No default spouse access.
Sensitive categories Mental wellbeing, sexual/reproductive health, HIV, and others the user tags as private are excluded from caregiver views by default The person can include them explicitly.
  1. Consent dashboard → select a consent (e.g. “Hospital X records”, “AI summaries”, “Health Connect”) → Withdraw.
  2. Show consequences in plain language: what stops, and what will be deleted. Offer “Download my data first” (FHIR bundle + PDFs + CSV).
  3. Execute:
    • ABDM: revoke via CM (or reflect revocation initiated in the ABHA app); stop pulls; erase data held under that artifact.
    • Platform: stop processing for that purpose.
    • Devices: revoke OS permission guidance + stop reads.
  4. Erasure:
    • Delete or crypto-shred raw and normalised data from that source.
    • Recompute derived insights.
    • Purge from search indexes, caches and analytics stores.
    • Schedule backup expiry (backups age out within a defined window; restored backups re-apply the deletion ledger).
  5. Exceptions disclosed. Data retained under legal obligation, e.g. transaction/billing records for tax law, and logs required by DPDP Rules / CERT-In for the prescribed period [S12][S20], is kept minimal and access-restricted.
  6. Receipt. Confirmation of what was deleted, with a reference number. The grievance officer’s contact is shown [S11].
  7. Account deletion. The same flow covers all data. Dependent profiles must be transferred or deleted first.

8. Insight and Summary Engine

8.1 Architecture of insights

Graph (consent-filtered view)
   │
   ├── Deterministic rules engine  (care gaps, waiting-period math, OOP, refill dates, duplicate molecule)
   ├── Statistical trend module     (direction, rate, variability; no prediction in MVP)
   ├── Knowledge services (T6)      (guidelines, drug knowledge, policy clauses, test explanations)
   └── LLM layer (constrained)      (summarise, explain, translate, answer Qs — ONLY over retrieved facts)
            │
            └── Safety filter → Citation checker → Tier labeller → UI

8.2 Capability specifications

Capability Inputs Logic Output example Label & guardrail
Health timeline Encounters, conditions, procedures, meds, labs, claims Chronological merge; group by episode (admission ± related tests/claims) “Mar 2026 — Admitted, City Hospital, 3 days: dengue fever (discharge summary). Claim approved ₹42,300.” Source badge on every event
Clinical summary Active conditions, meds, allergies, immunizations, latest results Active = clinical status active or recent med/test evidence; “unverified” section for T4-only See Snapshot §7.6 No new diagnoses inferred; “possible” items listed as “mentioned in documents”
Abnormal value highlighting Lab results + lab reference range Outside lab range → mark; repeated outside range → “repeatedly outside range” “LDL 168 mg/dL — above the lab’s reference range” No severity grading for the individual in MVP (clinician-reviewed critical-value list only triggers “contact your doctor promptly”)
Trends Time series (≥3 points) Direction via simple regression; change vs prior; variability (BP SD) “HbA1c has risen across your last 3 tests” Never “your diabetes is worsening”; wording approved by the clinical safety officer
Fitness & lifestyle summary Steps, active minutes, sleep, weight, nutrition logs, goals Weekly/monthly aggregates vs user goals; streaks “You averaged 6,100 steps (goal 8,000) and 6h 10m sleep this month” Device-measured label; goals user- or clinician-set
Preventive-care plan Age, sex, conditions, risk factors, last test dates; guideline rules (T6) Rule library sourced from Indian programmes (NP-NCD screening for 30+, UIP for children) [S37][S38] and clinician-approved chronic-care monitoring intervals ⚠ Validate guideline citations “Cervical cancer screening: no record found” “General guideline — your doctor may advise differently”; user can dismiss
Medication intelligence Med list, dispenses, adherence logs MVP: duplicate-molecule detection, refill-due estimate, adherence summary. Phase 2: licensed DDI checks filtered to clinically significant severity “Two of your medicines contain the same ingredient (paracetamol)” “Check with your doctor or pharmacist” — never “stop”
Insurance summary Policy, coverage, exclusions, waiting periods, claims Clause-cited extraction; waiting-period end dates; sum-insured consumption; renewal reminders “PED waiting period ends 15 Mar 2027” “Interpretation of your policy document, not a claim decision”
Coverage-to-Care gap Care gaps × coverage/exclusions/waiting periods × network Join a recommended service to benefit clauses “Retina check due · OPD not covered · est. ₹800–1,500 out-of-pocket (estimate)” Estimates shown as ranges with basis
Health spend analytics Expenses, claims, premiums Category totals; reimbursed vs OOP; recurring detection; annual premium vs claims “Net OOP in FY 2025-26: ₹52,300 (excluding premiums)” Amounts from documents; information only, no tax advice
Claim analytics Claims, letters Status history, TAT, deduction categories, denial taxonomy “Room-rent proportionate deduction caused 63% of your deductions” Information only
Risk flags Rule triggers (critical values, red-flag symptoms, dangerous combinations) Clinician-approved rules with severity levels “Your report shows a value the lab marked critical. Contact your doctor today.” Informational; escalation path; logged

8.3 Care-gap rule library: structure

rule_id: CG-DM-UACR-01
title: "Urine albumin (kidney check) for people with diabetes"
applies_if:
  condition_any: [SNOMED:44054006]          # Type 2 diabetes mellitus
evidence_of_completion:
  loinc_any: ["14959-1", "9318-7"]         # microalbumin / albumin-creatinine ratio (validate codes)
interval_months: 12                          # set by clinical governance
guideline_ref: "Clinician-approved, aligned to national/Indian society guidance (cite)"
message_key: cg.dm.uacr.due
coverage_lookup: [benefit:OPD, benefit:annual_health_check]
severity: reminder
owner: clinical_safety_officer
version: 1.2

The LOINC and SNOMED codes above are illustrative. They must be validated by the clinical informatics lead against current LOINC/SNOMED CT releases and NRCES value sets [S6][S50].

8.4 Explainability standard (every insight)

Each insight card must show:

  1. What: the statement.
  2. Based on: linked source items, with tier badges.
  3. How: the rule or method in plain words.
  4. Confidence / limitations: missing sources, unconfirmed values.
  5. What you can do: a non-directive action (discuss with doctor, book test, check policy clause).
  6. Feedback: “Is this wrong?”, which feeds the correction workflow.

Example (corrected from v1.0):

HbA1c trend. Your HbA1c was 6.4% on 12 Jan 2026 and 7.1% on 10 Sep 2026 (both from Lab A, verified clinical). Both values are above the lab’s reference range (4.0–5.6%), and the later value is higher. Method: comparison of your two most recent lab results. Limitation: results from other labs may not be included. Please discuss this trend with your doctor; this is not a diagnosis.

9. Trust, Safety, Privacy, and Compliance

Legal-advice flag. This section maps obligations for product design. It is not legal advice. Engage Indian counsel with DPDP, health-regulatory and insurance expertise before launch, and obtain written opinions on the items marked ⚖ Opinion required.

9.1 Regulatory map

Regime Applies to this product because… Key obligations / risks Status / validation
DPDP Act, 2023 [S11] The company processes digital personal data (including health data) of Data Principals in India Notice; consent or legitimate use; purpose limitation; accuracy; reasonable security safeguards; breach intimation; erasure; children’s data; grievance redressal; Data Principal rights (access, correction, erasure, nomination); penalties up to ₹250 crore for failure to take reasonable security safeguards [S11] Act in force in phases via the Rules
DPDP Rules, 2025 [S12] Operational detail for the Act Notice content; Consent Manager registration; breach intimation to each affected person and to the Board (detailed report within 72 hours, as understood); reasonable security safeguards (encryption/masking/tokenisation, access control, logging and monitoring, backups); log retention of at least one year; verifiable parental consent; exemptions for healthcare professionals/clinical establishments in specified cases; Significant Data Fiduciary duties Notified November 2025; phased commencement, with most fiduciary obligations ~18 months after notification ⚠ Validate exact dates and rule numbers
ABDM Health Data Management Policy [S16] and ABDM participation terms The product participates as PHR/HIU (and HIP) Consent per ABDM artifact; purpose limitation; erasure at dataEraseAt; security standards; audit; no use beyond consent ⚠ Validate latest HDM Policy version and HIU storage rules Condition of ABDM integration
CERT-In Directions (28 April 2022) [S20] The company is a body corporate / service provider operating IT systems in India Report specified cyber incidents to CERT-In within 6 hours; maintain ICT logs for 180 days within India; synchronise clocks to NTP sources designated by NIC/NPL In force ⚠ Validate any amendments
IT Act 2000 s.43A / SPDI Rules 2011 Health data is “sensitive personal data” under the SPDI Rules Reasonable security practices (e.g. ISO 27001-type), consent, disclosure limits Continue until DPDP provisions replace them (the DPDP Act omits s.43A on commencement) ⚠ Validate commencement
Telemedicine Practice Guidelines 2020 [S21] If the product connects users to doctors or enables tele-consults Only registered medical practitioners may consult; platforms must verify RMPs; AI/ML platforms may not counsel patients or prescribe Applies to any tele-consult feature; shapes the AI assistant’s scope
Medical Devices Rules 2017 / CDSCO [S22] Software intended for diagnosis, monitoring or treatment may fall within the medical-device definition Risk classification, licensing, quality management (ISO 13485), clinical evaluation ⚖ Opinion required on whether risk flags, interaction alerts or trend scoring are SaMD; design MVP to stay “informational”
Mental Healthcare Act 2017 [S23] Optional mental-wellbeing features Right to confidentiality of mental-health information Design: separate consent, exclusion from sharing by default
IRDAI regulations [S18][S19][S47] Insurance explainers, claim assistance, any renewal/recommendation Product and policyholder-protection norms (CIS, claim timelines); intermediary registration if soliciting/recommending/selling insurance ⚖ Opinion required before any “renew/compare/buy” feature
Consumer Protection Act 2019, E-Commerce Rules 2020, CCPA Dark Patterns Guidelines 2023 [S45] Consumer-facing subscription app Fair pricing disclosure; easy cancellation; no dark patterns (false urgency, basket sneaking, confirm-shaming, forced action, subscription traps, nagging, etc.) Applies to consent and subscription UX
Drugs and Cosmetics Act/Rules; Drugs and Magic Remedies (Objectionable Advertisements) Act Pharmacy referrals; health claims in marketing No facilitation of unlawful sale of prescription drugs; no prohibited health advertising claims Relevant to marketplace features
Clinical Establishments Act 2010 Only if the company itself provides clinical services Registration under state adoption Likely not applicable to a data platform ⚠ Validate per state
ICMR Ethical Guidelines for AI in Biomedical Research and Healthcare (2023) [S44] AI features; any research use Autonomy, safety, data privacy, accountability, transparency, ethics-committee oversight for research Adopt as governance baseline

9.2 DPDP obligations translated into product controls

DPDP principle Product control Owner Evidence artefact
Notice (s.5) Layered, multilingual notice per purpose; versioned Legal + Design Notice registry with hashes
Consent (s.6) Purpose catalogue (§5.4), granular toggles, consent ledger, withdrawal as easy as grant Product + Eng Consent ledger; UX test records
Legitimate uses (s.7) Narrow use, e.g. a medical emergency basis for showing emergency info; do not stretch it to cover analytics Legal Basis register per processing activity
Accuracy (s.8(3)) Correction workflow; conflict surfacing; source display Product Correction logs
Security safeguards (s.8(5)) Controls in §9.3 Security ISMS, pen-test reports
Breach intimation (s.8(6)) Incident runbook (§9.5) Security + Legal Incident register
Erasure (s.8(7)) Retention schedule + crypto-shredding Eng Deletion ledger
Grievance (s.8(10), s.13) In-app grievance with SLA tracking; published grievance officer Support Ticket metrics
Children (s.9) Verifiable parental consent; no tracking/behavioural monitoring/targeted ads for children Product + Legal VPC records
Rights (s.11–14) Self-serve access/export (FHIR + PDF), correction, erasure, nomination of a person to exercise rights on death/incapacity Product Rights-request logs
Significant Data Fiduciary (s.10) Prepare for possible notification as SDF (large volumes of sensitive health data): DPO in India, periodic DPIA, independent audit Legal DPIA, audit reports ⚖ Opinion required
Cross-border transfer (s.16) Host in India; restrict vendors (LLM/OCR) to India regions or contractually controlled processing; monitor any government restriction list Eng + Legal Data-flow map; vendor register

9.3 Security architecture controls

Control area Implementation
Data classification C4: raw clinical documents, FHIR bundles, mental-health data, identifiers (ABHA number, policy numbers). C3: normalised clinical/insurance graph. C2: derived insights. C1: de-identified aggregates.
Encryption TLS 1.2+ (1.3 preferred) in transit; AES-256 at rest; envelope encryption with per-person, per-source data keys in a cloud KMS/HSM (India region); ABDM transfers via per-request ECDH Curve25519 key pairs [S60]; mental-health data under a separate key hierarchy.
Crypto-shredding Erasure of a source = destroy its data key; backups become unreadable for that scope.
Identity & access Customer: OTP + device binding + optional biometric app lock. Workforce: SSO + MFA + just-in-time privileged access; no standing human access to C4 data; break-glass with dual approval and logging.
Authorisation Attribute-based access control: subject × purpose × consent × data category; enforced in a central policy engine (e.g. OPA-style) on every API, export and analytics job.
Tokenisation / masking ABHA number, Aadhaar-linked fields (never store the Aadhaar number), policy numbers, mobile numbers tokenised; masked in logs, support tools and analytics.
Audit Immutable, append-only audit log (WORM storage) of consent, access, disclosure, export, correction, revocation and deletion events; user-visible access log.
Logging & monitoring SIEM; anomaly detection on bulk access; log retention ≥ 180 days in India (CERT-In) and ≥ 1 year where DPDP Rules require [S12][S20]; time sync to designated NTP.
Application security Secure SDLC; SAST/DAST; dependency scanning; mobile app hardening (root/jailbreak detection, certificate pinning, encrypted local storage); OWASP MASVS alignment; annual third-party pen test + ABDM WASA [S59].
Vendor governance DPDP-aligned processing agreements; sub-processor register; security assessment; data-residency clauses; no-training clauses for LLM/OCR vendors; deletion SLAs.
Resilience Multi-AZ in an India region; encrypted backups with tested restore; RPO/RTO targets; DR in a second India region.
Certifications (target) ISO/IEC 27001 and 27701; SOC 2 Type II for B2B buyers (optional).

9.4 Clinical safety framework

Element Requirement
Clinical safety officer Named, registered medical practitioner accountable for clinical content, rules and thresholds
Hazard log Identify hazards (e.g. misread drug dose, missed critical value, false reassurance, outdated guideline, wrong-person record), assess severity × likelihood, define mitigations, track residual risk
Intended-use statement “Information and organisation of personal health records; general wellness and preventive reminders; not for diagnosis or treatment decisions.” Supports the SaMD position ⚖ Opinion required
Language rules Approved verbs: shows, recorded, outside the lab’s range, has risen, is due, discuss. Prohibited: you have, diagnosis, you should take/stop, safe, normal for you, cured
Critical values Clinician-approved list of lab-flagged critical values → prominent “contact your doctor today / seek care” message; no attempt at interpretation
Red-flag symptoms When a user types symptoms such as chest pain, breathlessness, stroke signs, severe bleeding, suicidal thoughts or loss of consciousness: emergency interrupt with 112 (national emergency), 108 (ambulance, in most states), and Tele-MANAS 14416 for mental-health support [S39] ⚠ Validate all numbers per state before release
No delayed care by design The app never says “no need to see a doctor”; absence of data is never presented as absence of disease
Human review High-risk outputs (critical values, interaction alerts at “major” severity) can be routed to a clinician-review queue in premium tiers
Post-market surveillance Track user-reported errors, near-misses and complaints; monthly safety review; incident reporting process

9.5 Breach response runbook (summary)

Phase Actions Clock
Detect & triage SIEM alert / report → incident commander; severity classification T0
Contain Revoke credentials/keys; isolate systems; preserve evidence ASAP
Report to CERT-In Reportable incident types Within 6 hours of noticing [S20]
Intimate Data Protection Board Initial intimation without delay; detailed report (facts, cause, mitigation, affected persons) Within 72 hours (or as extended by the Board) per DPDP Rules [S12] ⚠ Validate
Notify affected Data Principals Plain-language notice: nature, extent, likely consequences, mitigation, steps to protect themselves, contact Without delay [S12]
Notify ABDM / partners Per ABDM and partner agreements As contracted ⚠ Validate ABDM reporting requirement
Remediate & review Root cause; fixes; lessons learned; update hazard log/DPIA Post-incident
Feature Risk level Position
Show the user’s own policy terms, waiting periods, exclusions (from their document) Low Information service; label as interpretation; cite clauses
Claim tracker and deduction explainer Low–Medium Information; no representation before insurer on user’s behalf in MVP
Claim assistance (filing on the user’s behalf, follow-up with TPA) Medium Check whether this constitutes an intermediary activity or claims consultancy requiring authorisation ⚖ Opinion required
“Your cover may be inadequate” (coverage adequacy) Medium Information about own policy only; no product suggestion
Compare/recommend/sell/renew policies; earn commission High Requires IRDAI registration (e.g. corporate agent, web aggregator, broker) or partnership with a registered intermediary with clear disclosures [S47] ⚖ Opinion required
Share user data with insurers for underwriting High Not in roadmap. If ever considered: separate explicit consent, no penalty for refusal, legal review, and no use of data obtained for care purposes

9.7 Responsible AI controls

Control Implementation
Consent-aware retrieval The LLM context only receives facts from consents covering the clinical_summary / insurance_analysis purposes
Minimisation & redaction Direct identifiers (name, ABHA, phone, policy number, addresses) removed or pseudonymised before model calls; full histories never sent by default
Model hosting Prefer models hosted in India regions or self-hosted; contractual no-retention / no-training; logs within India ⚠ Validate vendor region availability
Grounding & citation Every generated sentence linked to a fact ID or rule output; unsupported sentences dropped; numbers copied from structured fields, never generated
Safety filters Red-flag detection before generation; prohibited-language filter after generation
Evaluation Golden test sets in English and Indian languages (lab explanations, policy Q&A, refusals); hallucination rate, citation accuracy and refusal correctness tracked per release; clinician review of samples
Auditability Store prompt template version, retrieved fact IDs, model version, output, and safety-filter results per response (retained per audit policy, access-restricted)
Transparency “AI-generated summary” label; “How this was made” link; easy error reporting
Low-confidence data Excluded from AI context unless user-confirmed

10. Technical Architecture

10.1 Logical architecture

 ┌──────────────────────── Channels ───────────────────────────────────────────────┐
 │ Android app (primary) · iOS app · Web (caregiver) · WhatsApp inbound · Email inbound │
 └───────────────┬─────────────────────────────────────────────────────────────────┘
                 │  (mTLS / OAuth2, device binding)
        ┌────────▼─────────┐
        │ API Gateway/BFF  │──── Policy Decision Point (consent × purpose × role)
        └────────┬─────────┘
   ┌─────────────┼─────────────────────────────────────────────────────────────┐
   │             │                    Core services                            │
   │  Identity & ABHA svc ── ABDM Gateway (V3) / HIE-CM ── HIPs                 │
   │  Consent svc + Consent Ledger (append-only)                                │
   │  HIU Data-flow svc (data-push endpoint, ECDH decrypt, ack, reconcile)      │
   │  Document Ingestion svc (upload, WA, email) → Document AI pipeline         │
   │  Device Sync svc (Health Connect/HealthKit on-device → batched upload)     │
   │  Partner Connectors (OAuth vendors; payer/TPA/employer feeds — later)      │
   │  Canonicalisation svc (FHIR R4 ⇄ Health Graph; terminology mapping)        │
   │  Health Graph store + Timeline/Query API                                   │
   │  Insight Engine (rules, trends, care gaps, coverage, spend)                │
   │  AI Gateway (redaction, grounding, safety filters, audit)                  │
   │  Notification svc (push, SMS, WhatsApp templates)                          │
   │  Sharing svc (time-bound doctor links; FHIR export)                        │
   │  Erasure & Retention svc (crypto-shred, purge, recompute)                  │
   └────────────────────────────────────────────────────────────────────────────┘
   ┌──────────────────────────── Data plane (India region) ─────────────────────┐
   │ Raw vault (encrypted blobs: FHIR bundles, PDFs, images)                    │
   │ FHIR store (canonical clinical/insurance resources)                        │
   │ Graph/relational store (person-partitioned) · Search index (per-person)    │
   │ Time-series store (device metrics, daily aggregates)                       │
   │ Audit log (WORM) · KMS/HSM (per-person/per-source keys)                    │
   │ De-identified analytics warehouse (consented P10/P11 only)                 │
   └────────────────────────────────────────────────────────────────────────────┘

10.2 ABDM HIU / PHR data flow (corrected sequence)

User (PHR UI)        Aggregator (HIU)            ABDM Gateway/CM           HIP
     │ select HIPs, range │                            │                    │
     │──────────────────▶ │ consent request (purpose,  │                    │
     │                    │ hiTypes, permission)──────▶│                    │
     │                    │◀───── ack (request id) ────│                    │
     │◀─ consent prompt (in PHR/ABHA app) ─────────────│                    │
     │── approve ─────────────────────────────────────▶│                    │
     │                    │◀─ consent notify (async callback: GRANTED,      │
     │                    │   artefact ids) ───────────│                    │
     │                    │── fetch artefact ─────────▶│                    │
     │                    │◀─ artefact (async) ────────│                    │
     │                    │── HI request (artefact id, │                    │
     │                    │   dateRange, dataPushUrl,  │                    │
     │                    │   keyMaterial: ECDH        │                    │
     │                    │   Curve25519 pubkey+nonce)▶│── HI request ─────▶│
     │                    │◀─ txn id (async) ──────────│                    │
     │                    │◀═══ encrypted FHIR bundles pushed to dataPushUrl ═│
     │                    │── decrypt (ECDH → HKDF → AES-GCM), validate,    │
     │                    │   store raw, canonicalise, ack per care context ─▶ (notify)
     │◀─ "3 of 4 records received" ─│                  │                    │
     │                    │ on revoke/expiry/dataEraseAt: stop + erase     │

Notes:

# 1. Create consent request (purpose codes per ABDM; validate against current spec)
req_id = abdm.consent_requests.init(
    patient_abha_address=person.abha_address,
    purpose_code="PATRQT",                 # self-requested (validate); CAREMGT for doctor-share
    hi_types=["DiagnosticReport", "Prescription", "DischargeSummary", "OPConsultation"],
    hips=selected_hip_ids,
    permission={
        "accessMode": "STORE",
        "dateRange": {"from": "2023-01-01T00:00:00Z", "to": "2026-10-21T00:00:00Z"},
        "dataEraseAt": "2027-10-21T00:00:00Z",
        "frequency": {"unit": "HOUR", "value": 1, "repeats": 0},
    },
)
ledger.record_pending(person.id, req_id, purpose="P2_abdm_self_access")

# 2. Asynchronous callback when the user approves or denies in the PHR/ABHA app
@callback("/v3/hiu/consent/request/notify")
def on_consent_notify(evt):
    verify_gateway_signature(evt)
    ledger.update_status(evt.consent_request_id, evt.status)          # GRANTED / DENIED / REVOKED / EXPIRED
    if evt.status == "GRANTED":
        for artefact_id in evt.consent_artefact_ids:
            abdm.consents.fetch(artefact_id)                          # result arrives via on-fetch callback
    elif evt.status in ("REVOKED", "EXPIRED"):
        erasure.schedule(scope=evt.consent_artefact_ids)

@callback("/v3/hiu/consents/on-fetch")
def on_artefact(evt):
    artefact = evt.consent
    ledger.attach_artefact(artefact)
    keys = crypto.new_x25519_keypair()                                # per request; never reused
    kms.hold_ephemeral(artefact.id, keys.private)
    abdm.health_information.request(
        consent_id=artefact.id,
        date_range=artefact.permission.dateRange,
        data_push_url=settings.DATA_PUSH_URL,
        key_material=keys.public_material(nonce=crypto.nonce32()),
    )

# 3. HIP pushes encrypted bundles to the data-push endpoint
@endpoint("/data/push", idempotent_on="transactionId+careContextReference")
def on_data_push(payload):
    priv = kms.get_ephemeral(payload.consent_id)
    for entry in payload.entries:
        bundle = crypto.decrypt_abdm(entry.content, priv, payload.keyMaterial)   # ECDH -> HKDF -> AES-GCM
        raw_ref = vault.put_encrypted(bundle, person_key(payload))
        result = fhir.validate(bundle, profiles="NRCES")
        nodes = canonicalise(bundle, provenance={
            "consent_id": payload.consent_id, "txn_id": payload.transactionId,
            "source": payload.hip_id, "raw_ref": raw_ref, "trust_tier": "T1"})
        graph.upsert(nodes, dedupe=True)
    abdm.health_information.notify(payload.transactionId, status="TRANSFERRED")
    insights.recompute(person=payload.person_id)

10.4 Document AI pipeline

Stage Function Build / buy note
Intake Virus scan, file type check, page split, de-duplication by hash Build
Pre-processing De-skew, de-noise, glare detection, orientation Buy/OSS
OCR Printed English + major Indian scripts; handwriting (low confidence) Buy/partner (evaluate India-region availability and accuracy on Indian lab/pharmacy layouts)
Classification Lab / prescription / discharge / bill / policy / CIS / claim letter / e-card / other Build (fine-tuned on own labelled data, with consent P10 or synthetic data)
Layout & table extraction Lab tables, bill line items, benefit tables Build + OSS
Clinical NER & normalisation Tests → LOINC, drugs → molecule/SNOMED, diagnoses → SNOMED/ICD-10, units → UCUM Build on licensed terminology + Indian drug dictionary
Policy clause extraction Benefits, limits, exclusions, waiting periods with spans Build (core IP)
Confidence scoring Field-level calibrated confidence Build
Human-in-the-loop User confirmation UI; internal review queue for premium/critical items (with consent and strict access controls) Build
Output Canonical nodes with provenance + page/bbox references Build

Accuracy targets (to validate in pilot): ≥ 98% field accuracy on confirmed lab values; ≥ 95% on printed prescriptions after confirmation; clause-citation precision ≥ 95% on policy Q&A. Track per lab chain and insurer template.

10.5 Technology choices (indicative)

Layer Options Notes
Cloud Hyperscaler India regions (Mumbai/Hyderabad/Delhi NCR/Pune) Keep all personal data in India; DR in a second India region
FHIR server HAPI FHIR (OSS) or a managed FHIR service available in an India region NRCES profile validation [S6]
ABDM integration Own implementation on V3, referencing NHA’s open-source ABDM-wrapper [S60] The wrapper is Spring Boot + MongoDB; evaluate for reuse vs reference
Graph/relational PostgreSQL (person-partitioned, row-level security) + edge tables; or a graph DB if query patterns justify it Simplicity favours Postgres for MVP
Time series Timescale/ClickHouse-class store for device data Daily aggregates by default
Search Per-person encrypted index Avoid cross-person indexes on C3/C4 data
Policy engine OPA-style ABAC Single enforcement point
Mobile Kotlin (Android) + Swift (iOS), or cross-platform with native Health Connect/HealthKit modules Health Connect and HealthKit need native code
Terminology SNOMED CT (India NRC), LOINC, ICD-10, UCUM, ATC [S50] Licence registrations
LLM India-hosted model endpoints with no-retention terms, or self-hosted open models Behind the AI Gateway
Messaging WhatsApp Business Platform via a BSP; SMS via DLT-registered templates (TRAI) Opt-in records; template approval

10.6 Non-functional requirements

NFR Target (MVP)
Availability 99.5% app/API; data-push endpoint 99.9% (ABDM pushes must not be lost)
Latency Timeline load < 2 s on 4G for 500 items; Snapshot < 5 s
Document turnaround Parsed result < 60 s for a typical 2-page lab report
Scalability Person-partitioned design; horizontal scaling of ingestion workers
Privacy 100% of reads pass the policy engine; zero C4 data in logs
Erasure Source-level erasure executed within a defined SLA; backups age out within the defined backup-retention window
Accessibility WCAG 2.1 AA-aligned; font scaling; screen-reader labels; read-aloud
Localisation All user-facing strings externalised; medical glossary per language reviewed by clinicians

11. MVP Scope and Roadmap

11.1 Source prioritisation (value × feasibility)

Source User value Feasibility MVP decision
Lab report PDFs/photos Very high (most common artefact; trends) High Include (core)
Prescriptions (printed + handwritten with confirmation) High (medicine list) Medium Include
Discharge summaries High (hospitalisation history) Medium Include
Insurance policy schedule + CIS + e-card Very high (paid wedge) High Include (core)
Claim letters / settlement advices High High Include
Pharmacy invoices High (medicines + spend) High Include
ABDM records (PHR + HIU) High where available; trust anchor Medium (certification on critical path) Include: launch with document flows if ABDM production access is still pending
Health Connect / HealthKit Medium (chronic-care context) High Include (steps, sleep, weight, BP, glucose, heart rate)
Manual entry (allergies, history, family history, vitals) Medium High Include
Vendor wearable clouds (Oura, Withings, Fitbit, Garmin) Low–Medium in India Medium Phase 2
Nutrition logging Medium (diabetes users) Medium (Indian dish DB needed) Phase 2
Mental-wellbeing check-ins Medium High technically / high governance need Phase 2 (after clinical safety sign-off)
Payer/TPA/employer feeds Very high Low (partnerships) Phase 2 (pilot partners)
NHCX (via provider SaaS / partner) High Low Phase 3
AA insurance data Medium Low–Medium Phase 2 discovery
Email-inbox scanning (Gmail API) Medium Low (restricted-scope assessment) [S33] Not planned; use forwarding instead

11.2 90-day MVP definition

Target users for the pilot: urban/semi-urban Android users aged 30–60 who manage a chronic condition, or who care for a parent with one, and hold at least one health insurance policy (retail or group). Pilot in 1–2 cities and 2 languages (English + Hindi; add one regional language if the pilot city requires).

Must-have features

# Feature Acceptance criteria
F1 Account, language choice, consent onboarding with purpose catalogue All purposes separately toggleable; consent ledger records every event; notice in 2 languages
F2 ABHA create/login + record discovery/linking + self-requested fetch (PHR + HIU) Sandbox-certified flows; production when credentials are granted; partial-delivery tracking
F3 Document ingestion: upload, camera, share-sheet, WhatsApp-forward, email-forward Classification + person-assignment prompt; offline queue
F4 Lab report, prescription, discharge summary, pharmacy invoice parsing with confirmation UI Confidence routing; source crop view; corrections logged
F5 Policy + CIS + e-card parsing; coverage card; waiting-period calculator; clause citations Every coverage statement cites a clause span
F6 Claim tracker (document-driven) + deduction/denial explainer Status timeline; denial taxonomy; deductions sum check
F7 Unified timeline + clinical summary + Snapshot (PDF + time-bound doctor-share link) Tier badges; data-coverage statement; share link expiry + access log
F8 Trends for key labs and vitals; duplicate-molecule check; refill estimates; medicine reminders Rules unit-tested; clinical safety officer sign-off on wording
F9 Care-gap reminders (limited, clinician-approved rule set: diabetes, hypertension, NP-NCD 30+ screening, child UIP) Each reminder cites rule + guideline; dismissible
F10 Spend dashboard (OOP vs reimbursed; recurring costs) Reimbursements offset correctly
F11 Family profiles: minors (verifiable parental consent) + adult invite with scoped caregiver grants Sensitive categories excluded by default
F12 Health Connect / HealthKit sync (selected types) Per-type permission; daily aggregates
F13 Consent dashboard, data export (FHIR + PDF), withdrawal + erasure, account deletion Erasure executes end-to-end, including derived insights
F14 Grounded Q&A assistant for the five reference question types (§7.7), with refusal patterns Golden-set pass thresholds; citations on every answer
F15 Emergency interrupt and helpline access Red-flag terms tested in both languages

Explicitly out of MVP: provider cohort dashboards; insurer/TPA integrations; policy recommendations or sales; drug-interaction alerts beyond duplicate molecules; mental-health screeners; nutrition logging; vendor wearable clouds; UHI booking; marketplace referrals.

MVP data model: the Section 6 entities Person, Relationship, ABHA Identity, Consent Record, Caregiver Grant, Facility, Clinician, Encounter, Condition, Allergy, Immunization, Medication, Medication Dispense, Lab Result, Diagnostic Report, Vital, Wearable Metric (daily), Activity, Sleep, Insurance Policy, Coverage, Exclusion, Waiting Period, Premium, Claim, Expense, Care Gap, Alert, all with the common provenance envelope.

MVP consent flows: onboarding purposes (P1, P3, P4, P5, P6 opt-in); ABDM self-access (P2, PATRQT/STORE); doctor-share (P7, VIEW, time-bound); caregiver grants (P8); P10/P11/P12 off and not prompted in MVP.

MVP workstreams and exit criteria (scheduling is left to the team)

Workstream Key deliverables Exit criterion
Compliance & legal DPIA; notices; consent catalogue; processor agreements; legal opinions (IRDAI, SaMD, HDM retention) Opinions received; DPIA signed by DPO
ABDM integration Sandbox registration (V3); M1 + PHR flows + M3; functional tests; WASA audit; sandbox-exit application Sandbox test cases passed; audit findings closed; production request filed
Document AI Classifiers; lab/prescription/invoice/policy/claim extractors; confirmation UI Accuracy targets met on pilot corpus
Graph & insights Canonical model; dedup/conflict rules; rule library v1; trends; spend Clinical safety officer sign-off; unit-test coverage on rules
App & UX Android-first app (iOS if resourced), 2 languages, accessibility Usability tests with low-literacy and elderly participants
Security KMS envelope encryption; ABAC; audit log; pen test Pen test with no open critical/high findings
Pilot ops Recruitment, support desk, grievance workflow Pilot cohort onboarded

11.3 MVP success metrics

Metric Definition Pilot target (hypothesis to validate)
Activation % of sign-ups adding ≥1 record within 24h ≥ 60%
ABHA link rate % of users who link/create ABHA ≥ 40%
HIP fetch yield Average consented HIP connections per ABHA-linked user; % yielding ≥1 record Measure (baseline)
Structured share % ABDM records with coded Observations vs PDF-only Measure (baseline)
Parse accuracy Field accuracy after confirmation (labs / prescriptions / policies) ≥ 98% / ≥ 95% / ≥ 95%
Time to Snapshot From first upload to Snapshot < 10 minutes
Insurance engagement % users with a parsed policy who open a coverage answer ≥ 50%
Care gaps Gaps surfaced per user; % marked done within the pilot Measure
Doctor-share % of users sharing a Snapshot with a clinician ≥ 15%
Retention D30 retention (caregiver and chronic segments) ≥ 35%
Trust % of users reporting they understand who can see their data (survey) ≥ 80%
Safety Clinically significant errors reaching users 0 tolerated; all near-misses reviewed
Conversion (if paywall tested) Premium/family plan conversion Measure by price-point A/B

11.4 Phase 2 (after MVP validation)

Theme Capabilities
Insurance depth Payer/TPA/employer pilot integrations (claim-status events, e-cards, network lists); AA insurance-data discovery; cashless pre-admission checklist; renewal-risk alerts (information only)
Clinical depth Licensed drug-interaction checks with clinician-approved severity filtering; expanded care-gap library (cardiac, thyroid, CKD, maternal, adult vaccination); clinician-review queue (premium)
Lifestyle Indian-dish nutrition logging (IFCT-based, dietitian-curated); vendor wearable clouds (Oura, Withings, Fitbit, Garmin with approval); CGM CSV import
Mental wellbeing Opt-in mood check-ins and validated screeners with crisis routing (after safety sign-off)
ABDM Act as HIP to re-share user documents and wellness records; Scan & Share support
B2B2C Employer benefits-navigation offering (aggregated reporting only)
Languages 6–8 Indian languages with clinician-reviewed glossaries; voice-first navigation

11.5 Phase 3

Theme Capabilities
Provider SaaS Clinician dashboard for patient-shared summaries; consented care-gap lists; NHCX participation as provider technology partner (eligibility/pre-auth/claim) [S59]
Care navigation UHI-based booking for diagnostics/consults to close gaps [S13]; neutral ranking with conflict disclosures
Insurance (regulated) If licensing obtained or via a registered partner: compliant comparison/renewal journeys
Research & population health Opt-in, ethics-approved, de-identified research programme [S44]; aggregated benchmarks for employers/payers
Advanced analytics Validated risk models only with regulatory clearance where required (CDSCO)
Environment AQI-aware respiratory insights (opt-in, coarse location)

12. Build vs Partner Decisions

Capability Decision Rationale Candidate partner types Licence / regulatory note
ABDM integration (M1, PHR, M3, M2) Build (reference NHA’s open-source wrapper [S60]) Core competency; certification owned by the company NHA integration support ABDM certification + WASA audit
Consent ledger & policy engine Build Core trust IP — DPDP
Health Graph & canonicalisation Build Core IP; quality differentiator — Terminology licences
FHIR server Adopt OSS / managed Commodity HAPI FHIR; managed FHIR in India region Data residency
OCR engine Partner/buy Commodity; multilingual handwriting is hard Cloud OCR vendors (India region), Indian OCR specialists Processor agreement; no-training clause; residency
Document understanding (lab/prescription/policy/claim) Build on top of OCR Indian templates and clause extraction are differentiators Annotation vendors (under strict DPAs, synthetic/de-identified data) Consent P10 or synthetic data only
Indian brand→molecule dictionary License + curate Coverage and update burden Indian drug-information publishers Commercial licence
Drug-interaction knowledge base License Clinical liability; maintenance Global DDI knowledge-base vendors Commercial licence; clinical governance
Terminology (SNOMED CT, LOINC, ICD-10) License (free/affiliate) Standards NRCES (SNOMED CT India NRC) [S50]; Regenstrief (LOINC) Affiliate registrations
Nutrition database License + curate Indian dish coverage ICMR-NIN (IFCT) [S34]; academic datasets; dietitians Licence for commercial use ⚠ Validate
Wearable connectivity Build on Health Connect/HealthKit; partner with vendors later OS aggregators give broad coverage cheaply Garmin/Samsung partner programmes Platform policies [S25][S26]
Multilingual (ASR/TTS/translation) Partner Speed to multi-language Bhashini [S58]; commercial Indian-language AI vendors Clinician review of medical translations
LLM Partner/self-host behind own AI Gateway Rapid evolution; keep control layer in-house India-region model hosting; open-weight models No-retention, residency
WhatsApp / SMS Partner Regulated channels WhatsApp BSPs; DLT-registered SMS aggregators Opt-in; template approval; TRAI DLT
Payer/TPA/employer data Partner No public APIs Insurers, TPAs, employee-benefit platforms Data-sharing agreements; IRDAI considerations
Insurance distribution Partner (if pursued) Licensing IRDAI-registered intermediary ⚖ Opinion required
Tele-consult / clinician review Partner Regulated practice RMP networks/telemedicine providers compliant with TPG 2020 [S21] Telemedicine guidelines
Diagnostics/pharmacy fulfilment Partner Not core; conflict-of-interest risk Lab chains, licensed pharmacies Disclosures; drug-law compliance
Security operations Hybrid 24×7 coverage MSSP for SOC; CERT-In empanelled auditors CERT-In directions

13. Key Risks and Mitigations

# Risk Category Likelihood Impact Mitigation Owner
R1 Breach of health data Security Medium Severe Envelope encryption + crypto-shredding; no standing access; SIEM; pen tests; runbook (6h CERT-In, 72h Board) [S12][S20]; cyber insurance CISO
R2 Invalid or bundled consent (DPDP non-compliance) Privacy Medium High Purpose catalogue; unbundled toggles; versioned multilingual notices; consent receipts; UX audits against dark-pattern guidelines [S45] DPO
R3 Purpose creep (using care data for marketing, model training, underwriting) Privacy Medium High Policy engine enforces purpose; P10–P12 off by default; data-use review board DPO
R4 Over-retention of ABDM data beyond dataEraseAt Compliance Medium High Automated erasure scheduler tied to artifacts; audits; erasure tests in CI [S60] Eng lead
R5 Cross-border transfer via LLM/OCR vendors Compliance Medium High India-region processing; redaction; contractual controls; vendor register DPO
R6 Misread lab value or drug dose from OCR Clinical safety High High Confidence thresholds; mandatory confirmation for drug lines; plausibility checks; source crops; low-confidence data excluded from insights Clinical safety officer
R7 Misinterpretation of lab values / false reassurance Clinical safety Medium High Lab-specific ranges; approved language; no “normal for you”; “absence of data ≠ absence of disease” CSO
R8 LLM hallucination AI safety Medium High Rules-first; grounding + citation checker; numbers from structured fields; golden-set evaluation per release AI lead
R9 Delayed care / missed emergency Clinical safety Low–Medium Severe Red-flag interrupts; helplines; no triage claims; critical-value prompts CSO
R10 Product classified as a medical device (SaMD) Regulatory Medium High Intended-use statement; informational design; early CDSCO-focused legal opinion; QMS readiness if needed [S22] Regulatory lead
R11 Insurance features deemed unlicensed intermediation Regulatory Medium High No recommendations/sales; legal opinion; partner with registered intermediary if needed [S47] Legal
R12 Low ABDM record yield (few HIPs, PDF-only bundles) Data access High Medium Document-first strategy; measure yield; HIP onboarding partnerships; re-request flows Product
R13 ABDM certification delays / API version changes Delivery Medium Medium Start sandbox early; build on V3; abstraction layer; track NHA advisories [S59][S60] Eng lead
R14 Insurer/TPA refuse data partnerships Commercial High Medium Value proposition for payers (complete documents, fewer queries); start with employer channel; document intelligence remains self-sufficient BD
R15 Wearable API changes (Google Fit shutdown, Fitbit migration) Technical High Low–Medium Health Connect/HealthKit as primary; vendor APIs optional [S26][S27][S28] Mobile lead
R16 Wrong-person records on shared family phones Data quality / privacy High High Person-match checks; “whose report is this?”; separate profiles; ABHA demographic checks Product
R17 Elderly/low-literacy users can’t give meaningful consent Trust / legal Medium High Vernacular audio notices; assisted consent with own-device confirmation; legal review of sufficiency Design + Legal
R18 Commercial bias in recommendations (marketplace) Trust Medium High Neutral ranking; disclosures; marketplace out of MVP; ethics review CEO
R19 Becoming an attractive central data honeypot Security / trust Medium Severe Minimise retention; user-controlled deletion; per-person keys; optional on-device-only mode for sensitive categories CISO
R20 DPDP Rules interpretation changes / SDF designation Regulatory Medium Medium Regulatory watch; design to strictest reading; DPIA cadence DPO
R21 Incompatible formats / terminology gaps (local test names, Indian brands) Data quality High Medium Terminology team; mapping tables with review; feedback loops from corrections Clinical informatics
R22 Trust erosion from opaque insights Trust Medium High Explainability standard (§8.4); source badges; easy corrections; transparency reports Product

14. Open Questions and Validation Plan

# Question Why it matters Validation experiment Decision informed
Q1 Will consumers link ABHA and approve ABDM consents? Determines the weight of ABDM vs documents Pilot with 100–500 users: measure ABHA link rate, consent approval, drop-off per step ABDM prominence in onboarding
Q2 What is the actual ABDM record yield and structure? Determines parsing and insight design Instrument HIP discovery/fetch: % with records, records/user, % coded vs PDF; break down by facility type Investment in FHIR vs PDF pipelines
Q3 Which sources create the most perceived value? Prioritisation In-pilot ranking + usage analytics across lab PDFs, policies, claims, pharmacy invoices, wearables Phase 2 priorities
Q4 Is insurance clarity the paid wedge? Business model Fake-door + price A/B tests (individual vs family plan); willingness-to-pay interviews Pricing and packaging
Q5 Can policy parsing be accurate across insurers? Core feature feasibility Build a corpus of retail + group policy documents from top insurers (user-contributed with consent, or public sample wordings); measure clause-citation precision Insurer coverage roadmap
Q6 Do clinicians use patient-shared Snapshots? Provider value Pilot with 5–20 clinicians: time-to-history, completeness, satisfaction; qualitative interviews Phase 3 provider SaaS
Q7 Will insurers/TPAs/employers partner? Data access Structured interviews with insurers, TPAs, benefits platforms; propose a document-completeness pilot Phase 2 partnerships
Q8 Is NHCX ABHA↔︎policy linking going to be exposed to consumer apps? Future insurance data rail Written query to NHA integration support; monitor NHCX documentation [S59] Insurance architecture
Q9 Are risk flags/interaction alerts SaMD under CDSCO? Scope and cost Legal/regulatory opinion; review CDSCO software guidance ⚠ Validate Phase 2 clinical features
Q10 Do claim-assistance features require IRDAI authorisation? Scope Insurance counsel opinion Claim assistance roadmap
Q11 Do users understand trust tiers and consent screens? Trust, legal validity Comprehension tests (teach-back) in 2–3 languages with elderly and low-literacy participants Consent UX
Q12 Which analytics are trusted and acted upon? Insight design A/B explainability variants; measure care-gap closure and doctor-share Insight engine roadmap
Q13 What retention is permitted for ABDM-sourced data in a personal vault? Data architecture Review the current HDM Policy + ABDM terms; written clarification from NHA ⚠ Validate Erasure defaults
Q14 Does the company risk SDF designation? Compliance cost Legal assessment against DPDP criteria; scenario planning DPO/audit budget
Q15 Can WhatsApp-based document intake meet privacy expectations? Ingestion channel Security review of BSP data handling; user trust survey Channel strategy

15. Appendix: Source Links, Standards, and Glossary

Appendix A — Source register

Status: V = verified in-session from an official NHA repository · A = author-supplied in Playbook v1.0, not re-opened · K = official source known to the research team; URL/document to be confirmed.

ID Source URL / locator Status
S1 ABDM HIP/HIU Guidelines https://abdm.gov.in/strapicms/uploads/HIP_HIU_Guidelines_f85df336ec.pdf A
S2 ABDM Building Blocks (v8.3 external) https://abdm.gov.in/strapicms/uploads/ABDM_Building_Blocks_v8_3_External_Version_eabbc5c0f3_4_a96f40c649_57c96f3_4de80a9b9c.pdf A
S3 ABDM Guidance Document for ABDM-compliant HMIS/LMIS (FHIR) https://abdm.gov.in/strapicms/uploads/Guidance_Document_for_ABDM_Compliant_HMIS_LMIS_d066e52a6d.pdf A
S4 Building HIU System — API Flow (M3) https://sandboxcms.abdm.gov.in/uploads/Building_HIU_System_API_Flow_M3_Documentation_5327c6f3_4de80a9b9c.pdf A
S5 ABDM consent manager & gateway documentation (community) https://docs.coronasafe.network/abdm-documentation/building-blocks/consent-manager-and-gateway A (secondary)
S6 NRCES FHIR R4 profiles for ABDM https://www.nrces.in/ndhm/fhir/r4/profiles.html A
S7 NHCX package (GitHub) https://github.com/nha-in/nhcx-package A
S8 PIB release on NHCX workflows https://www.pib.gov.in/PressReleseDetail.aspx?PRID=2236366 A
S9 Account Aggregator Framework (DFS) https://financialservices.gov.in/account-aggregator-framework A
S10 RBI Master Direction — NBFC-Account Aggregator https://www.rbi.org.in/Scripts/NotificationUser.aspx?Id=10598 A
S11 Digital Personal Data Protection Act, 2023 https://www.meity.gov.in/static/uploads/2024/06/2bf1f0e9f04e6fb4f8fef35e82c42aa5.pdf A
S12 Digital Personal Data Protection Rules, 2025 https://static.pib.gov.in/WriteReadData/specificdocs/documents/2025/nov/doc20251117695301.pdf A
S13 Unified Health Interface (UHI) https://abdm.gov.in/uhi A
S14 ABDM portal and public dashboard https://abdm.gov.in K
S15 ABDM Sandbox (V3) https://sandbox.abdm.gov.in/sandbox/v3/ (URL as cited in NHA’s NHCX guide [S59]) V (URL)
S16 ABDM Health Data Management Policy (latest version) abdm.gov.in → Resources/Policies K
S17 NRCES FHIR Implementation Guide (ABDM) https://www.nrces.in/ndhm/fhir/r4/index.html K
S18 IRDAI Master Circular on Health Insurance Business (2024) and policyholder-protection circulars https://irdai.gov.in (Circulars) K
S19 IRDAI (Insurance Products) Regulations, 2024 https://irdai.gov.in (Regulations) K
S20 CERT-In Directions under s.70B(6) IT Act, 28 April 2022 https://www.cert-in.org.in K
S21 Telemedicine Practice Guidelines, March 2020 https://www.mohfw.gov.in/pdf/Telemedicine.pdf K
S22 Medical Devices Rules, 2017 and CDSCO notifications on software https://cdsco.gov.in K
S23 Mental Healthcare Act, 2017 https://www.indiacode.nic.in K
S24 Apple HealthKit documentation https://developer.apple.com/documentation/healthkit K
S25 Apple App Review Guidelines (5.1.3 Health and Health Research) https://developer.apple.com/app-store/review/guidelines/ K
S26 Android Health Connect developer guide https://developer.android.com/health-and-fitness/guides/health-connect K
S27 Google Fit APIs (deprecation / migration) https://developers.google.com/fit K
S28 Fitbit Web API reference https://dev.fitbit.com/build/reference/web-api/ K
S29 Garmin Connect Developer Program https://developer.garmin.com/gc-developer-program/ K
S30 Samsung Health developer (Data SDK) https://developer.samsung.com/health K
S31 Oura Cloud API v2 https://cloud.ouraring.com/v2/docs K
S32 Withings developer API https://developer.withings.com K
S33 Google OAuth restricted-scope verification (CASA) https://developers.google.com/identity/protocols/oauth2/production-readiness/restricted-scope-verification K
S34 ICMR-National Institute of Nutrition (IFCT 2017; Dietary Guidelines for Indians 2024; Nutrient Requirements 2020) https://www.nin.res.in K
S35 National List of Essential Medicines 2022 https://cdsco.gov.in / https://main.mohfw.gov.in K
S36 Pradhan Mantri Bhartiya Janaushadhi Pariyojana product list https://janaushadhi.gov.in K
S37 National Programme for Prevention and Control of NCDs (NP-NCD) operational guidelines https://main.mohfw.gov.in K
S38 Universal Immunization Programme schedule https://main.mohfw.gov.in K
S39 Tele-MANAS (National Tele Mental Health Programme) https://telemanas.mohfw.gov.in K
S40 Anjana RM et al. Metabolic NCDs in India (ICMR-INDIAB), Lancet Diabetes & Endocrinology, 2023 doi:10.1016/S2213-8587(23)00119-5 K
S41 National Health Accounts Estimates for India (latest, NHSRC/MoHFW) https://nhsrc.in K
S42 National Family Health Survey (NFHS-5) https://rchiips.org/nfhs/ K
S43 PM-JAY (NHA) and Ayushman Vay Vandana releases https://nha.gov.in/PM-JAY; https://pib.gov.in K
S44 ICMR Ethical Guidelines for Application of AI in Biomedical Research and Healthcare (2023) https://main.icmr.nic.in K
S45 CCPA Guidelines for Prevention and Regulation of Dark Patterns, 2023 https://consumeraffairs.nic.in K
S46 Sahamati (AA industry alliance) and ReBIT AA technical standards https://sahamati.org.in; https://api.rebit.org.in K
S47 IRDAI regulations on insurance intermediaries (web aggregators, corporate agents, brokers) https://irdai.gov.in (Regulations) K
S49 Health Claims Exchange (HCX) protocol specification Swasth/HCX protocol documentation (locate) K
S50 SNOMED CT National Release Centre, India (NRCES) https://www.nrces.in K
S52 Draft Drugs and Cosmetics (Amendment) Rules on sale of drugs by e-pharmacy, 2018 https://cdsco.gov.in / e-Gazette K
S58 Bhashini (National Language Translation Mission, MeitY) https://bhashini.gov.in K
S59 NHA-ABDM/NHCX repository: NHCX Postman collections (sandbox/production onboarding, participant and use-case APIs) and “NHCX – Guide for Providers” integrator deck; last commit 27 Sep 2024 https://github.com/NHA-ABDM/nhcx V
S60 NHA-ABDM/ABDM-wrapper repository: ABDM reference HIP/HIU implementation (V1→V3), consent Permission model, HI-type enum, ECDH/AES-GCM crypto, V3 gateway usage; last commit 21 Feb 2025 https://github.com/NHA-ABDM/ABDM-wrapper V

IDs S48, S51 and S53–S57 are reserved and unused in this edition.

Appendix B — Verification checklist (clear before external use)

# Item Where used Check against
B1 DPDP Rules 2025: notification date, commencement phases, breach-report timing (72h), log retention, notice content, Consent Manager conditions, children/healthcare exemptions §1, §9 S12 (official gazette text)
B2 DPDP Act section numbers and penalty amounts §9.2 S11
B3 ABDM purpose codes (esp. PATRQT), frequency units, current V3 consent and PHR-app APIs §5, §10 ABDM sandbox V3 docs (S15), S4
B4 Current HI-type list (incl. Invoice) and NRCES profile versions §4.1, §5 S6, S17
B5 HDM Policy version; HIU storage/retention rules; ABDM breach-reporting obligations §5, §9 S16
B6 Sandbox exit steps; WASA/CERT-In auditor requirement §5.6 S15, S59, NHA integration support
B7 ABHA/linked-record/HFR/HPR scale figures §2.1 S14 dashboard; PIB
B8 Third-party consent managers in ABDM; DPDP Consent Manager interplay §3.2 S2, S16, S12
B9 NHCX: current onboarding counts; IRDAI direction; ABHA↔︎policy linking access for consumer/PHR apps §3.4 S8, S59, NHA
B10 IRDAI 2024 health insurance norms (cashless timelines, PED/moratorium limits, CIS format, grievance/Ombudsman wording) §2, §7 S18, S19
B11 AA insurance FIP availability §4.2 S9, S46
B12 IRDAI intermediary triggers for recommendations/claim assistance §9.6 S47 + counsel
B13 CDSCO position on clinical decision support software §9.1 S22 + counsel
B14 Health Connect permission rules (history >30 days, background read), Play health-apps policy §4.3, §7.3 S26
B15 Google Fit shutdown date; Fitbit API migration; Oura token policy; Samsung SDK partner terms §4.3 S27, S28, S31, S30
B16 Apple Guideline 5.1.3 wording; Health Records availability in India §4.3 S24, S25
B17 IFCT/Indian food DB licensing §4.4 S34
B18 E-pharmacy rules status §4.5 S52
B19 SNOMED CT India licence terms and drug extension §4.5 S50
B20 NP-NCD screening protocol, UIP schedule §8 S37, S38
B21 Helpline numbers (112, 108, Tele-MANAS 14416) per state §4.7, §9.4 S39, state sources
B22 OOP share, NCD prevalence, PM-JAY figures, NFHS-5 insurance coverage §1, §2 S40–S43
B23 Competitor ABDM integration status and features §2.3 ABDM integrator list; company sources
B24 Bhashini API terms §7.1 S58
B25 LOINC/SNOMED codes in examples and rule library §8.3, App. D LOINC, SNOMED CT browsers

Appendix C — Standards reference

Standard Use in product
HL7 FHIR R4 + NRCES ABDM profiles Exchange, canonical clinical/insurance resources, export
NHCX FHIR profiles (Claim, ClaimResponse, CoverageEligibility, InsurancePlan…) Normalisation target for policy/claim documents; future NHCX participation
SNOMED CT (India NRC) Conditions, procedures, findings, drugs (India extension)
LOINC Lab tests and vital-sign observations
ICD-10 Diagnoses on claims and discharge summaries
UCUM Units of measure
ATC Drug classification (molecule level)
ECDH (Curve25519) + HKDF + AES-GCM ABDM health-information transfer encryption [S60]
JWE (RFC 7516) NHCX payload encryption [S59]
UHI / Beckn Discovery and booking (later phases)
ReBIT AA specifications If integrating Account Aggregator data (insurance FIPs)
ISO/IEC 27001 / 27701 Information security and privacy management
OWASP ASVS / MASVS Application and mobile security verification

Appendix D — FHIR R4 examples (corrected)

Illustrative only. Validate against current NRCES profiles; ABDM bundles wrap clinical content in Composition-based document bundles (e.g. DiagnosticReportRecord, PrescriptionRecord, DischargeSummaryRecord) [S6].

D.1 Patient (ABHA identifier)

{
  "resourceType": "Patient",
  "id": "patient-001",
  "identifier": [{
    "type": {"text": "ABHA address"},
    "system": "https://healthid.abdm.gov.in",
    "value": "ravi.kumar@abdm"
  }],
  "name": [{"text": "Ravi Kumar"}],
  "gender": "male",
  "birthDate": "1985-06-12"
}

The identifier system URI is illustrative. Use the URI prescribed by the current NRCES Patient profile.

D.2 Observation: HbA1c (unchanged from v1.0 except reference-range structure)

{
  "resourceType": "Observation",
  "id": "observation-hba1c-001",
  "status": "final",
  "category": [{"coding": [{"system": "http://terminology.hl7.org/CodeSystem/observation-category", "code": "laboratory"}]}],
  "code": {"coding": [{"system": "http://loinc.org", "code": "4548-4",
            "display": "Hemoglobin A1c/Hemoglobin.total in Blood"}], "text": "HbA1c"},
  "subject": {"reference": "urn:uuid:patient-001"},
  "effectiveDateTime": "2026-09-10",
  "valueQuantity": {"value": 7.1, "unit": "%", "system": "http://unitsofmeasure.org", "code": "%"},
  "referenceRange": [{"low": {"value": 4.0, "unit": "%"}, "high": {"value": 5.6, "unit": "%"},
                      "text": "4.0 - 5.6% (laboratory-specific)"}]
}

D.3 DiagnosticReport: laboratory (corrected: report code matches its content)

{
  "resourceType": "DiagnosticReport",
  "id": "diagnostic-report-lab-001",
  "status": "final",
  "category": [{"coding": [{"system": "http://terminology.hl7.org/CodeSystem/v2-0074", "code": "LAB", "display": "Laboratory"}]}],
  "code": {"coding": [{"system": "http://loinc.org", "code": "4548-4",
            "display": "Hemoglobin A1c/Hemoglobin.total in Blood"}], "text": "HbA1c report"},
  "subject": {"reference": "urn:uuid:patient-001"},
  "effectiveDateTime": "2026-09-10",
  "issued": "2026-09-10T18:00:00+05:30",
  "performer": [{"reference": "urn:uuid:organization-lab"}],
  "result": [{"reference": "urn:uuid:observation-hba1c-001"}],
  "presentedForm": [{"contentType": "application/pdf", "title": "Laboratory Report",
                     "url": "urn:uuid:binary-lab-report-001"}]
}

v1.0 used LOINC 24323-8 (comprehensive metabolic panel) as the report code while the only result was HbA1c. Use a panel code only when the report contains that panel’s results.

D.4 DocumentReference: discharge summary (corrected LOINC code)

{
  "resourceType": "DocumentReference",
  "id": "document-discharge-summary-001",
  "status": "current",
  "type": {"coding": [{"system": "http://loinc.org", "code": "18842-5", "display": "Discharge summary"}],
           "text": "Discharge Summary"},
  "subject": {"reference": "urn:uuid:patient-001"},
  "context": {"encounter": [{"reference": "urn:uuid:encounter-ip-001"}]},
  "content": [{"attachment": {"contentType": "application/pdf", "title": "Discharge Summary",
                              "url": "urn:uuid:binary-discharge-001"}}]
}

v1.0 used 34117-2 (“History and physical note”). The NRCES DischargeSummaryRecord is a Composition-based document. A SNOMED CT type such as 373942005 (Discharge summary) may be used there; validate against the profile.

D.5 MedicationRequest. The v1.0 structure is retained. Recommendations: code medications with SNOMED CT (as used in NRCES profiles) plus ATC where useful; keep the brand name in medicationCodeableConcept.text; never store dose inferred from OCR without user confirmation.

D.6 Claim (FHIR). The v1.0 example is retained as an internal normalisation target. Note that NHCX claim bundles follow NRCES/NHCX-specific profiles and are exchanged only between participants (PAYER/PROVIDER/TPA) using JWE encryption [S59].

D.7 Internal Coverage-clause object (non-FHIR helper)

{
  "policy_id": "pol-001",
  "benefit": "day_care_procedure",
  "item": "Cataract surgery",
  "limit": {"amount": 40000, "currency": "INR", "per": "eye"},
  "waiting_period": {"type": "specific_disease", "months": 24,
                     "continuous_cover_start": "2025-03-15", "ends_on": "2027-03-15"},
  "clause_ref": {"document_id": "doc-policy-wording-001", "page": 14, "span": [1203, 1388],
                 "text_hash": "sha256:…"},
  "extraction": {"confidence": 0.96, "verification": "user_confirmed"},
  "trust_tier": "T2"
}

Appendix E — Glossary

Term Meaning
AA Account Aggregator (RBI-regulated NBFC-AA; data-blind consent conduit for financial data)
ABDM Ayushman Bharat Digital Mission (National Health Authority)
ABHA Ayushman Bharat Health Account (14-digit number; ABHA address as a handle)
Care context Provider-side grouping of health records (e.g. a visit) linked to an ABHA address
CIS Customer Information Sheet: IRDAI-mandated summary of a health policy’s key terms
CM / HIE-CM Health Information Exchange–Consent Manager
Consent artifact Machine-readable, signed record of a patient’s consent in ABDM
Data Fiduciary / Data Principal DPDP terms for the entity determining purpose and means of processing / the individual
DPB Data Protection Board of India
DPDP Digital Personal Data Protection Act, 2023 (and Rules, 2025)
FHIR HL7 Fast Healthcare Interoperability Resources
FIP / FIU Financial Information Provider / User (AA framework)
HFR / HPR Health Facility Registry / Healthcare Professionals Registry
HIP / HIU Health Information Provider / User
HI type ABDM health-information type (OPConsultation, Prescription, DischargeSummary, DiagnosticReport, ImmunizationRecord, HealthDocumentRecord, WellnessRecord)
M1 / M2 / M3 ABDM integration milestones: ABHA; HIP linking and sharing; HIU consent and import
NHCX National Health Claims Exchange (provider↔︎payer/TPA claims rail)
NRCES National Resource Centre for EHR Standards (FHIR IG; SNOMED CT NRC for India)
OOP Out-of-pocket expenditure
PED Pre-existing disease (insurance)
PHR Personal Health Record (app)
RMP Registered Medical Practitioner
SaMD Software as a Medical Device
SDF Significant Data Fiduciary (DPDP)
TPA Third-Party Administrator (health insurance claims)
Trust tier This report’s labelling of data provenance (T1 verified clinical … T6 reference knowledge)
UHI Unified Health Interface (open protocol for discovery, booking, fulfilment)
WASA Web Application Security Assessment (ABDM security testing)