Skip to main content
Expert Talk
Amakomaya — Aamako Maya Apps Pvt. Ltd.

Maternal and Child Health

Maternal and child health is the domain where the architectural properties discussed elsewhere stop being abstract. A pregnancy spans nine months, several facilities, a community health worker and — at delivery — often an unplanned provider. If longitudinal identity, scheduling and referral do not work, women are lost to follow-up, and the loss is measurable in outcomes.

It is also the domain with the most mature computable guideline content: WHO's antenatal care work is among the most developed SMART Guidelines material available.

The pregnancy as an episode​

The concept that most systems fail to model, and that everything else depends on.

A pregnancy is not a series of unrelated encounters. It is an episode of care with a start, an expected end, a schedule, a risk status that evolves, and an outcome. Systems that model only encounters cannot answer "how many contacts has this woman had?", "is she overdue?", or "what was her blood pressure trend?" without fragile reconstruction.

   Pregnancy episode
        │
        ├── LMP / estimated date of delivery / gestational age
        ├── Risk classification (evolving)
        ├── ANC contacts        1 … 8+ (per national schedule)
        ├── Referrals           (with outcomes)
        ├── Delivery            place, mode, attendant, complications
        ├── Outcome             live birth(s), stillbirth, loss
        └── Postnatal contacts  mother and newborn, on separate schedules
                │
                └──▶ Child record(s), linked to the mother

In FHIR, EpisodeOfCare carries this, with Encounter referencing it. Many implementations instead use a custom pregnancy resource or a DHIS2 tracker programme. Any of these works; not modelling it at all does not.

Gestational age is a computed, time-varying value​

Getting this right is disproportionately important because almost every clinical decision in the domain depends on it.

  • LMP (last menstrual period) is the usual basis, and is frequently uncertain or unknown
  • Ultrasound dating is more accurate and often unavailable
  • Symphysis-fundal height is an estimate used where nothing else exists
  • EDD (estimated date of delivery) may be revised as better information arrives

Architectural requirements:

  1. Store the basis, not just the number. Record LMP, ultrasound date and scan-derived gestational age separately, each with its own date and method.
  2. Compute gestational age at the point of use. Storing "24 weeks" without a reference date produces a value that is wrong the following week. This is a real and common defect.
  3. Support revision with history. When the EDD is revised, the previous value and the reason must remain visible — a schedule computed on the old EDD needs to be recomputed, and someone needs to understand why appointments moved.
  4. Handle "unknown". A woman presenting at an unknown gestation is common, and a system that requires an LMP will get a fabricated one.

Scheduling and recall​

The contact schedule comes from national policy, derived from WHO recommendations (the 2016 WHO ANC model recommends eight contacts; national schedules vary and the national one governs).

Scheduling logic is guideline content, and belongs in the DAK rather than hard-coded in an application — because it changes, and when it changes it must change everywhere at once.

The recall workflow drives real architecture:

   Scheduled contact due
        │
        ├──▶ SMS / voice / push reminder to the client
        ├──▶ Task in the CHW's worklist
        └──▶ Facility due-list
                │
        (contact not attended)
                │
        ├──▶ Escalating reminders
        └──▶ CHW home visit task
                │
        (still not attended)
                │
        └──▶ Defaulter list for supervisor review

This requires a notification service, a task/worklist model in the CHW application, and a defaulter-tracking capability. See community health.

Danger signs and referral​

Danger sign detection is the highest-value decision support in the domain, and it is also the most demanding of the architecture, because it must work where the woman is — which is frequently offline.

Design implication: danger sign logic must run locally on the device, not in a cloud decision service. A CHW in a village with no signal cannot wait for a CDS Hooks response. The logic is distributed with the application, versioned, and updated when connectivity permits. See offline-first.

The referral loop is the part that most often fails:

  1. CHW or facility identifies a danger sign
  2. Referral is created, with clinical context, and given an identifier
  3. The receiving facility is notified
  4. The woman travels — the step the system cannot control
  5. Arrival is recorded at the receiving facility, against the referral
  6. Outcome is recorded and fed back to the referring worker
  7. Non-arrivals appear on a follow-up list

Steps 5 to 7 are what makes it a loop rather than a one-way message. Most implementations build 1 to 3 and stop, and consequently cannot tell whether referral works. The feedback is also what sustains CHW engagement — a worker who never learns what happened to their referrals stops making them carefully.

At delivery, one record becomes two or more, and the linkage must be established at a moment when the system is least likely to be attended to.

Requirements:

  • A newborn identity created immediately, before any national identifier exists — a temporary identifier with a defined path to permanence. See client registry.
  • A persistent link between mother and child records (FHIR RelatedPerson, or Patient.link semantics as appropriate), because postnatal care is delivered to both.
  • Multiple births handled correctly — twins share a mother, a delivery, a date of birth and often a name pattern, and are the classic false-positive case for patient matching.
  • Adverse outcomes modelled properly: stillbirth and neonatal death must be recordable without creating a spurious ongoing child record, and without the system continuing to send immunisation reminders. This is a dignity issue as much as a data one, and it is routinely handled badly.

Child health continuity​

The child record then carries:

  • Immunisation — doses, schedule, due dates, catch-up logic. The strongest case for a shared immunisation registry, because children are vaccinated wherever they happen to be.
  • Growth monitoring — weight, height, MUAC over time, plotted against standards. Requires reliable longitudinal identity and accurate age.
  • Nutrition and feeding practice
  • Sick child care — IMCI-style assessment and classification

Each of these needs the same episode/longitudinal properties as the pregnancy.

Data elements that need agreed definitions​

The recurring source of non-comparable data between districts:

ElementThe question that must be answered
ANC contactDoes a blood-pressure check at an outreach post count?
ANC 1First contact this pregnancy, or first contact in this facility?
Skilled birth attendanceWhich cadres count as skilled, per national policy?
Live birthBoundary against stillbirth, and gestational-age threshold
Postnatal contactWithin how many days, for mother and for newborn separately?
Facility deliveryThe facility of delivery, or of admission?

These belong in the national data dictionary, derived from the WHO DAK, and bound to value sets in the terminology service.

Standards and content​

  • WHO SMART Guidelines — antenatal care DAK and related computable artefacts. Check the current catalogue and each item's publication status: https://www.who.int/teams/digital-health-and-innovation/smart-guidelines
  • WHO ANC recommendations (2016) — the clinical basis: https://www.who.int/publications/i/item/9789241549912
  • FHIR — EpisodeOfCare, Encounter, Observation, Condition, Immunization, ServiceRequest (referral), RelatedPerson, Patient.link
  • Terminology — LOINC for measurements, SNOMED CT for findings and danger signs, ICD for outcome coding
  • Immunisation FHIR implementation guides — WHO SMART Guidelines immunization content, and various national IGs. Verify status per guide.

Architecture checklist​

  • Pregnancy modelled as an episode, not as loose encounters
  • LMP, EDD and dating method stored separately, with revision history
  • Gestational age computed at point of use, never stored as a bare number
  • Contact schedule sourced from national policy, not hard-coded
  • Danger sign logic executes offline on the device
  • Referral loop closes, with arrival and outcome recorded and fed back
  • Newborn identity issuable before a national identifier exists
  • Mother–child link persistent and queryable
  • Multiple births and adverse outcomes handled explicitly
  • Indicator definitions agreed and shared with the reporting layer
  • Immunisation record accessible wherever the child presents

References​

Last updated on