Case & Cohort Intelligence · Concept Paper

Backward-Generated Case & Cohort Intelligence for Clinical Audits and Research

Instead of asking clinicians to specify what to look for, start from the outcome that already happened — an escalation, a readmission, a flagged claim — and let the pipeline walk backward through the coded timeline and the cohort to generate the questions, the comparisons, and the evidence trail an auditor or researcher would otherwise have to build by hand.

01 Forward audits ask questions. Backward generation finds them.

Most clinical audit and research tooling is forward-directed: someone decides what to check for — a guideline, a hypothesis, a compliance rule — and the system queries records to confirm or deny it. That works well when the reviewer already knows what to suspect. It works poorly when the interesting signal is exactly the thing nobody thought to ask about.

Forward (status quo)
Hypothesis → Query → Records → Confirm/deny

An auditor picks a rule ("was antibiotic X given within 4 hours?") and the system checks compliance one chart at a time. Coverage is limited to rules someone already wrote.

Backward generation
Outcome → Timeline replay → Candidate causes → Cohort test → Question

The system starts from an outcome that already occurred, replays the coded timeline leading into it, and surfaces the comorbidities, delays, and interventions that precede it most often — then asks whether that pattern holds across the cohort.

The mechanism is deliberately unglamorous: no free-text generation stands between the outcome and the finding. Every candidate cause is a SNOMED CT / FHIR-coded item that already sits on the patient's or the cohort's timeline; the pipeline's job is ordering, filtering, and comparing those items against a baseline, not inventing explanations.

Working definition — Backward-generated intelligence takes a realized outcome as the seed, reconstructs the coded events that preceded it (case-level), and repeats that reconstruction across every patient who shares the outcome to see what recurs more than chance would predict (cohort-level). The output is a ranked set of candidate audit questions or research hypotheses, each traceable to an auto-generated query.

02 Architecture: outcome-in, evidence-trail-out

The pipeline sits downstream of a standard FHIR + SNOMED CT store. It doesn't replace forward querying — it adds a second entry point that starts at the end of the timeline instead of the beginning.

1 · SEED OUTCOMEescalation, readmission, mortality flag, denied claim, adverse event report
2 · TIMELINE REPLAY (per case)pull all coded Conditions, MedicationRequests, Procedures, Observations in a bounded lookback window
3a · CASE TRACEorder the window into a reverse chronology; tag comorbidity / symptom / intervention / escalation lanes
3b · COHORT ROLL-UPrepeat the replay for every patient sharing the seed outcome; pool coded items into frequency tables
4 · BASELINE COMPARISONcompute outcome rate in the general population or a matched control window; compute lift for each recurring item
5 · AUDIT / RESEARCH ARTIFACTranked findings, each with its auto-generated SQL/ECL query, ready for reviewer sign-off — not a conclusion, a lead

Because every stage stays inside coded data, the artifact produced at step 5 is auditable in the literal sense: a reviewer can re-run the attached query against the source records and get the same table back.

03 Case-level: replaying one patient backward from the event

At the single-patient level, backward generation looks like a reverse chart review that a human would normally do by scrolling up through the record. The difference is that the pipeline does the scrolling, tags each item by clinical lane, and proposes the question a reviewer would eventually ask anyway.

Worked trace — ICU escalation, day 0

Illustrative, not a real patient
Day 0 · Seed outcome
Escalation to ICU coded on the timeline.
Day −1 to −2
Two coded vital-sign Observations trending toward the escalation threshold; no coded intervention logged in between.
Candidate question: was the trend acted on, or only recorded?
Day −5
MedicationRequest for a antibiotic changed to a second-line agent.
Candidate question: did the switch correlate with a culture result, or precede one?
Day −14
Pre-existing Condition (comorbidity) already on the problem list before this admission.
Candidate question: does this comorbidity recur across other patients who escalated?

What makes this "generated" rather than manual

04 Cohort-level: turning recurring traces into lift

A single backward trace is an anecdote. Running the same replay across every patient who shares the seed outcome turns it into a comparison: does this comorbidity, this delay, this medication pattern show up more often among patients who escalated than among the baseline?

Comorbidity presentPatientsEscalation rate within groupvs. cohort baseline
Chronic kidney disease1861%2.4×
Type 2 diabetes2646%1.8×
COPD1139%1.5×
No flagged comorbidity3425%1.0× (baseline)

This is the same arithmetic an epidemiologist runs by hand for a case-control study — baseline rate, subgroup rate, ratio — just triggered automatically by the seed outcome instead of a pre-registered hypothesis. It doesn't replace a designed study; it tells a reviewer or a researcher where a designed study might be worth running.

Audit framing

Which finding would a quality committee want first?

  • Rank by lift, then by cohort size, so rare-but-huge-lift artifacts don't crowd out common, actionable patterns.
  • Surface the counterfactual too — patients with the same comorbidity who did not escalate — so the finding reads as a lead, not a verdict.
Research framing

Which pattern is worth a pre-registered hypothesis?

  • Treat the ranked list as hypothesis-generation output, explicitly separate from any confirmatory analysis.
  • Attach the exact cohort-selection query so a second team can reproduce the same pool before designing a real study on top of it.

05 Where this fits in audits and research

Use caseSeed outcomeWhat backward generation adds
Retrospective quality reviewAdverse event, code blue, unplanned ICU transferAuto-drafts the "what led up to this" section reviewers currently assemble manually from scrolling charts.
Claims / billing auditDenied or flagged claimTraces the coded justification chain backward to see whether the denial pattern clusters around a specific documentation gap.
Readmission root-cause30-day readmissionCompares discharge-window coded events across readmitted vs. non-readmitted patients with the same index diagnosis.
Hypothesis generation for researchAny outcome of interest in a registryProduces a ranked, reproducible shortlist of candidate exposures before committing study design and funding to one.
Signal detection for pharmacovigilanceReported adverse drug eventSurfaces which pre-event medication changes or comorbidities recur across reports at a rate above baseline.

06 Build order

Phase 1

Single-case backward trace

  • Fixed lookback window per outcome type; lane-tag coded items (comorbidity, symptom, intervention, escalation).
  • Render the trace with the query attached to every step — no narrative generation yet.
Phase 2

Cohort roll-up and baseline comparison

  • Repeat Phase 1 across all patients sharing the seed outcome; pool into frequency tables.
  • Compute baseline outcome rate and per-item lift; require a minimum group size before surfacing a lift figure.
Phase 3

Templated audit / research artifacts

  • Turn ranked findings into a standard brief format (presentation → intervention → tracking) built entirely from computed fields.
  • Attach a reproducibility footer: exact query, cohort size, and whether the result came from cache or a fresh run.
Phase 4

Reviewer sign-off loop

  • Every generated question routes to a human reviewer before it becomes an audit finding or a study hypothesis.
  • Log accept/reject decisions to see which finding types are actually useful, and prune the templates that never survive review.

07 What this deliberately does not claim

Correlation, not causation. Lift computed against a baseline is a screening signal. It tells a reviewer where to look, not what caused what — confounding by indication is expected and undisclosed confounders are not ruled out.
Not a predictive model. The pipeline surfaces observed historical rates, not a validated risk score. Treat any outcome-prediction framing as illustrative until a real model is built, calibrated, and reviewed on held-out data.
Coded data only. If a contributing factor was never coded — a hallway conversation, an unlogged clinical judgment call — it cannot appear in a backward trace. The absence of a finding is not evidence of absence.
Human-in-the-loop by design. Every artifact is a candidate question for a reviewer, not a finding a committee should act on unread. The templates exist to standardize how a lead is presented, not to replace the judgment of whoever decides what to do with it.