Prevention Adoption Initiative · Data-infrastructure pillar
The colorectal-cancer screening cascade, completion & follow-up
Screening only counts when the whole cascade closes: an overdue patient completes a stool test, and, if it is positive, completes the follow-up colonoscopy that turns a warning into a diagnosis. Nationally, that second step is where people fall through: between 24% and 75% of positive stool tests get a follow-up colonoscopy, and in Italian screening-program data people who never completed one were about twice as likely to die of colorectal cancer.1,2 This is the registry the initiative would install to close both gaps, with an explainable risk model, a next-best action, and a time-to-colonoscopy clock. Work the reds first and watch the numbers move.
Synthetic single-site view · 30 patients shown of 247 screened · last sync: static demo · no live EHR
If you have one minute, do these three things
- Read the top row of the Urgent queue. One patient is 194 days past a positive stool test — already outside the 180-day window of the HEDIS measure NCQA has proposed. Most EHRs have no screen that shows you this.
- Press “Complete colonoscopy” on that patient. Watch the follow-up rate, the median days-to-colonoscopy and the equity gap all move together. That is the loop a site would work each week.
- Open “Equity, follow-up by language.” The gap between English and non-English follow-up is the number a QI committee is graded on, and the number outreach automation most often makes worse.
Why the follow-up step is the one that kills
24-75%
of positive stool tests actually get a follow-up colonoscopy, a gap so wide that NCQA has proposed a dedicated HEDIS measure for it (COF-E, MY2027: colonoscopy within 180 days).1
~2×
higher colorectal-cancer mortality among people with a positive test who never completed a follow-up colonoscopy, hazard ratio 2.03.2 An association, not a measured causal effect.
12 months
is the danger line: beyond it, the odds of any colorectal cancer are 2.25× and of advanced-stage disease 3.22× those of a colonoscopy done within a month.3 Sooner is better.
Urgent: positive FIT, awaiting colonoscopy
The cohort that decides whether screening saves a life. Sorted by days since the positive result, a live time-to-colonoscopy clock, banded against the 90-day mark CDC’s Colorectal Cancer Control Program uses as its quality indicator and the 180-day COF-E follow-up window.5 This queue does not exist in most EHRs today.
≤ 90d · on target> 90d · past best practice> 180d · fails COF-ECOF-E: the follow-up-after-positive measure NCQA has proposed for HEDIS MY2027 — colonoscopy within 180 days, ages 45–85.
Screening queue, overdue for a stool test
Ranked by an explainable non-completion risk = care gap + prior no-shows + SDOH barriers + language mismatch. Each score shows its contributions; the next-best action is matched to the barrier.
The screening cascade
Where the panel is right now, and where it leaks.
Equity, follow-up by language
Colonoscopy-after-positive-FIT completion, watched as a fairness guardrail so automation never widens the gap.
Community health workers, the human layer for the hardest cases
When two low-touch attempts (reminder, mailed FIT) don’t close the gap (or a positive FIT stalls behind transport, language, or trust) the case escalates to a CHW. Higher cost per touch, far higher yield; the intervention with the strongest evidence for non-responders. Escalate one from the screening queue and it appears here.
Why this is state-of-the-art, and honest
Design claims the initiative would build toward. Everything here is illustrative; nothing is deployed.
Cascade, not a checkbox
Most registries mark screening "done" at the stool test. This one tracks the full cascade to a closed diagnosis, and treats a positive FIT with no colonoscopy as an open, time-critical item, aligned with the proposed NCQA follow-up HEDIS measure (COF-E, MY2027).1
Explainable risk
The score is decomposable: every point traces to a driver (a SHAP-style contribution breakdown). In production the additive score is replaced by a trained model (logistic regression → gradient boosting) with the same explainability, human-in-the-loop, never a black box.
Next-best action
Not just "who is high-risk," but which action most changes completion for this person, mailed FIT for access barriers, centralized navigation for positive-FIT follow-up (the intervention with the strongest evidence), transport support where that is the blocker.4
FHIR-native
Ingest overdue panels and results via Bulk FHIR / SMART-on-FHIR; write actions back as FHIR Communication / Task resources. Phase 1 runs on a flat export; deep write-back is Phase 2.
Fairness monitoring
The follow-up equity gap is a first-class metric, tracked so outreach automation never widens disparities for patients with unstable contact, low portal access, or a language mismatch.
How it would work in production
The demo is static and synthetic. In a real deployment, three things carry it: the AI that ranks and recommends, the pipe that gets your data in, and the security that keeps it compliant.
01 · The AI’s role
Predict, prioritize, explain, never decide
- Predicts non-completion for each overdue patient, and, via an uplift model, which one action most changes their odds, not just who is high-risk.
- Explainable by design: every score decomposes to its drivers (SHAP), shown per patient. Logistic-regression baseline → gradient boosting.
- Human-in-the-loop: the model ranks and suggests; staff and CHWs decide and act. No autonomous clinical decisions.
- Guardrails: fairness-audited (the equity gap is a monitored metric), no protected attributes as direct features, and drift monitoring on live performance.
Signals that would sharpen it: prior colonoscopy history, insurance follow-up cost-sharing, comorbidity burden, transportation access, appointment-slot availability, portal engagement, area deprivation index, and prior FIT-kit return behavior.
02 · Feeding data in
Start with a flat file; grow into FHIR
- Pilot with the least effort: a nightly de-identified export (CSV) of the overdue registry + results. No deep integration needed to begin.
- Standards path: FHIR Bulk Data ($export) for panels and results; a SMART-on-FHIR app for in-EHR use.
- Write-back: recommended actions return as FHIR Communication / Task resources into the EHR worklist.
- Mapping (patient id, screening status, last result, contact prefs, SDOH) is a one-time setup; refresh nightly to start, near-real-time later.
Effort: a single-site flat-file pilot can run in weeks, not quarters. The FHIR and write-back work comes only when the value is proven.
The exact file we would ask for — 19 columns
| patient_id | Site-generated pseudonym. No name, MRN, DOB, address or phone leaves your system. |
| birth_year | Year only — enough to apply the 45–85 eligibility window. |
| sex, preferred_language | Reporting and outreach-channel matching. |
| zip3 | First three digits only, per the HIPAA Safe Harbor rule. |
| payer_type | Coded category, not a policy number. |
| screening_status | overdue | fit_negative | fit_positive | colonoscopy_completed |
| months_overdue | Months since the patient became due. |
| last_fit_date, last_fit_result | Most recent stool test and its result. |
| positive_fit_date | Starts the COF-E clock. The single most important field. |
| colonoscopy_status, colonoscopy_date | open | scheduled | completed — closes the cascade. |
| prior_no_shows_24mo | Count. Drives the non-completion risk score. |
| portal_active, preferred_contact | Which channel actually reaches this person. |
| sdoh_flags | Pipe-separated: transport | language | food | housing | cost. Optional. |
| pcp_id, site_id | Panel attribution for multi-provider or multi-site reporting. |
03 · Build & security (HIPAA)
Compliant by construction
- HIPAA-eligible cloud (AWS / GCP / Azure) under a signed BAA; PHI stays inside the covered entity’s environment or a BAA-covered tenant.
- Encryption in transit (TLS 1.2+) and at rest (AES-256); role-based access + SSO/MFA; full audit logging of every view and action.
- Minimum-necessary access; the model trains on de-identified data; no PHI in analytics, or in this demo.
- Access reviews, breach monitoring, and a documented incident-response plan; aligns with the HIPAA Security Rule (and 42 CFR Part 2 where behavioral data applies).
Rollout: (1) flat-file pilot at one site → (2) FHIR ingest + write-back → (3) multi-site + the AI layer, each gated by an evaluation.
Full panel
All 30 synthetic patients across the cascade. Stage shows where each one is.
| Patient | Lang | ZIP | Payer | Barriers | Risk | Stage |
|---|
Sources
- NCQA. Follow-Up After Positive Colorectal Cancer Non-Invasive Screening Test (COF-E), proposed for HEDIS MY2027: colonoscopy within 180 days of a positive stool-based test, ages 45–85. The published literature puts follow-up completion between 24% and 75%; NCQA’s own field testing found plan-level rates of 21.7% to 37.5% at 180 days.
- Zorzi M, et al. Non-compliance with colonoscopy after a positive faecal immunochemical test, and colorectal cancer mortality. Gut 2022;71(3):561-567. doi:10.1136/gutjnl-2020-322192. Hazard ratio 2.03 (95% CI 1.68-2.44) for colorectal-cancer death among non-compliers. Observational; a published critique (Winters & Ford, Gut 2023;72(9):1799-1800) argues immortal-time bias may inflate the estimate.
- Corley DA, et al. Association between time to colonoscopy after a positive fecal test result and risk of colorectal cancer and cancer stage at diagnosis. JAMA 2017;317(16):1631-1641. 70,124 Kaiser Permanente patients across Northern and Southern California: beyond 12 months, the odds of any colorectal cancer were 2.25 and of advanced-stage disease 3.22, against colonoscopy within one month. The study measures incidence and stage, not mortality. Summary.
- Centralized patient navigation shortens time-to-colonoscopy and raises follow-up completion after a positive FIT (randomized and pragmatic trials in safety-net settings). Community Preventive Services Task Force recommends mailed-FIT outreach and patient navigation.
- Nadel MR, et al. Variations in screening quality in a federal colorectal cancer screening program for the uninsured. Prev Chronic Dis 2019;16:180452. doi:10.5888/pcd16.180452. CDC’s Colorectal Cancer Control Program set a quality indicator of 80% of positive stool tests followed by colonoscopy within 90 days; the 180-day mark reflects the more lenient US Multi-Society Task Force guidance.
A demonstration of the Prevention Adoption Initiative (an HSREP initiative in development). Synthetic data; no real patients, no PHI, no EHR connection. Positive-FIT share is amplified for illustration; a real panel would surface these from hundreds of screened patients. Illustrative only. One figure on this page was corrected on 3 August 2026; the record is in the corrections log.