Care gap identification data quality: how trust scoring improves population health AI
Population health AI models flag care gaps using data that is often incomplete, outdated, or coded inconsistently across systems. Trust scoring each record before it enters a care gap identification engine reduces false positives, eliminates phantom gaps, and produces outreach lists that clinicians actually trust. Without a trust layer, care gap data quality degrades every downstream decision from risk stratification to quality measure reporting.
The care gap identification problem is a data quality problem
A care gap is a clinical service that evidence says a patient should receive but has not. Colorectal cancer screening after age 45. HbA1c testing every six months for a patient with diabetes. An annual eye exam for someone on a GLP-1 receptor agonist. Population health AI models scan claims, EHR data, lab results, and pharmacy records to find these gaps at scale.
The problem is not the algorithm. The problem is what the algorithm eats.
When a health plan runs a care gap engine against its membership, the output is only as reliable as the input data. And that data is routinely stale, duplicated, miscoded, and fragmented across systems that do not talk to each other. The result: alert fatigue, wasted outreach dollars, and clinicians who stop trusting the gap lists entirely.
What breaks in care gap data
Care gap identification depends on the convergence of multiple data streams. Each stream carries its own quality risks.
Claims data arrives 30 to 90 days after a service is rendered. A patient who completed a mammogram in January may still appear as a gap in March because the claim has not been adjudicated. As we detailed in Claims data lag: what 30-90 day reporting delays cost AI models, this lag creates phantom gaps that trigger unnecessary outreach and erode provider trust in the system.
EHR data is structured inconsistently across vendors. A colonoscopy documented in Epic at one health system and in Cerner at another may use different procedure codes, different date formats, and different clinical note structures. If the population health platform cannot reconcile these, the patient looks like they have an open gap when they do not.
Lab data depends on LOINC code standardization. A fasting glucose result coded with one LOINC identifier at LabCorp and a different one at Quest can cause the same test to appear as two separate entries, or worse, to be missed entirely. The challenges here are significant enough that we wrote an entire post on LOINC code standardization and the lab data trust problem.
Pharmacy claims can confirm medication adherence but miss samples, discount programs, and cash-pay prescriptions. A patient flagged for statin non-adherence may be using a GoodRx coupon that never touches their insurance record.
Each of these failure modes produces false positives (flagging gaps that do not exist) or false negatives (missing gaps that do exist). Both are dangerous. False positives waste resources and annoy patients. False negatives leave real clinical needs unaddressed.
Key statistics
Population health programs lose measurable value when care gap data quality is not assessed before AI models act on it.
Why traditional data quality checks fail for care gaps
Most health plans and ACOs run data quality checks that focus on completeness and format validation. Is the field populated? Is the date in the right format? Is the code a valid ICD-10 code?
These checks are necessary. They are not sufficient.
A record can pass every completeness check and still be wrong. A diagnosis code can be valid but inaccurate. A procedure date can be properly formatted but refer to a service that was billed twice. A lab result can be present but mapped to the wrong LOINC code.
The systematic review literature on data quality in healthcare identifies multiple dimensions: completeness, accuracy, timeliness, consistency, and concordance. But most operational systems only check one or two of these dimensions before feeding data into a care gap engine. The gap between what researchers define as quality and what operational systems actually measure is where care gap identification breaks down.
As we wrote in Data quality vs data trust: what is the difference and why it matters for healthcare AI, quality is a property of a data element. Trust is a property of a data element in context. A lab result that is accurate but 18 months old has quality but no trust for a current care gap assessment.
How trust scoring changes care gap identification
Trust scoring applies a multi-dimensional assessment to every record before it enters a care gap identification engine. The Data Trust Index scores each record 0 to 100 across eight dimensions: Provenance (25%), Consent (20%), Recency (15%), Quality (10%), Concordance (10%), Validation (10%), Breadth (5%), and Stability (5%).
For care gap identification specifically, three dimensions matter most.
Recency determines whether a data point reflects the patient's current clinical state. A mammogram completed 14 months ago is clinically relevant. One completed 38 months ago is not. Claims data that is 60 days old carries a different trust weight than data confirmed through an EHR feed yesterday. Recency scoring lets the care gap engine weight recent confirmations higher and flag stale data for verification rather than treating it as ground truth.
Concordance measures whether multiple data sources agree. If a claims record says a patient has not had a diabetic eye exam, but an EHR note from their ophthalmologist documents a dilated fundus exam three months ago, concordance scoring catches the conflict. Without concordance checks, the care gap engine treats the claims record as authoritative and generates a false alert.
Provenance tracks where the data came from and how it was transformed. A lab result that originated from a CLIA-certified lab, was transmitted via a standard HL7 interface, and mapped to a validated LOINC code has higher provenance than a manually entered result in a free-text clinical note. Provenance scoring tells the care gap engine how much to trust each data point's origin.
When every record carries a trust score, the care gap engine can make smarter decisions. High-trust records confirm or close gaps definitively. Low-trust records trigger verification workflows instead of outreach. Medium-trust records get flagged for reconciliation.
The population health AI trust cascade
Care gap identification is not a standalone function. It feeds into a cascade of downstream decisions that affect patient outcomes, quality scores, and financial performance.
Risk stratification uses care gap density as one input. A patient with five open gaps looks higher risk than one with zero. But if three of those five gaps are phantoms created by claims lag, the risk score is inflated, and care management resources get misallocated.
Quality measure reporting for HEDIS, CMS Stars, and value-based contracts depends on accurate care gap closure rates. A health plan that cannot distinguish real gaps from phantom gaps will underreport its closure rate, costing millions in quality bonus payments. Or it will over-close gaps by counting unverified data as evidence of service completion, creating compliance risk.
Patient outreach driven by unscored care gap data wastes money and trust. A patient who receives a reminder to schedule a mammogram they already completed does not just ignore the message. They lose confidence in their health plan's competence. Multiply this across thousands of members and the brand damage is measurable.
Provider performance measurement becomes unreliable when care gaps are miscounted. A physician whose panel shows a 65% gap closure rate when the real number is 82% will disengage from the quality program entirely. Provider engagement in value-based care depends on data they can trust.
Trust scoring at the data layer prevents errors from cascading through all of these functions. This is why we describe the DTI as a FICO score for health data: it gives every downstream consumer of that data a standardized signal about how much to rely on it.
Clinical relevance: what clinicians see when trust scoring is absent
Clinicians are already skeptical of care gap lists. A 2024 qualitative study published in MedPage Today found that patients with hidradenitis suppurativa avoided biologics partly because of information quality concerns. The same dynamic plays out on the provider side: when physicians receive care gap alerts that repeatedly conflict with what they see in the chart, they stop paying attention to all alerts.
Recent coverage of genomic classifier testing in prostate cancer illustrates a related problem. Genomic classifier tests do not appear to consistently influence risk classification or treatment decisions, suggesting that even when data is available, its integration into clinical workflows fails when providers do not trust the underlying information. The same trust breakdown happens with care gap alerts built on low-quality data.
The fix is not better algorithms. The fix is better data going into the algorithms. Trust scoring makes care gap alerts credible by ensuring that only high-confidence data drives outreach.
The care gap data quality dimensions most systems ignore
Beyond recency, concordance, and provenance, several other trust dimensions affect care gap identification in ways most population health platforms do not address.
Consent matters because care gap identification often requires combining data across entities. A health plan combining claims data with third-party lab results needs documented consent for that combination. Without consent governance, the care gap engine may be operating on data it is not authorized to use for that purpose. SuperTruth's ConsentOS addresses this with five-tier consent architecture.
Validation checks whether the data has been verified against an authoritative source. A self-reported blood pressure reading from a patient portal has a different validation level than one recorded by a nurse during an office visit. Care gap engines that treat both identically will produce unreliable outputs.
Stability measures how much a data element changes over time. A patient's diabetes diagnosis is stable. Their medication list is not. Care gap engines need to weight stable attributes differently from volatile ones when determining gap status.
Most population health platforms check none of these dimensions. They ingest data, run rules, and produce gap lists. The gap lists look authoritative because they come from a system, but the system never asked whether the data was trustworthy enough to act on.
What a trust-scored care gap workflow looks like
A trust-scored care gap identification workflow operates differently from a traditional one.
Step 1: Data ingestion with trust scoring. Every record entering the system receives a DTI score at the point of ingestion. Claims, EHR extracts, lab feeds, pharmacy data, and SDOH data all get scored across all eight dimensions.
Step 2: Trust-tiered gap identification. The care gap engine runs its clinical rules but tags each gap with the trust level of the underlying evidence. A gap identified from Platinum-grade data (DTI 90+) gets treated as confirmed. A gap identified from Bronze-grade data (DTI 50-69) gets routed to a verification queue.
Step 3: Concordance resolution. When multiple data sources conflict on gap status, the system uses concordance scoring to determine which source to trust. A recent EHR-confirmed procedure with high provenance overrides a stale claims absence.
Step 4: Outreach prioritization. Only gaps supported by data above a defined trust threshold trigger patient or provider outreach. Gaps below the threshold trigger data reconciliation workflows instead.
Step 5: Feedback loop. When outreach confirms or denies a gap, the result feeds back into the trust scoring model, improving future accuracy.
This workflow eliminates phantom gaps before they reach a clinician or patient. It reduces outreach waste. And it builds the feedback loops that make the system smarter over time.
The financial case for trust-scored care gaps
The financial impact of care gap data quality is direct and measurable.
A mid-size health plan with 500,000 members might generate 2 million care gap alerts per year. If 30% are phantom gaps, that is 600,000 false alerts. At a conservative cost of $5 per outreach attempt (including call center time, mailing costs, and staff overhead), phantom gaps cost $3 million annually in wasted outreach alone.
The quality score impact is larger. CMS Stars ratings drive billions in quality bonus payments. A plan that underreports its care gap closure rate by even 0.5 stars due to data quality issues can lose tens of millions in annual revenue. Trust scoring the data before it enters the quality measurement pipeline directly protects that revenue.
SuperTruth demonstrated this ROI with imaware, where trust scoring 105,000 diagnostic records reduced standardization time from three weeks to two hours and saved 200+ hours per month. The identified data patterns also revealed a customer segment driving 20% of revenue that had been invisible in the unscored data. As imaware CEO Brodie Flanders said: "The lab industry has never had a trust standard. DTI created one."
Why population health AI trust requires a trust layer, not better algorithms
The current approach to improving care gap identification focuses on algorithmic sophistication. Machine learning models, natural language processing of clinical notes, and predictive analytics all aim to find gaps more accurately.
But no algorithm can compensate for data it cannot trust. A neural network trained on claims data with a 30-day lag will learn to predict gaps that do not exist. An NLP model parsing unstructured notes will extract information that was never validated. A predictive model trained on ICD-10 codes with a 10-30% error rate will inherit that error rate in its predictions.
The answer is not algorithmic. The answer is infrastructural. Every data element needs a trust score before any model processes it. This is what the DTI Engine does: it sits between the data sources and the AI models, scoring every record so the models operate on data whose reliability is known and quantified.
This is the difference between population health AI that generates reports and population health AI that generates action. Clinicians act on data they trust. Trust is not a feeling. It is a score.
The DTI Engine scores every health data record 0 to 100 across 8 trust dimensions before your AI model sees it. If your team is building or buying population health AI and needs care gap identification intelligence that clinicians will actually trust, contact Louis Simeonidis at louis@supertruth.ai or (215) 918-4140.
Further reading:
Jason Alan Snyder
Co-founder of SuperTruth and Artists & Robots, and an inventor on the Data Trust Index patents. Twenty-plus years building technology inside Interpublic Group. He writes here nearly every day on data trust, provenance, and what AI should be allowed to act on, and publishes essays on his Substack.
About SuperTruth · LinkedIn · Substack · jasonalansnyder.com
See it in practice
The FICO score for health data.
8 dimensions. 0–100. Travels with every record permanently.