Suspect Conditions Risk Adjustment: Identification, Validation, and Capture

What Are Suspect Conditions

Suspect conditions in risk adjustment are diagnoses that clinical evidence — drawn from prior claims history, lab results, medication records, or prior-year coding — indicates a member likely has but that have not yet been documented through a qualifying face-to-face encounter in the current payment year, representing an HCC gap between probable clinical status and what the RAF score currently reflects.

Suspect conditions risk adjustment workflows bridge the gap between claims data and clinical reality. Suspect conditions are diagnoses that clinical evidence suggests a member has but that have not been documented and coded through a qualifying face-to-face encounter in the current payment year. They represent the gap between a member's actual clinical status and what the risk adjustment data reflects.

In the CMS-HCC model, every Hierarchical Condition Category must be re-documented annually. A member diagnosed with diabetes, heart failure, and COPD in 2025 starts 2026 with none of those HCCs unless each condition is evaluated, assessed, and coded in a new qualifying encounter. Chronic conditions do not carry over automatically, which means every chronic condition in the population is a potential suspect at the start of each payment year.

• Prior-Year Gaps: HCCs captured in the prior payment year that have not yet been recaptured in the current year. These are the highest-confidence suspects because the condition was previously documented and coded. For chronic conditions, the clinical likelihood that the condition still exists exceeds 95%
• Pharmacy-Based Suspects: Conditions inferred from active medication fills without a corresponding diagnosis code in the current year. A member filling metformin and insulin has a near-certain diabetes diagnosis, but if no provider encounter has documented it this year, the HCC is missing from the RAF score
• Lab-Based Suspects: Conditions suggested by laboratory results without corresponding diagnosis codes. An HbA1c of 8.5% without a diabetes diagnosis, or an eGFR of 28 without a CKD stage 4 code, represents a clinical reality not reflected in the coded data
• Clinical Algorithm Suspects: Conditions predicted by statistical models analyzing patterns across multiple data sources. A member with a history of stroke, current anticoagulant use, and recent echocardiogram orders likely has atrial fibrillation, even if no AF diagnosis appears in current-year claims

How Suspect Conditions Are Identified

Suspect identification requires integrating multiple data sources and applying clinical logic to detect the gap between probable conditions and documented conditions. The methods range from straightforward year-over-year comparison to sophisticated predictive modeling.

• Year-Over-Year HCC Comparison: The most basic and highest-confidence method. Compare each member's prior-year HCCs against current-year captured HCCs. Any prior-year HCC not yet recaptured becomes a suspect. This approach identifies the recapture gap that HCC gap analysis quantifies at the population level
• Pharmacy Crosswalk Analysis: Map active prescription fills to expected diagnosis codes using drug-to-disease crosswalk tables. Members with active prescriptions for condition-specific medications (insulin for diabetes, ACE inhibitors for heart failure, bronchodilators for COPD) who lack the corresponding diagnosis code in current-year claims are pharmacy-based suspects
• Lab Value Inference: When lab results are available through direct feeds or health information exchange, abnormal values directly suggest conditions: elevated creatinine suggests kidney disease, elevated TSH suggests hypothyroidism, and abnormal lipid panels suggest metabolic conditions. Lab-based suspects have lower confidence than pharmacy suspects because abnormal values may reflect acute rather than chronic conditions
• Procedure-Based Inference: Certain procedures strongly imply underlying conditions. A member receiving dialysis has ESRD. A member with a below-knee amputation history likely has peripheral vascular disease or diabetes with complications. Procedure data corroborates other suspect signals
• Predictive Modeling: Machine learning models trained on historical data predict which members are likely to have undocumented conditions based on patterns across demographics, utilization, pharmacy, and claims. These models identify suspects that simpler rule-based methods miss, particularly for conditions with indirect clinical indicators
• Clinical Clustering: Conditions that commonly co-occur create inference opportunities. A member with documented diabetes, obesity, and hypertension but no hyperlipidemia diagnosis has a high probability of having undiagnosed or undocumented dyslipidemia based on known comorbidity patterns

Understanding the relationship between prospective and retrospective risk adjustment approaches is essential for designing suspect identification programs that operate on the right time horizon.

Validation Workflows

Not every identified suspect will be validated as a current, active condition. Validation workflows ensure that only clinically legitimate conditions proceed to provider engagement, protecting both compliance standing and provider trust.

• Clinical Plausibility Review: Before routing suspects to providers, apply clinical plausibility rules. A diabetes suspect for a member with no history of diabetes, no diabetic medications, and normal HbA1c values should not reach a provider. False suspects waste provider time and erode program credibility. Target a false positive rate below 15%
• Confidence Scoring: Assign confidence levels to each suspect based on the strength and number of supporting data signals. A suspect supported by prior-year HCC, active medication, and abnormal lab values has very high confidence. A suspect based solely on a clinical clustering algorithm has lower confidence. Confidence scores drive prioritization and routing
• Duplicate and Overlap Detection: Check suspects against encounters already scheduled or recently completed. A member with an endocrinology appointment next week does not need a separate outreach for their diabetes suspect; instead, the suspect should be attached to the upcoming visit for the provider to address
• Compliance Screening: Verify that each suspect meets CMS requirements for legitimate risk adjustment activity. Suspects must be routed for independent clinical evaluation, not presented as directives to code specific diagnoses. The provider must exercise independent medical judgment in determining whether the condition is present and active
• Prioritization Logic: With potentially thousands of suspects in a population, prioritization determines which suspects receive attention first. Prioritize by RAF impact (higher-value HCCs first), capture probability (higher-confidence suspects first), and time remaining in the payment year (urgency increases as the year progresses). Common coding errors should also inform which suspect categories receive additional validation scrutiny

Provider Engagement for Capture

Suspect condition capture ultimately depends on provider action. The provider must see the member, evaluate the suspected condition, determine whether it is clinically present and active, document their findings, and assign appropriate diagnosis codes. How plans engage providers determines capture success rates.

• Pre-Visit Alerts: The most effective delivery mechanism. Before a scheduled appointment, the provider receives a list of suspected conditions for that specific member with supporting evidence (prior-year diagnosis history, current medications, relevant lab values). Providers who review suspects before the encounter address them 70-80% of the time; providers receiving suspects after the visit address them less than 20% of the time
• EHR Integration: Embed suspect alerts directly in the EHR workflow so they appear during the encounter without requiring providers to access a separate system. EHR-integrated alerts achieve the highest engagement rates because they meet providers where they work. Integration with risk stratification systems enriches the clinical context
• Chart Review Programs: For suspects that cannot be addressed during routine visits, deploy retrospective chart review teams. Certified coders review existing documentation to determine if the suspected condition is already present in the clinical record but was not coded. Chart reviews capture 15-25% of suspects without requiring additional provider encounters
• Annual Wellness Visits: Medicare Annual Wellness Visits provide a dedicated opportunity to review and document chronic conditions comprehensively. Plans that coordinate suspect condition lists with AWV scheduling achieve systematically higher recapture rates because the visit is specifically designed for comprehensive health assessment
• Provider Education: Providers who understand why suspect conditions matter are more engaged in capture. Education should cover three points: the CMS-HCC annual recapture requirement, the clinical value of comprehensive problem list documentation, and how accurate coding supports appropriate resource allocation for their patients

Technology for Suspect Condition Management

At scale, suspect condition management requires technology platforms that automate identification, validation, routing, tracking, and measurement. Manual suspect management is feasible for plans with fewer than 5,000 members but breaks down rapidly above that threshold.

• Data Integration Layer: A platform that ingests and normalizes data from claims, pharmacy, lab, eligibility, and encounter systems into a unified member record. Without this foundation, suspect identification operates on incomplete data and produces unreliable results. Data refresh frequency matters: weekly or more frequent updates are necessary for timely suspect generation
• Suspect Engine: Algorithmic components that apply identification rules (year-over-year comparison, pharmacy crosswalk, lab inference, predictive models) to the integrated data and generate member-level suspect lists with confidence scores and supporting evidence. The engine must support both CMS-HCC V28 and V24 mapping to accurately calculate the RAF impact of each suspect
• Provider Portal: A web-based interface where providers access their patient-level suspect lists, review supporting evidence, and indicate disposition (confirmed, ruled out, deferred). The portal should integrate with or supplement existing EHR workflows rather than creating a separate workflow that competes for provider attention
• Workflow Tracking: End-to-end tracking from suspect generation through provider delivery, encounter completion, coding, and claims submission. Tracking reveals bottlenecks: are suspects reaching providers? Are providers addressing them? Are addressed suspects being coded? Are coded suspects appearing in submitted claims? Each step has its own failure mode
• Analytics Dashboard: Real-time visibility into capture rates by provider, condition category, suspect type, and time period. Dashboards enable program managers to identify underperforming providers, problematic condition categories, and seasonal patterns that inform intervention timing
• Compliance Controls: Built-in guardrails ensuring that suspect routing complies with CMS guidelines. Suspects must be presented as clinical prompts for evaluation, not coding directives. Audit trails document every step from identification through capture to demonstrate program integrity

Measuring Capture Rates

Capture rate measurement provides the performance metrics that drive program optimization. Without measurement, suspect programs operate blindly, unable to distinguish effective strategies from ineffective ones.

• Overall Capture Rate: The percentage of identified suspects that are ultimately captured (documented, coded, and submitted) within the payment year. Industry benchmarks: 60-70% for mature programs, 35-50% for emerging programs. Plans below 35% have fundamental delivery or engagement issues requiring investigation
• Capture Rate by Type: Differentiate performance by suspect source. Prior-year HCC recapture should achieve 80-90% for chronic conditions. Pharmacy suspects typically achieve 40-60%. Lab suspects achieve 30-50%. Clinical algorithm suspects achieve 25-40%. Significant deviation from these benchmarks by type indicates issues with specific identification methods or provider engagement approaches
• Provider-Level Performance: Track capture rates by individual provider and provider group. Variation is expected, but providers consistently below 30% capture rates need targeted support: more effective alert delivery, additional education, or assessment of whether their patient panel matches the suspects being routed
• Time-to-Capture: Measure the elapsed time from suspect generation to capture. Suspects captured within 60 days of generation have the highest value because they provide maximum runway for the condition to contribute to the full-year RAF score. Suspects captured in the last 30 days of the payment year may miss CMS submission deadlines
• RAF Impact: Calculate the actual RAF score improvement attributable to captured suspects. This metric connects the suspect program directly to revenue: total RAF improvement multiplied by approximately $10,400 per 1.0 RAF point yields the program's revenue contribution. Risk stratification accuracy also improves as captured suspects produce more complete member risk profiles
• False Positive Rate: Track the percentage of suspects that providers review and rule out as not clinically present. Rates above 20% indicate that identification algorithms need refinement. High false positive rates damage provider trust and reduce engagement with legitimate suspects