Workers' Compensation Red-Flag Score with Limited Resources
A workers' compensation carrier insures a 300-employee construction firm. The SIU analyst pulls 620 indemnity claims from 2021-2023. Lost-time claims average 97 days versus a 55-day book-of-business benchmark. Forty-one percent of claims have a Monday date of injury, 22 claims name the same site supervisor as the only witness, and 14 claimants were treated by the same chiropractor within 10 miles of the job site. The firm has high turnover, overtime, and a 2022 safety citation. The claims vice president wants a red-flag score to refer the worst 50 claims to investigation. There is no surveillance budget for all 620 claims, and the carrier must avoid targeting injured workers based on protected characteristics.
1. Design a scored triage using only the fields described. Identify four to six factors, how you would weight them, and how you would validate the score.
Read the worked answer
Use a transparent points score, not a black-box probability. Candidate factors: Monday injury report (1 point), supervisor named as sole witness (1), same chiropractor (1), lost days above the 55-day benchmark (2), tenure under 90 days (1), prior claim in the past 24 months (1), attorney representation within 30 days (1), and wage statement inconsistent with reported job (2). Weights should reflect univariate lift against closed claims, not intuition alone. Compute the score for all 620, sort descending, and take the top 50 only as a review queue. Validate by back-testing on 2021-2022 closed claims with known outcomes and by auditing a random sample of low-scoring claims to estimate missed cases. Track precision and false-positive themes. Remove any factor that is a proxy for age, sex, race, national origin, or disability. The score supports prioritization; it does not establish fraud.
2. The Monday rate is 41%. Is that evidence of fraud? Show the reasoning and what comparison you need.
Read the worked answer
41% is much higher than the roughly 20% expected if injury days were uniform across a five-day week, and with 620 claims that gap is unlikely to be chance alone. But the comparison is not proof of fraud. Construction injuries may be reported on Monday because weekend symptoms surface then, because Monday is the first administrative day, or because date-of-injury fields are defaulted. I would compare Monday rates by injury type, by reporting lag, by supervisor, and against the firm's prior years and industry peers. I would also test whether the date of injury differs from the date reported. The Monday pattern is a reason to examine reporting practices, not to conclude that workers are fabricating injuries.
3. What three metrics would you track after referrals to see whether the score works?
Read the worked answer
Track referral precision: the share of referred claims that SIU accepts and the share with a substantiated finding. Track outcomes: average paid and lost days for referred versus matched non-referred claims, and the time from referral to closure. Track process: how many top-50 referrals are closed with no action and what themes explain them. Also audit a random sample of non-referred claims to estimate false negatives. If the score only produces denials and no accepted cases, the weights or threshold are wrong. If it produces accepted cases but also large numbers of legitimate claims, refine the factors. These metrics keep the triage honest and show whether the carrier is finding fraud or just punishing injury reports.