Missing Data is Informative
Over 30% of metrics are missing. In healthcare PE diligence, sellers who don't provide denial/AR data often have worse-than-average performance. Bayesian estimates below are shrunk toward peer priors — treat as conservative baselines, not forecasts.
Missing: Denial Rate, Clean Claim Rate, Net Collection Rate, First Pass Resolution, Appeals Overturn Rate, Days In Ar
Bayesian-Calibrated KPI Estimates
Each metric combines peer-group prior (from 167-bed hospital benchmarks) with observed data. Shrinkage shows the weight given to prior vs observed: 0% = all data, 100% = all prior. Green bar = data weight, amber = prior weight.
| Metric | Prior | Observed | Posterior | 90% CI | Shrinkage | Data vs Prior | Quality |
|---|---|---|---|---|---|---|---|
| Denial Rate | 8.5% | — | 8.5% | [0.3%, 16.7%] | 100% | Prior Only | |
| Clean Claim Rate | 92.5% | — | 92.5% | [85.7%, 99.3%] | 100% | Prior Only | |
| Net Collection Rate | 96.3% | — | 96.3% | [91.1%, 100.0%] | 100% | Prior Only | |
| First Pass Resolution | 78.0% | — | 78.0% | [64.6%, 91.4%] | 100% | Prior Only | |
| Appeals Overturn Rate | 45.0% | — | 45.0% | [27.1%, 62.9%] | 100% | Prior Only | |
| Days In Ar | 42.5 | — | 42.5 | [32.0, 53.0] | 100% | Prior Only | |
| Cost To Collect | 2.1% | — | 2.1% | [1.6%, 2.6%] | 100% | Prior Only | |
| Dnfb Days | 5.2 | — | 5.2 | [3.9, 6.5] | 100% | Prior Only | |
| Charge Lag Days | 2.8 | — | 2.8 | [2.1, 3.5] | 100% | Prior Only |
Methodology: Hierarchical Bayesian Partial Pooling
Rate metrics (denial rate, collection rate, clean claim rate) use Beta-Binomial conjugate updating. Continuous metrics (AR days, cost to collect) use Gamma/Normal approximation. Priors are stratified by hospital type (large/medium/small/rural). With zero observations, the posterior equals the prior. As n increases, the posterior converges to the observed value and the credible interval narrows. This prevents unstable estimates from thin data while letting real evidence dominate.