Answers are grounded in this paper and cite their sources.
Background: We investigate the policy implications of statistically valid upper limits on anesthesia durations. We test whether these tolerance limits categorized by claims data (procedure, surgeon, and facility) differ in practical consequence from limits calculated with data pooled among surgeons and facilities. We evaluate whether using valid tolerance limits at typical insurer sample sizes would differ functionally from no limits. Methods: Simulations had 100,000 replicates per configuration with 8 surgeons per hospital: equal n = 10 or 300 cases per-surgeon or unequal (geometric) median n = 10 and 60. The within-surgeon coefficient of variation was 36.2%. The between-surgeon variance accounted for 22.4% of total log-time variance (intraclass correlation). Five tolerance-limit methods (pooled Weibull generalized pivotal quantity, pooled nonparametric, per-surgeon log-normal, per-surgeon 3-sigma, and per-surgeon nonparametric) were evaluated by coverage probability (95% was correct), case-volume-weighted percentage bias (0% best), functional quantile (90 th for 90 th percentile), and payment satisfaction (100% ideal). Results: Pooled mixtures of cases by procedure were best fit by log-normal distributions. The noncentral-t–based log-normal tolerance limit, calculated per combination of procedure, surgeon, and facility, attained nominal 95% coverage in every configuration. Pooled Weibull and pooled nonparametric limits failed often (coverage 70.9%–79.2% and 62.9%–83.8%). The nonparametric method with all claim data attained 95.4%–97.1% coverage but was unavailable for 85% of surgeons at the realistic condition of unequal sample sizes, median n = 10. Applying the exact log-normal-based tolerance limits on 90 th percentiles functionally achieved percentiles ranging from the 92.1-97.4 th , with resulting payment satisfaction rates of 97.9%-100%. Conclusions: The per-surgeon log-normal 95% upper confidence limit for the 90 th percentile on log-transformed durations was the only valid method computable at realistic sample sizes. Applying those actuarially valid tolerance limits would pay ≈99.4% of historical case-time in full, thus functionally equivalent to retaining no paid time limit.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: