Key result
Risk-adjusted hierarchical modeling characterizes facility-specific differences in surgical duration while accommodating case mix differences, though it cannot predict the longest case durations.
Answers are grounded in this paper and cite their sources.
Risk-adjusted hierarchical modeling is defended as a valid and necessary approach for benchmarking facility-level surgical durations, despite limitations in predicting specific case extremes.
We would like to thank Dexter and Epstein1 for their careful review of our study examining the Variability in Case Durations for Common Surgical Procedures.2 As we describe in our article, we constructed hierarchical parametric proportional hazards models to estimate the facility-level variability in surgical durations for common surgeries. Using this framework, we estimated facility-specific hazard ratios that represent the relative risk (for each facility) that a patient will exit the operating room compared to the average facility per unit time. We assumed that the hazard function followed a Weibull distribution, and we estimated the effect of facilities on surgical duration by specifying facilities as a random effect. We appreciate Dexter and Epstein’s1 suggestion that we could have used alternative approaches, such as generalized pivotal methods or binning, as described in their papers, to compare surgical duration times among hospitals. However, using risk-adjusted hierarchical modeling to benchmark provider performance is considered a best practice by the Society of Thoracic Surgeons, the American Heart Association, and the American College of Surgeons.3–5 Centers for Medicare and Medicaid Services uses this methodology in public reporting for Hospital Compare and in programs linking payment to performance such as the Hospital Readmission Reduction Program and bundled care.6 The approaches suggested by Dexter and Epstein1 to compare surgical duration times across facilities do not incorporate risk adjustment. Because providers may not care for similar patient populations, it is generally accepted that risk adjustment is necessary to ensure that differences in provider performance are not simply due to differences in patient case mix. Although the need for risk adjustment has strong face validity, it should be noted that because the National Anesthesia Clinical Outcomes Registry (NACOR) does not currently include information on body mass index, our risk adjustment is incomplete. However, our goal was to develop a risk-standardized measure of surgical duration that could serve as the basis for using NACOR to benchmark surgical duration once NACOR data quality is sufficiently robust to account for differences in case mix between facilities. Dexter and Epstein1 also suggest that we may have based our conclusion regarding the heterogeneity of facility performance on the comparison of one of the parameters of the Weibull distribution and not the other. In making this comment, it appears they have assumed that hierarchical parametric survival models estimate separate Weibull distribution for each of the facilities: “The authors’ survival analysis may have compared only one of the parameters, and assumed homogeneity of the second (‘ancillary’) parameter among hospitals.” However, the time-dependent portion of the hazard function for a parametric proportional hazards survival analysis model is restricted to follow a single distribution with identical parameters that do not vary across hospitals. For a Weibull distribution, the shape and scale parameters will not vary across hospitals by construction in proportional hazards analysis. Instead, the facility effect on surgical duration is modeled by specifying the facility as a random effect, which is a parsimonious way of distinguishing variation across hospitals. We agree, however, that unlike the methods suggested by Dexter and Epstein,1 which do not incorporate risk adjustment, our survival models cannot be used to predict the longest duration that cases at a specific facility will take. Our approach allows us to characterize facility-specific differences in surgical duration while accommodating differences in facility case mix, but they cannot be used to predict the shape of the distribution of surgical durations at specific facilities. We would again like to thank Dexter and Epstein,1 both of whom are recognized experts in this field, for their thoughtful letter. Laurent G. Glance, MDDepartments of Anesthesiology and Public Health SciencesUniversity of Rochester School of MedicineRochester, New YorkRAND HealthRANDBoston, Massachusetts[email protected] Richard P. Dutton, MD, MBAUS Anesthesia PartnersDallas, Texas Changyong Feng, PhDDepartment of Biostatistics and Computational BiologyUniversity of Rochester School of MedicineRochester, New York Yue Li, PhDDepartment of Public Health SciencesUniversity of Rochester School of MedicineRochester, New York Stewart J. Lustik, MD, MBADepartment of AnesthesiologyUniversity of Rochester School of MedicineRochester, New York Andrew W. Dick, PhDRAND HealthRANDBoston, Massachusetts
No takes yet. Share an insight, caveat, or question.
Glance et al. (2018) conducted a letter in Surgical procedures. Risk-adjusted hierarchical modeling was evaluated. Risk-adjusted hierarchical modeling characterizes facility-specific differences in surgical duration while accommodating case mix differences, though it cannot predict the longest case durations.
Synapse has enriched 10 related papers. Consider them for comparative context: