Pitfalls of calculating hospital readmission rates based on nonvalidated administrative data sets

Abstract

Object Administrative databases are increasingly being used to establish benchmarks for quality of care and to compare performance across peer hospitals. As proposals for accountable care organizations are being developed, readmission rates will be increasingly scrutinized. The purpose of the present study was to assess whether the all-cause readmissions rate appropriately reflects the University of California, San Francisco (UCSF) Medical Center hospital's clinically relevant readmission rate for spine surgery patients and to identify predictors of readmission. Methods Data for 5780 consecutive patient encounters managed by 10 spine surgeons at UCSF Medical Center from October 2007 to June 2011 were abstracted from the University HealthSystem Consortium (UHC) using the Clinical Data Base/Resource Manager. Of these 5780 patient encounters, 281 patients (4.9%) were rehospitalized within 30 days of the previous discharge date. The authors performed an independent chart review to determine clinically relevant reasons for readmission and extracted hospital administrative data to calculate direct costs. Univariate logistic regression analysis was used to evaluate possible predictors of readmission. The two-sample t-test was used to examine the difference in direct cost between readmission and nonreadmission cases. Results The main reasons for readmission were infection (39.8%), nonoperative management (13.4%), and planned staged surgery (12.4%). The current all-cause readmission algorithm resulted in an artificially high readmission rate from the clinician's point of view. Based on the authors' manual chart review, 69 cases (25% of the 281 total readmissions) should be excluded because 39 cases (13.9%) were planned staged procedures; 16 cases (5.7%) were unrelated to spine surgery; and 14 surgical cases (5.0%) were cancelled or rescheduled at index admission due to unpredictable reasons. When these 69 cases are excluded, the direct cost of readmission is reduced by 29%. The cost variance is in excess of $3 million. Predictors of readmission were admission status (p < 0.0001), length of stay (p = 0.0001), risk of death (p < 0.0001), and age (p = 0.021). Conclusions The authors' findings identify the potential pitfalls in the calculation of readmission rates from administrative data sets. Benchmarking algorithms for defining hospitals' readmission rates must take into account planned staged surgery and eliminate unrelated reasons for readmission. When this is implemented in the calculation method, the readmission rate will be more accurate. Current tools overestimate the clinically relevant readmission rate and cost.