Identifying complications and low provider adherence to normative practices using administrative data.

Abstract

OBJECTIVE: This study investigated whether unexpected length of stay (LOS) could be used as an indicator to identify hospital patients who experienced complications or whose care exhibited low adherence to normative practices. DATA SOURCES AND STUDY SETTING: We analyzed 1,477 cases admitted for one of three medical conditions. All cases were discharged from one of nine participating Department of Veterans Affairs (VA) hospitals from October 1987 through September 1989. Analyses used administrative data and information abstracted through chart reviews that included severity of illness indicators, complications, and explicit process of care criteria reflecting adherence to normative practices. STUDY DESIGN: We developed separate multiple linear regression models for each disease using LOS as the dependent measure and variables that could be assumed present at the time of admission as explanatory variables. Unexpectedly long LOS (i.e., discharges with high residuals) was used to target complications and unexpectedly short LOS was used to target cases whose care might have exhibited low adherence to normative practices. Information gleaned from chart reviews served as the gold standard for determining actual complications and low adherence. PRINCIPAL FINDINGS: Analyses of administrative data showed that unexpectedly long LOS identified complications with sensitivities ranging from 40 through 62 percent across the three conditions. Positive predictive values all were at greater than chance levels (p < .05). This represented substantial improvement over identification of complications using ICD-9-CM codes contained in the administrative database where sensitivities were from 26 through 39 percent. Unexpectedly short LOS identified low provider adherence with sensitivities ranging from 33 through 45 percent with positive predictive values all above chance levels (p < .05). The addition to the LOS models of chart-based severity of illness information helped explain LOS, but failed to facilitate identification of complications or low adherence beyond what was accomplished using administrative data. CONCLUSIONS: Administrative data can be used to target cases when seeking to identify complications or low provider adherence to normative practices. Targeting can be accomplished through the creation of indirect measures based on unexpected LOS. Future efforts should be devoted to validating unexpected LOS as a hospital-level quality indicator. RELEVANCE/IMPACT: Scrutiny of unexpected LOS holds promise for enhancing the usefulness of administrative data as a resource for quality initiatives.