East Coast Ballroom
Estimating the causal effect of an observation versus inpatient stay on 30-day readmission: Comparison to risk-standardized estimates and implications for quality measurement (307904)Mike Baiocchi, Stanford University
Gabriel Escobar, Division of Research, Kaiser Permanente Northern California
*Ben Marafino, Stanford University
Alejandro Schuler, Division of Research, Kaiser Permanente Northern California
Keywords: risk adjustment,causal inference,quality measurement,readmission,prediction modeling
Since the implementation of the Hospital Readmissions Reduction Program (HRRP) in 2012, the use of observation stays as an alternative to inpatient hospitalization has significantly increased. From the viewpoint of quality measurement, this practice may be problematic as these stays are not considered “index” events, and so are not included in 30-day readmission measures used by the HRRP and similar initiatives. The extent to which diverting observation stay patients from the “denominator” of quality measures may artificially deflate, or even inflate, these measures remains an open question. Using a rich dataset derived from a cohort of patients admitted to 21 KPNC hospitals, together with a form of inverse-propensity weighting, we estimate the causal effects of an observation stay on 30-day readmission across a range of conditions, and compare these to risk-standardized estimates. We find that, compared to causal estimates, risk standardization appears to underestimate risk among patients admitted for observation. As a result, diverting would-been inpatients to observation stays, and thus excluding these stays from risk-adjusted measures, may overstate hospital performance.