Computing hospitalization rates in presence of repeated events: impact and countermeasures to avoid misinterpretation

RATIONALE, AIMS AND OBJECTIVES: The admission rate, including both first and recurrent events, is a clear overall measure of hospital utilization, its variability accounting for individual propensity to disease recurrence. METHOD: In this paper, we compared two variance estimators derived from the Poisson and negative binomial distribution of directly and indirectly age/gender-standardized hospitalization rates allowing for multiple events. The latter approach accommodates departures from the assumption of randomness of repeated events required by the Poisson distribution. We apply these methods to a retrospective cohort based on hospital discharge data in 2001 of Piedmont (north-western Italy) residents. RESULTS: Estimated standard errors under the negative binomial for both directly and indirectly standardized rates result in almost twice those under the Poisson distribution. CONCLUSION: Our analysis confirms that ignoring the typical non-random nature of repeated events underestimates the true variance of rates and can lead to biased optimistic interpretation of study results.