Estimation of extra hospital stay attributable to nosocomial infections: heterogeneity and timing of events

Infections acquired in hospital are likely to affect the duration of hospitalization. Suitable statistical methods for estimating the extra days spent in hospital due to nosocomial infections should allow modeling of the heterogeneity of the patient population and the timing of events, as failure to account for important covariates and failure to model adequately the timing of events may lead to biased results. Three approaches have been used in the past to estimate the extra stay: a comparison of duration of stay of infected and uninfected patients, matching of infected and uninfected patients with respect to potentially important determinants of the length of hospital stay, and matching for time-to-infection in addition to the other factors. While these approaches can allow for the heterogeneity of the patient population, none takes sufficient account of the real timing of events and may overestimate the effect of nosocomial infections. We explored the statistical methods available for analyzing time-to-event data and derived alternative methods to estimate the extra stay that appropriately account for heterogeneity and timing. Data from two prospective cohort studies on postoperative wound infection and on nosocomial pneumonia showed that the two-group comparison yields the highest estimates of extra stay (21 and 14 extra days), while matching for confounders and time reduced the estimates to 11 and 8 extra days; our methods yield even lower results (10–12 and 3–4 extra days).