LAUSR

Abstract Objective To estimate the effect of treatment duration on in-hospital mortality in patients with Staphylococcus aureus blood stream infection and demonstrate the biases that can arise when immortal-time bias is ignored. Exposure We compared three treatment strategies: short therapy (<10 days), intermediate (10–18 days) and long (>18 days). Main outcome measures Twenty-eight-day all-cause in-hospital mortality. Methods Using data from the BSI-FOO study, we implemented an approach proposed by Hernán to overcome confounding and immortal-time biases. The first stage is to clone all participants, so that each participant is assigned to each treatment strategy. Second, observations are censored when their data becomes inconsistent with their assigned strategy. Finally, inverse-probability weights are applied to adjust for potential selection. We compared our results to a naïve approach where immortal-time bias is ignored. Results Of the 1903 participants in BSI-FOO, 587 were eligible and included in the analysis. After cloning, the weighted estimates of hazard ratio of mortality for short versus long therapy was 1.74 (95% CI 1.36, 2.24) and for intermediate versus long therapy was 1.09 (0.98, 1.22). In the naïve approach, the hazard ratios with reference to the long therapy group are 37.4 (95% CI 18.9 to 74.4) in the short therapy group and 4.1 (95% CI 1.9 to 8.9) in the intermediate therapy group. Conclusions Our findings suggest that duration of therapy >18 days is beneficial with respect to 28-day in-hospital mortality, however, there remains uncertainty around the efficacy of reducing duration of treatment to 10–18 days.