Observational studies of early versus late salvage therapies in critical care exhibit intrinsic selection bias: two meta-analyses

Abstract

It is difficult to determine the optimal timing of salvage therapies, such as initiation of renal replacement therapies (RRT), using non-experimental designs. Therefore, using timing of RRT as a motivating example, we performed meta-analyses comparing observational and experimental studies assessing timing of RRT and timing of invasive mechanical ventilation (IMV). We performed two meta-analyses of observational and experimental studies testing the association of early versus late initiation of RRT and IMV on mortality. We included 72 studies for RRT (57 observational, 15 experimental) and 50 for IMV (48 observational, 2 experimental). For RRT, observational studies showed mortality benefit with early RRT (OR 0.52, 95% CI 0.42-0.63) that was not seen in experimental studies (OR 0.94, 95% CI 0.76-1.17). For IMV, observational studies demonstrated harm with early IMV (OR 1.25, 95% CI 1.03-1.52), although not to the degree of experimental studies (OR 1.86, 95% CI 0.90-3.86). When observational studies were restricted to subjects who all received IMV, conclusions were further biased towards benefit favoring early IMV (OR 0.75, 95% CI 0.55-1.02). Studies that also included subjects who were never intubated showed harm with early IMV (OR 1.63, 95% CI 1.30-2.04). There were significant differences in the results of observational and experimental studies looking at timing of salvage therapies, partly due to selection bias in observational studies. This issue was worsened by only including subjects who receive the therapy. Randomized trials using objective eligibility criteria remain the best method to determine optimal timing of salvage therapies.