Approaches to Analysis of Length of Hospital Stay Related to Antibiotic Therapy in a Randomized Clinical Trial: Linezolid versus Vancomycin for Treatment of Known or Suspected Methicillin‐Resistant Staphylococcus Species Infections

Abstract

As length of hospital stay (LOS) represents about 70-90% of the total cost of treating serious infections, it represents a key variable in analyzing the health economic differences between treatments for hospitalized patients. In a retrospective analysis using LOS data from a multinational, randomized, phase III clinical trial, we examined two methods (the log-logistic model and Kaplan-Meier survival function) and three approaches (unadjusted total LOS, total LOS adjusted for nontreatment factors, and adjusted LOS based on antibiotic treatment [the antibiotic treatment LOS]) for estimating antibiotic treatment effect on LOS and determined if these approaches could reduce the variation in LOS and control for the imbalance between treatment groups. The trial enrolled patients who were hospitalized with known or suspected Staphylococcus species infections who received at least one dose of linezolid or vancomycin (intent-to-treat sample) and who continued taking the study drug for at least 7 days (clinically evaluable sample). In the intent-to-treat sample, the linezolid group had a 2- (unadjusted) or 4-day (adjusted for nontreatment factors) shorter LOS at the 25th percentile; a 1- or 2-day advantage, respectively, at the 50th percentile (median); and a 0.6- or 1.6-day mean LOS advantage, compared with the vancomycin group. With the antibiotic treatment LOS approach, the linezolid group had mean and median LOS reductions comparable to or greater than those seen in the nontreatment-factor-adjusted results. Results for the clinically evaluable sample were similar to those of the intent-to-treat sample, but the differences between the treatment groups were greater. Linezolid-treated patients had significant LOS reductions that otherwise would be masked without the use of more appropriate, but less commonly used, methods.