Comparison of censored regression and standard regression analyses for modeling relationships between antimicrobial susceptibility and patient- and institution-specific variables.

In order to identify patients likely to be infected with resistant bacterial pathogens, analytic methods such as standard regression (SR) may be applied to surveillance data to determine patient- and institution-specific factors predictive of an increased MIC. However, the censored nature of MIC data (e.g., MIC < or = 0.5 mg/liter or MIC > 8 mg/liter) imposes certain limitations on the use of SR. In order to investigate the nature of these limitations, simulations were performed to compare a regression tailored for censored data (censored regression [CR]) and one tailored for an SR. By using a model relating piperacillin-tazobactam MICs against Enterobacter spp. to patient age and hospital bed capacity, 200 simulations of 500 isolates were performed. Various MIC censoring patterns were imposed by using 26 left- or right-censored (L,R) pairs (i.e., MICs < or = 2 mg/liter(L) [2L] or MICs > 2 mg/liter(R) [2R], respectively). Data were fit by CR and SR for which censored MICs were either (i) excluded, (ii) replaced by 2L or 2R, or (iii) replaced by 2(L - 1) or 2(R + 1). Total censoring for the 26 pairs ranged from 7 to 86%. By CR, deviations of average parameter estimates from the true parameter values were <0.10 log2 (mg/liter) for all parameters for each of the 26 pairs. By SR, these deviations were >0.10 log2 (mg/liter) for at least 18 of the 26 pairs for all but one parameter. Two-standard-error confidence intervals for individual parameters contained as little as 0% of cases for all SR approaches but > or = 91.5% of cases for the CR approach. When censored MIC data are modeled, CR may reduce or eliminate biased parameter estimates obtained by SR.