Attributable mortality from extensively drug-resistant gram-negative infections using propensity-matched tracer antibiotic algorithms

Background: Tracer antibiotic algorithms using administrative data were investigated to estimate mortality attributable to extensively drug-resistant gram-negative infections (GNIs). Methods: Among adult inpatients coded for GNIs, colistin cases and 2 comparator cohorts (non-carbapenem β-lactams or carbapenems) treated for ≥4 consecutive days, or died while receiving the antibiotic, were separately propensity score-matched (1:2). Attributable mortality was the in-hospital mortality difference among propensity-matched groups. Infection characteristics and sepsis severity influences on attributable mortality were examined. Algorithm accuracy was assessed by chart review. Results: Of 232,834 GNIs between 2010 and 2013 at 79 hospitals, 1,023 per 3,350 (30.5%) colistin and 9,188 per 105,641 (8.7%) β-lactam (non-carbapenem) comparator cases died. Propensity-matched colistin and β-lactam case mortality was 29.2% and 16.6%, respectively, for an attributable mortality of 12.6% (95% confidence interval 10.8-14.4%). Attributable mortality varied from 11.0% (7.5%-14.7%) for urinary to 15.5% (12.6%-18.4%) for respiratory (P <.0001), and 4.6% (2.1%-7.4%) for early (≤4 days) to 16.6% (14.3%-18.9%) for late-onset infections (P <.0001). Attributable mortality decreased to 7.5% (5.6%-9.4%) using a carbapenem comparator cohort but increased 9-fold in patients coded for severe sepsis or septic shock (P <.0001). Our colistin algorithm had a positive predictive value of 60.4% and sensitivity of 65.3%. Conclusions: Mortality attributable to treatment-limiting resistance during GNIs varied considerably by site, onset, and severity of infection.