Attributable Mortality from Extensively Drug Resistant Gram-Negative Infections using Tracer Antibiotic Algorithms

Sameer Kadri; Jeffrey Stritch; Stephanie Bonne; Bruce Swihart; John Dekker; Rebecca Prevots; Tara Palmore; Bradley Freeman; Jillian Raybould; Nirav Shah; Devang Patel; Jennifer Husson; Robert Danner

The impact of extensive drug-resistance (XDR) on survival from gram-negative infections (GNIs) remains unclear. Tracer antibiotic algorithms may enable the estimation of mortality attributable to XDR among patients with GNIs using large administrative databases.

In a propensity-matched analysis, patients receiving colistin, marking them as having an extended spectrum gram negative infection will have higher attributable mortality than patients with gram negative infections without significant extended antibiotic resistance.

Adult inpatient encounters coded for GNIs and associated antibiotic administrations in the Vizient database were analyzed. Colistin cases were defined by >3 consecutive days of intravenous colistin or death while receiving colistin; comparator cases were similarly defined using select non-carbapenem ß-lactams instead. Colistin cases were matched (1:2) to comparator cases by propensity of receiving colistin. XDR attributable mortality was calculated as the difference in in-hospital mortality between propensity-matched groups and 95% confidence intervals (CI) using bootstrapping. Variation in attributable mortality by infection site and onset, sepsis strata, and a propensity-matched carbapenem comparator group was examined. Algorithm accuracy was tested using chart review at 3 hospitals.      

Of 232,834 GNI encounters between 2010-2013 at 79 hospitals, 905 of 3,350 (27%) colistin and 9,188 of 105,641 (8.7%) comparator cases died. Mortality among propensity-matched colistin (n=3,099) and comparator cases (n=6,198) was 29.2% and 16.6% respectively [attributable mortality =12.6% (CI 10.8-14.4%)]. This estimate of XDR attributable mortality varied considerably by site and onset of infection, ranging from 1.1% (-7.6-8.2%)  for bloodstream to 15.5 (12.6-18.4%) for respiratory (p<0.0001), and 4.6% (2.1-7.4%) for community vs. 16.6% (14.3-18.9%) for hospital onset (p<0.0001). Mortality attributable to XDR increased 3-fold when coded for sepsis and 9-fold when coded for severe sepsis/septic shock (p<0.0001). Using a carbapenem comparator group, attributable mortality decreased to 7.5 (5.6-9.4)%. Chart reviews demonstrated that colistin cases had a positive predictive value of 60.4% and sensitivity of 65.3% for detecting XDR GNIs.  The mortality varied significantly within antibiotic groups with the coding for sepsis and severe sepsis/shock.

Using tracer antibiotic algoirthms, mortality attributable to XDR during GNIs was estimated at 12.6%, but varied considerably by site and onset of infection and coding for sepsis or severe sepsis/septic shock.