P-782. Developing a Multivariable Prediction Model of Antibiotic Heteroresistance to First-Line Therapies for Urinary Tract Infection

TL;DR

This study creates a model to predict antibiotic heteroresistance in urinary tract infections caused by Escherichia coli, using clinical factors like cognitive disease and age.

Contribution

A novel multivariable model using clinical risk factors to predict antibiotic heteroresistance in urinary Escherichia coli isolates.

Findings

Cognitive disease, age, and sex were significant predictors of heteroresistance in a multivariable model.

The model had modest predictive performance with an AUC of 0.64 in the main analysis and 0.69 in a subgroup analysis excluding resistant isolates.

Abstract

Heteroresistance (HR), where a subpopulation of bacteria is phenotypically resistant while most of the population appears susceptible, is not routinely tested for despite its potential association with antibiotic failure. Understanding clinical predictors of HR may improve antibiotic selection. Thus, we aimed to develop and validate a model predicting HR to oral antibiotics for urinary tract infection (UTI) in urinary Escherichia coli isolates.

We analyzed urinary E. coli isolates collected during a surveillance pilot performed by the CDC-funded Georgia Emerging Infections Program. HR to nitrofurantoin, fosfomycin, or sulfamethoxazole-trimethoprim was assessed by population analysis profiling (PAP) and defined as bacterial survival >10-6 but < 50% at 1x minimum inhibitory concentration. Covariates were collected by chart review. We used logistic regression with forward selection to build models predicting HR to ≥1 antibiotic, assessed performance using area under the curve (AUC) and Hosmer-Lemeshow (HL) goodness-of-fit, and performed n-fold cross-validation in SAS. All models controlled for age and sex. A subgroup analysis excluded isolates resistant by PAP (comparing susceptible to HR).

Of 349 isolates included, 275 (79%) were from females with median age 56 (Table 1). Forty-four (13%) were HR to ≥1 antibiotic. In unadjusted analyses, an indwelling device in place prior to collection, chronic kidney disease, and cognitive disease were significantly associated with HR to ≥ 1 antibiotic (Table 1). The final multivariable model included cognitive disease, age, and sex (AUC = 0.64, HL = 0.81) (Figure 1). After cross-validation the mean AUC was 0.53. In a subgroup analysis excluding resistant isolates, 38/271 (14%) were HR to ≥1 antibiotic. In this subgroup, the model predicting HR to ≥1 antibiotic included cognitive disease, neuropathy, age, and sex (AUC = 0.69, HL = 0.86). Cross-validation yielded a mean AUC of 0.55 (Figure 2).

We created a novel model using clinical risk factors to predict HR to first-line antibiotics used for UTIs. This model provided modest prediction of HR but may have limited generalizability when used on other populations. Future work should increase sample size for model development and validate on outside populations.

Lucy S. Witt, MD, MPH, Merck & Co: Grant/Research Support

Figures