Machine Learning Validates Guidelines for Treating UTI

Analyses based on machine learning-derived, or on infectious diseases expert-curated confounders agree that guidelines for treating uncomplicated urinary tract infection, last issued by the Infectious Diseases Society of America (IDSA) in 2011, remain valid despite changes in epidemiology since their release.

Sanjat Kanjilal, MD, MPH, Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Healthcare Institute, Boston, MA, and David Sontag, PhD, Institute for Medical Engineering and Sciences, Massachusetts Institute of Technology, Cambridge, MA, and colleagues found the two analyses congruent in determining that the guideline treatments are appropriate against currently circulating pathogen strains and patterns of infection.

"In this cohort study of patients with uncomplicated UTI derived from a large regional claims dataset, national treatment guidelines published almost 14 years ago continue to recommend optimal treatments.These results also provide proof-of-principle that automated feature extraction methods for OMOP (Observational Medical Outcomes Partnership common data model) formatted data can emulate manually curated models," Kanjilal, Sontag and colleagues report.

The investigators accessed an electronic health records (EHR) dataset containing inpatient, outpatient, laboratory and pharmacy claims made between 2012 and 2021.They identified a cohort of 57,585 episodes of UTI occurring in 49,037 nonpregnant female adult patients.Antibiotics designated first line per guideline--nitrofurantoin or trimethoprim-sufamethoxazole--were prescribed in 61% of episodes, fluoroquinolones in 37%, and ß-lactams in 2%.

The machine learning analysis was accomplished with a software package developed by Sontag to build features from the OMOP-formatted dataset automatically. The development was facilitated by the EHR dataset being formatted in the OMOP common data model, Kanjilal explained in an interview² with JAMA Associate Editor Yulin Hswen, ScD, MPH.

"A common data model is simply a way to translate institution-specific data into a common syntax that allows universal algorithms to work on it," Kanjilal said.

Separate from the machine learning analysis, Kanjilal collaborated with study co-author Sonali Advani, MBBS, MPH, Division of Infectious Diseases, Department of Medicine, Duke University School of Medicine, Durham, NC, in constructing an analysis of treatment outcomes that accounted for conditions they thought would be most predictive of treatment response.

"They're what you think they would be: comorbidities, visits to the hospital, prior infections, etc," Kanjilal commented.

The primary outcomes of guideline-concordant and guideline-discordant antibiotic treatments was a composite end point for treatment failure--defined as outpatient or inpatient revisit within 30 days for UTI, pyelonephritis or sepsis; and adverse events commonly associated with antibiotics—gastrointestinal symptoms, rash, kidney injury, and Clostridium difficile infection.

After adjustment, receipt of the guideline first-line treatments for uncomplicated UTI, was associated with an absolute risk difference of -1.78% (95% CI -2.37 to -1.06%) relative to fluoroquinolones for having a revisit for UTI within 30 days of diagnosis. Relative to ß-lactam antibiotic, the guideline first-line treatments were associated with an absolute risk difference of -6.40% (-10.14 to -3.24%).Differences in adverse events were similar between all comparators.

The investigators conclude that both analyses show that guideline first-line treatments for uncomplicated UTI remain the drugs of choice on measures of effectiveness and adverse events—barring a patient history of drug resistance or intolerance, or in regions with high local rates of resistance.

"That's really good news for antimicrobial stewardship programs that want newer evidence to support using guideline-concordant therapy," Kanjilal told Hswen.

The congruence of the analyses with machine learning-derived, or expert-curated factors also suggests opportunities for wider application of machine learning evaluations at the health center level.

"That raises the possibility of doing these types of studies with observational data on a more turn-key basis, where we only need my support to validate the results, not to pick the features," Kanjilal remarked. "We hope that will lower the barrier for these types of analyses and allow us to constantly update the results as new data accrue, which I think is the core of any learning health system."

References

1.Jones N, Shih M-C, Healey E, et al. Use of machine learning to assess the management of uncomplicated urinary tract infection. JAMA Netw Open 2025 Jan 31;8(1):e2456950. doi:10.1001/jamanetworkopen.2024.56950.Accessed Feb 18, 2025.

2.Hswen Y, Rubin R. Researchers use machine learning to put older clinical guidelines to the test. JAMA 2025 Feb 14. doi:10.1001/jama.2024.28105. Online ahead of print.Accessed Feb 18, 2025.