Machine-Learning Clinical Decision Support Tool Improves UTI Treatment
A new study shows that implementing a machine learning-based clinical decision support system in primary care practices significantly improved UTI treatment success.
Source: Getty Images
By - Researchers have found that a machine learning-based clinical decision support system (CDSS) significantly impacted treatment success and antibiotic prescription behavior for urinary tract infections (UTIs) when implemented in primary care practices.
According to the study, which was published earlier this month in JMIR Medical Informatics, UTIs are a significant health burden worldwide, but most clinical trials on UTI treatments are conducted on female patients with uncomplicated infections. This limits the scientific evidence for effective treatments of complicated UTIs, which can impact clinical decision-making.
The researchers further noted that machine-learning CDSS methods are rarely evaluated in clinical practice, so they sought to assess how a CDSS would impact the antibiotic prescription behavior of general practitioners (GPs) and the success of their UTI treatments. Because the CDSS would contain data on all types of UTI patients, including those with complications, the researchers hypothesized that its use could facilitate better clinical decision-making.
To test their hypothesis, the researchers provided CDSS software designed to aid GPs with treatment choices for UTI patients to 36 primary care practices in the Netherlands.
The dataset for the development of the CDSS was based on EHR data from patients over the age of 12 who had received antibiotic treatment for at least one UTI between 2012 and 2014. Treatment success within this dataset was defined as a subsequent period of 28 days in which no new treatment was necessary. From this, classifiers were constructed to estimate the probability of success for eight antibiotics commonly used for UTI treatment.
The final dataset consisted of 122,203 UTIs diagnosed at 264 primary care facilities. Because of the anatomical differences between male and female patients, the 15 machine-learning models that comprised the CDSS were split based on sex: eight for female patients and seven for male.
Following CDSS development, clinicians from the practices participating in the study were trained on how to use the software. They used the CDSS for four months. Also included was a control group of 29 practices that provided a benchmark for evaluating the study’s findings. UTI data from before the study began were gathered from the participating practices for comparison purposes.
In practices using the CDSS, the proportion of successful treatments rose from 75 percent to 80 percent throughout the four-month implementation period. In the control practices, the proportion remained at 76 percent. For treatments in which the software was used, the proportion of successful treatments was 83 percent.
When measuring treatment outcomes based on sex, age, comorbidities, and whether the UTI was complicated, the researchers found that using the CDSS significantly improved outcomes for female patients and patients over 70.
But GP antibiotic prescription behavior was not significantly impacted by CDSS use.
These findings indicate that there is potential for machine-learning methods to assist with clinical decision-making and improve outcomes in primary care practices. However, further study and validation of machine learning-based CDSS models in clinical practice are needed, researchers said.
