Machine-Learning Algorithms Predict C. Diff Risk Soon After Hospital Admission

Machine-Learning Algorithms Predict C. Diff Risk Soon After Hospital Admission

NEW YORK (Reuters Health) – Machine-learning algorithms (MLAs) can accurately predict which hospitalized patients will become infected with Clostridioides difficile (CDI), allowing for increased monitoring, timely implementation of appropriate infection-control practices and treatment earlier in the disease course.

“Healthcare-associated infections are a great target for machine learning and data analysis, because there is all of this patient data that we have now and it’s just not being leveraged or processed and presented to physicians in a way that can really help them make better decisions and anticipate which patients are most at risk,” Dr. Jana Hoffman, vice president of science for Dascena, Inc, told Reuters Health by phone.

Currently, there is no gold standard clinical risk assessment tool to predict CDI in hospitalized patients.

Dr. Hoffman and her colleagues used real-world electronic health record (EHR) data from over 700 U.S. hospitals to train and evaluate three different MLAs to predict CDI.

The development dataset comprised more than 13 million inpatient encounters with 80,046 CDI events and the external dataset comprised 1,149,088 inpatient encounters with 7,107 CDI encounters.

MLAs can predict CDI with “excellent discrimination” at any point during an inpatient stay based on the first six hours of hospitalization data, the Houston, Texas-based researchers report in the American Journal of Infection Control.

The highest performing MLA in terms of AUROC values was the XGBoost model (AUROC, 0.815).

Many of the key features used by the algorithms to predict CDI were similar across MLAs, and have previously been identified as risk factors for CDI.

Age was the leading CDI risk factor, followed by clinical measurements such as sodium, BMI, white blood cell count, and heart rate; active treatment with antibiotics or proton-pump inhibitors; glycated hemoglobin; and race.

This MLA approach “doesn’t require any additional effort on the behalf of the doctors who are already overwhelmed,” Dr. Hoffman told Reuters Health.

The algorithm “automatically monitors and analyzes all the data coming in through the electronic health system without any human intervention required and provides alerts for patients that are most at risk,” she said.

Looking ahead, the researchers hope to validate MLA-based CDI prediction tools on prospectively collected, live data, and get feedback from clinicians to optimize the usefulness and acceptability of MLA alerts of CDI risk.

“We’re definitely interested in deploying (this tool) and see how it performs in a prospective setting and we’re exploring conversations with different healthcare partners,” Dr. Hoffman told Reuters Health.

“There have been good strides made with C. difficile,” Linda Dickey, president of the Association for Professionals in Infection Control (APIC) told Reuters Health by phone.

“But being able to predict who is at risk using a tool like this would be really helpful because then you could maybe tailor some of the approaches for patients that are at higher risk. This a very elegant tool to help identify some of those patients,” Dickey said.

“There is already this type of algorithmic learning that has helped us see patients, for example, that are at higher risk for readmission, helping us focus on good discharge planning and that type of thing,” she added.

The study authors are Dascena employees or contractors.

SOURCE: American Journal of Infection Control, online January 19, 2022.

Source: Read Full Article