Home Medizin Evaluierung eines maschinellen Lerntools zur Vorhersage einer im Krankenhaus erworbenen akuten Nierenschädigung

Evaluierung eines maschinellen Lerntools zur Vorhersage einer im Krankenhaus erworbenen akuten Nierenschädigung

von NFI Redaktion


A hospital-acquired acute kidney injury (HA-AKI) is a common complication among hospitalized patients that can lead to chronic kidney disease and is associated with longer hospital stays, higher health care costs, and increased mortality. Preventing HA-AKI can improve treatment outcomes for patients while predicting its onset can be challenging due to various contributing factors.

Researchers at Mass General Brigham Digital tested a commercial machine learning tool, the Epic Risk of HA-AKI prediction model, and found it to be moderately successful in predicting the risk of HA-AKI in recorded patient data. The study revealed lower performance compared to internal validation by Epic Systems Corporation, emphasizing the importance of validating AI models before clinical implementation.

The Epic model assesses adult inpatient encounters for the risk of HA-AKI characterized by a predefined increase in serum creatinine levels. After training the model on MGB hospital data, the researchers tested it with data from nearly 40,000 inpatient hospitalizations over a five-month period between August 2022 and January 2023. This extensive dataset included various patient encounter information, such as patient demographics, comorbidities, primary diagnoses, serum creatinine levels, and length of hospital stay. Two analyses were conducted to assess model performance at the encounter and prediction levels.

The researchers found that the tool was more reliable in assessing patients with lower HA-AKI risk. While the model could accurately identify low-risk patients who would not develop HA-AKI, it struggled to predict when higher-risk patients might develop HA-AKI. The results also varied depending on the evaluated HA-AKI stage – predictions were more successful for Stage 1 HA-AKI than for severe cases.

The authors concluded that the implementation could lead to high false positive rates and called for further investigation into the clinical impact of the tool.

Sayon Dutta, MD, MPH, lead study author of the clinical informatics team at Mass General Brigham Digital and an emergency physician at Massachusetts General Hospital stated, „We found that the Epic prediction model could better exclude low-risk patients than high-risk patients. Identifying HA-AKI risk with predictive models could help support clinical decisions, such as warning providers before ordering nephrotoxic medications. However, further studies are needed before clinical implementation.“

Source:

Journal reference:

Dutta, S., et al. (2024). External validation of a commercial prediction model for acute kidney injuries. NEJM AI. doi.org/10.1056/aioa2300099.

Related Posts

Adblock Detected

Please support us by disabling your AdBlocker extension from your browsers for our website.