Machine learning models are being developed to help physicians in intensive care units by alerting them to rapidly deteriorating patient conditions. However, a recent study from Virginia Tech published in Communications Medicine has revealed that these models are failing to detect key health deteriorations, with in-hospital mortality prediction models missing 66% of critical injuries.
Lead researcher Danfeng “Daphne” Yao, along with Ph.D. student Tanmoy Sarkar Pias, collaborated with other researchers to evaluate the responsiveness of machine learning models to critical or deteriorating health conditions. The study found that patient data alone is not sufficient to train these models effectively. By calibrating the models with test patients, the researchers were able to uncover the limitations of the current models.
Using innovative medical testing approaches such as the gradient ascent method and neural activation map, the team assessed the ability of machine learning models to respond to serious medical conditions. These methods helped identify deficiencies in the responsiveness of models for in-hospital mortality prediction and five-year breast and lung cancer prognosis.
The study emphasizes the need to incorporate medical knowledge into clinical machine learning models to improve their accuracy and effectiveness. Yao’s team is actively testing other medical models, including large language models, to ensure their safety and efficacy in time-sensitive clinical tasks like sepsis detection.
In the rapidly evolving field of AI and healthcare, transparent and objective testing of machine learning models is crucial to protect patients’ lives. Yao’s group is committed to conducting rigorous testing to ensure the safety and reliability of AI-powered medical products.
For more information on the study, “Low Responsiveness of Machine Learning Models to Critical or Deteriorating Health Conditions,” published in Communications Medicine, visit the DOI link provided. This research highlights the importance of enhancing the predictive capabilities of machine learning models in healthcare settings and the need for interdisciplinary collaboration between computing and medical experts.
