A radiologist who interprets magnetic resonance imaging. Images by an editor of Medical Futurist. – The future of radiology and artificial intelligence. Medical Futurist (2017-06-29) CC4.0
According to a multi-hospital study by Mount Sinai Health System, artificial intelligence (AI) helps emergency department (ED) teams better predict patients who need to be hospitalized earlier than they can now. This illustrates the strength of the predictive models in this field.
By providing clinicians with advance notice, this approach can enhance patient care and patient experience, reducing overcrowding and “boarding” (if patients are recognized but remain in ED because beds are not available).
This is the biggest prospective assessment of AI in an emergency environment to date, and the study has been published in Mayo Clinic Proceedings: Digital Healthentitled “Comparison of machine learning and nurse predictions for hospitalization in a multisite emergency health system.”
For this study, the researchers collaborated with over 500 ED nurses across the seven hospital health system. Together, they evaluated machine learning models trained with data from the past over 1 million patient visits. Over two months, we compared AI generation predictions with nurse triage assessments to see whether AI could identify potential hospitalizations immediately after the patient arrives.
This study showed that AI models were reliably implemented in these diverse hospital settings, with nearly 50,000 patient visits across urban and suburban hospitals in Mount Sinai. Surprisingly, researchers show that combining human and machine predictions indicates that AI systems are the only powerful predictors.
By training the algorithm with over 1 million patient visits, researchers aimed to capture meaningful patterns that would help predict admissions faster than traditional methods.
The output was that machine learning-based predictions outperformed the estimated hospital triage nurses, indicating high sensitivity and specificity for hospitalization predictions. These findings suggest that admission prediction systems fixed by machine learning can be reliably executed using data available in triage.
Although the study was limited to one healthcare system for two months, researchers hope that the findings will serve as a springboard for future live clinical testing. The next step is to implement AI models in real-time workflows to measure outcomes such as reduced boarding times, improved patient flow, and operational efficiency.
This can trigger a new set of systems that allow AI to make complex predictions. This study also has immediate practical benefits. By predicting admission early, results can give the care team the time they need to plan, coordinate and ultimately provide better, more compassionate care. It's exciting to see AI appear not as a futuristic idea, but as a practical, real-world solution shaped by people who provide care every day.
