Overview
During the COVID-19 pandemic, hospitals faced significant challenges with ICU bed shortages. This issue has persisted due to an aging population, with approximately 11% of hospital admissions involving ICU stays.
AI as a Solution
According to Indranil Bardhan, a professor at Texas McCombs, artificial intelligence (AI) can help address these challenges. AI models are capable of predicting the duration of patient stays in the ICU, which can assist hospitals in managing their resources more effectively and potentially reduce costs.
Understanding AI Predictions
While AI excels at predicting lengths of stay, it often lacks the ability to explain its predictions, leading to skepticism among healthcare professionals. Bardhan emphasizes the importance of explainable artificial intelligence (XAI) to enhance trust and usability among ICU doctors.
Research Findings
In collaboration with doctoral student Tianjian Guo and other researchers, Bardhan developed a model trained on a dataset of 22,243 medical records from 2001 to 2012. The model analyzes 47 attributes of patients upon admission, such as:
- Age
- Gender
- Vital signs
- Medications
- Diagnosis
This model generates graphs indicating the probability of patient discharge within seven days and highlights the factors influencing these outcomes.
Practical Application
In testing, six ICU physicians evaluated the model’s explanations. Four out of six agreed that the model could enhance staffing and resource management, aiding in patient scheduling.
Limitations and Future Directions
One significant limitation of the model is the age of the data, as the medical coding system transitioned from ICD-9-CM to ICD-10-CM in 2014, which introduced more detailed coding. Bardhan expressed a desire to update the model with more recent data.
Moreover, the model’s application is not restricted to adult ICUs; it can also be adapted for pediatric and neonatal ICUs, as well as emergency room settings.
Conclusion
By making AI outputs more understandable, hospitals can optimize ICU bed management, reduce costs, and improve patient care.