Uses a Dual-Model Machine Learning Model to Predict Urinary Retention Following Spinal Surgery
This stacked machine learning algorithm accurately predicts post-operative urinary retention (POUR) following lumbar spine surgery by analyzing patient medical history and surgical details. In the United States, more than 1.62 million spine surgeries occur annually. Post-operative urinary retention (POUR) is a common complication affecting 6%-38% of patients who undergo elective spine surgery. POUR significantly increases hospital length of stay and impedes functional recovery. POUR increases hospitalization costs by $3565 to $4100. Currently, there are no reliable methods to identify patients with an increased risk for POUR.
Researchers at the University of Florida developed this machine learning model to support surgeons and care teams in preventing POUR. The model integrates logistic regression and a neural network into a logical stacking framework, enhancing diagnostic accuracy without requiring additional patient testing. The logical stacking system, using strict or loose cutoff criteria depending on desired diagnostic sensitivity or specificity, allows clinicians to tailor predictions to their clinical context. Unlike traditional stacking methods that average model outputs, this approach applies a logical filter to determine outcomes, ensuring higher flexibility and performance in real-world settings. By leveraging the stacking mechanism and a thorough analysis of patient-specific factors, this approach provides predictive risk analysis in perioperative care.
Application
This stacked machine learning model provides personalized risk assessments for post-operative urinary retention, enabling proactive patient management and improved surgical outcomes
Advantages
- Combines logistic regression and neural networks, enhancing prediction accuracy for post-operative urinary retention
- Uses a logical stacking method, offering customizable sensitivity and specificity based on clinical needs
- Integrates seamlessly with patient data, providing real-time, personalized risk assessments during surgical planning
Technology
This technology uses a stacked machine learning algorithm to predict post-operative urinary retention (POUR) following lumbar spine surgery. It combines logistic regression and a neural network to assess a patient's risk based on medical history and surgical factors. The logical stacking method evaluates the models' predictions using optimized threshold values. Upon inputting clinical data, each model generates a probability that the patient will experience POUR. These outputs are then evaluated using optimized threshold cutoffs to make the final prediction. This allows clinicians to tailor predictions to their preferred balance of sensitivity and specificity. By identifying high-risk patients early, the model supports proactive interventions and improves surgical outcomes. The tool is designed for integration into hospital systems, making it practical for real-time clinical use.
Brochure
