A deep learning tool developed at Tulane University outperformed five algorithms designed to predict osteoporosis risk, a recently published study found. The cutting edge of artificial intelligence.
Osteoporosis is characterized by a decrease in bone mineral density (BMD) and deterioration of bone tissue. Silent Illness Many people with osteoporosis never experience symptoms, but the weakened bones associated with osteoporosis are a leading cause of fractures and disability, especially in older adults.
To mitigate these adverse outcomes, we need to accurately stratify patient risk and effectively diagnose at-risk individuals.However, early diagnosis of osteoporosis remains a major public health challenge because dual-energy X-ray absorptiometry, the gold standard for diagnosing osteoporosis, is expensive and inaccessible to most people.
To combat this, researchers have been working to build algorithms that use routinely collected demographic and clinical data to predict osteoporosis risk. Currently, many of these models have been shown to be inaccurate, and a team at Tulane University is evaluating whether deep neural networks (DNNs) can improve predictive performance.
The model was built by combining extensive demographic, clinical, and BMD data from 8,134 people aged 40 and over drawn from the Louisiana Osteoporosis Study. From there, the DNN was compared to four machine learning approaches (random forest (RF), artificial neural networks, k-nearest neighbors, and support vector machines) and one regression model (known as the Osteoporosis Self-Assessment Tool) to analyze 16 variables.
The performance of each tool in predicting osteoporosis risk was assessed based on the area under the curve (AUC) and various accuracy indices.
DNN was more effective at classifying osteoporosis compared to other approaches, reaching an AUC of 0.848, sensitivity of 0.740, and specificity of 0.793. The prediction accuracy of DNN was 0.753, surpassed only by the RF model at 0.757.
Furthermore, the analysis revealed 10 significant factors predicting osteoporosis risk: weight, age, sex, grip strength, height, beer intake, diastolic blood pressure, alcohol consumption, years of smoking, and income.
These findings indicate that the DNN model performed best overall predictive performance, providing an opportunity to facilitate early diagnosis of osteoporosis.
“The earlier osteoporosis risk is detected, the more time patients have to take preventive measures,” lead author Chuan Qiu, M.D., Ph.D., a research assistant professor in the Center for Biomedical Informatics and Genomics at the Tulane University School of Medicine, said in a press release. “We are pleased to see that our DNN model outperformed other models in accurately predicting osteoporosis risk in an aging population.”
Despite the success of their model, the researchers cautioned that further work to refine and validate the AI-based risk prediction tool is needed before it can be deployed in clinical settings to stratify the risk of individual patients.
“Our ultimate goal is for people to be able to enter their information, receive a highly accurate osteoporosis risk score, and then receive treatment to strengthen their bones and reduce further damage,” Qiu said.
Shania Kennedy has been covering healthcare IT and analytics related news since 2022.
