Using Artificial Intelligence to Predict Residual Distal Lumbosacral Pain Post Percutaneous Kyphoplasty for Osteoporotic Vertebral Compression Fractures.

BACKGROUND

Percutaneous kyphoplasty (PKP) can restore spinal stability and relieve pain in patients with osteoporotic vertebral compression fractures (OVCF). However, in some cases, distal lumbosacral pain (DLP) persists postoperatively, affecting patients' expectations of the surgery and their recovery to activities of daily life.

OBJECTIVE

To use artificial intelligence to predict DLP post-PKP for OVCF, thereby providing personalized treatment plans for patients with OVCF.

STUDY

DESIGN

Retrospective study.

SETTING

The study was carried out at a university hospital.

METHODS

A univariate analysis was performed to identify the risk factors for DLP post-PKP. A heatmap analysis was conducted to examine the relationships between variables in the dataset. A random forest model was established, and its performance was evaluated using a confusion matrix. After validating and tuning the model, features were ranked based on their contribution to prediction accuracy.

RESULTS

A total of 179 patients completed this study. Patients were divided into 2 groups (Group 0 without DLP; Group 1 with DLP). The univariate analysis indicated statistically significant differences in terms of bone density, intravertebral vacuum cleft, sarcopenia, bone cement distribution, interspinous ligament degeneration, and Hounsfield unit (P < 0.05). The heatmap analysis revealed a moderate correlation between DLP and both sarcopenia and interspinous ligament degeneration. A random forest model was built. The confusion matrix showed that the model exhibited strong performance across all metrics. The random forest model showed that the preoperative Cobb angle and sarcopenia were the most critical features.

LIMITATIONS: This was a retrospective study, which may be prone to selection and recall bias. Single-center noncontrolled studies may also introduce bias.

CONCLUSION

Our random forest model can effectively predict DLP post-PKP for OVCF, assisting in the selection of treatment plans.