Research Scientist Ascension Texas Spine and Scoliosis
Introduction: Using machine learning identify factors associated with the likelihood of a patient will choose surgical intervention for lumbar disc herniation.
Methods: 300 retrospective charts were reviewed. Clinical and demographic data were collected. Models were optimized by assessing multicollinearity of independent variables and employing Area Under the Curve/Receiver Operating Characteristic (AUC-ROC) and Precision-Recall techniques.
Results: Overall surgical rate was 25.3%, with median time to surgery eight weeks from intake. Men chose surgery at a higher rate than women (29% vs 21.2%) Comorbidities and clinical presentation were not found to have an effect on odds of surgical intervention. Median patient reported back/leg pain and disability scores were higher in patients that did have surgery (38 vs 34 and 7 vs 6, respectively, p>0.05). The final optimized model yielded a test recall of 0.4. Only three variables were found to impact the odds of surgical intervention. Patients who received a prescription for non-narcotic medication and had at least one epidural injection, increased the odds of surgical intervention by 3.4 and 2.6, respectively. Prescribing opioids decreased odds of surgical intervention by 0.37.
Conclusion : Non-narcotic medication and injections increased the odds of later surgical intervention. Given the current standard of care, this is not surprising as one should exhaust all non-operative measures before culminating in surgical intervention. Prescribing opioid medications during this minimal six-week time frame lowered the odds of surgical intervention. This could be due to the fact that opioids were prescribed to patients that had comorbidities or other factors that do not lend themselves to the preferred physical rehabilitation plan of care. These patients would already be unlikely surgical candidates due to increased risk associated with medical comorbidities.
How to Improve Patient Care: A larger, prospective study would be the next step in utilizing machine learning to predict which patients are most likely to report satisfaction and a successful outcome.
