Use of Predictive Models at the Population Level Has the Potential to Save Cost by Directing Economic Resources to Those Likely to Improve Most: Spinal Deformity Simulation Analysis Stratified by Risk

Introduction: Healthcare costs in first-world economies are continuing to accelerate at an unsustainable rate and present significant economic burden, especially for management of complex disease such as adult spinal deformity. Machine learning models have the ability to preoperatively predict which patients may benefit most from surgery prior to incurring the expense of their care. By using robust machine learning models to identify optimal surgical candidates, resources can be directed towards patients most likely to benefit from surgical correction before incurring costs.

Methods: A prospective cohort of patients treated at 17 ASD specialty centers in US and Europe from 2008-2016 was queried. Clinical outcomes including minimum clinically important difference (MCID), complication, and reoperation rates were predicted using gradient boosting classification. Thresholds were applied to simulate patients who met specific outcomes criteria for surgical utilization and applied to public health data from the US and Spain.

Results: A cohort of 1,245 patients with 195 variables was used to train a predictive model (complications AUC = 0.68, MCID AUC = 0.70). Example criteria of < 20% expected complication rate and >50% chance of MCID corresponded to 33% of existing patients qualifying for surgery. Using $120,000 as average hospital cost for ASD correction, this utilization rate translated to total hospital savings of $541 million in the US (year 2013 data) in direct cost. In Spain (year 2015 data), only 244 patients out of 740 would have received surgery, reducing surgery rate per 100,000 adults from 1.64 to 0.54.

Conclusion : Accurate prognostic models that predict outcomes for ASD patients can be used to guide clinical decision-making by preoperatively identifying patients who would benefit most from surgery. While complication/outcome profiles should not solely determine if patients warrant or may benefit from surgery, optimized ASD candidate selection may reduce societal costs and maximize outcomes especially in countries with cost-constrained healthcare systems.