In The spine journal : official journal of the North American Spine Society
BACKGROUND CONTEXT : Reimbursement has slowly transitioned from a Fee-for-Service model to a bundled payment model after introduction of the United States Centers for Medicare and Medicaid Services bundled payment program. To minimize healthcare costs, some surgeons are trying to minimize healthcare expenditures by transitioning appropriately selected lumbar decompression patients to outpatient procedure centers.
PURPOSE : To prepare a risk stratification calculator based on machine learning algorithms to improve surgeon's preoperative predictive capability of determining whether a patient undergoing lumbar decompression will meet inpatient versus outpatient criteria. Inpatient criteria was defined as any overnight hospital stay.
STUDY DESIGN/SETTING : Retrospective single-institution cohort PATIENT SAMPLE: 1,656 patients undergoing primary lumbar decompression OUTCOME MEASURES: Postoperative outcomes analyzed for inclusion into the risk calculator included length of stay.
METHODS : Patients were split 80-20 into a training model and a predictive model. This resulted in 1,325 patients in the training model and 331 into the predictive model. A logistic regression analysis ensured proper variable inclusion into the model. C-statistics were used to understand model effectiveness. An odds ratio and nomogram were created once the optimal model was identified.
RESULTS : A total of 1,656 patients were included in our cohort with 1,078 dischared on day of surgery and 578 patients spending ≥ 1 midnight in the hospital. Our model determined older patients (OR = 1.06, p<0.001) with a higher BMI (OR = 1.04, p<0.001), higher back pain (OR = 1.06, p=0.019), increasing ASA score (OR = 1.39, p=0.012), and patients with more levels decompressed (OR = 3.66, p<0.001) all had increased risks of staying overnight. Patients who were female (OR = 0.59, p=0.009) and those with private insurance (OR = 0.64, p=0.023) were less likely to be admitted overnight. Further, weighted scores based on training data were then created and patients with a cumulative score over 118 points had a 82.9% likelihood of overnight. Analysis of the 331 patients in the test data demonstrated using a cut-off of 118 points accurately predicted 64.8% of patients meeting inpatient criteria compared to 23.0% meeting outpatient criteria (p<0.001). Area under the curve analysis showed a score greater than 118 predicted admission 81.4% of the time. The algorithm was incorporated into an open access digital application available here: https://rothmanstatisticscalculators.shinyapps.io/Inpatient_Calculator/?_ga=2.171493472.1789252330.1671633274-469992803.1671633274 CONCLUSIONS: Utilizing machine-learning algorithms we created a highly reliable predictive calculator to determine if patients undergoing outpatient lumbar decompression would require admission. Patients who were younger, had lower BMI, lower pre-operative back pain, lower ASA score, less levels decompressed, private insurance, lived with someone at home, and with minimal comorbidities were ideal candidates for outpatient surgery.
Canseco Jose A, Karamian Brain A, Lambrechts Mark J, Issa Tariq Z, Conaway William, Minetos Paul D, Bowles Daniel, Alexander Tyler, Sherman Matthew, Schroeder Gregory D, Hilibrand Alan S, Vaccaro Alexander R, Kepler Christopher K
2023-Jan-12
inpatient stay, lumbar decompression, machine learning, outpatient procedure, predictive modeling
