External validation of a predictive model of adverse events following spine surgery

• Published in: The spine journal Vol. 22; no. 1; pp. 104 - 112
• Main Authors: Fatemi, Parastou, Zhang, Yi, Han, Summer S., Purington, Natasha, Zygourakis, Corinna C., Veeravagu, Anand, Desai, Atman, Park, Jon, Shuer, Lawrence M., Ratliff, John K.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.01.2022
• Subjects: Aged; Humans; Machine learning; Medicare; Postoperative adverse events; Postoperative Complications - epidemiology; Postoperative Complications - etiology; Predictive model; Prospective Studies; Risk Assessment; Screening tool; Spine - surgery; United States; United States; Screening tool; Postoperative adverse events; Machine learning; Predictive model; Spine surgery
• ISSN: 1529-9430; 1878-1632
• DOI: 10.1016/j.spinee.2021.06.006

We lack models that reliably predict 30-day postoperative adverse events (AEs) following spine surgery. We externally validated a previously developed predictive model for common 30-day adverse events (AEs) after spine surgery. This prospective cohort study utilizes inpatient and outpatient data from a tertiary academic medical center. We assessed a prospective cohort of all 276 adult patients undergoing spine surgery in the Department of Neurosurgery at a tertiary academic institution between April 1, 2018 and October 31, 2018. No exclusion criteria were applied. Incidence of observed AEs was compared with predicted incidence of AEs. Fifteen assessed AEs included: pulmonary complications, congestive heart failure, neurological complications, pneumonia, cardiac dysrhythmia, renal failure, myocardial infarction, wound infection, pulmonary embolus, deep venous thrombosis, wound hematoma, other wound complication, urinary tract infection, delirium, and other infection. Our group previously developed the Risk Assessment Tool for Adverse Events after Spine Surgery (RAT-Spine), a predictive model of AEs within 30 days following spine surgery using a cohort of approximately one million patients from combined Medicare and MarketScan databases. We applied RAT-Spine to the single academic institution prospective cohort by entering each patient's preoperative medical and demographic characteristics and surgical type. The model generated a patient-specific overall risk score ranging from 0 to 1 representing the probability of occurrence of any AE. The predicted risks are presented as absolute percent risk and divided into low (<17%), medium (17%-28%), and high (>28%). Among the 276 patients followed prospectively, 76 experienced at least one 30-day postoperative AE. Slightly more than half of the cohort were women (53.3%). The median age was slightly lower in the non-AE cohort (63 vs. 66.5 years old). Patients with Medicaid comprised 2.5% of the non-AE cohort and 6.6% of the AE cohort. Spinal fusion was performed in 59.1% of cases, which was comparable across cohorts. There was good agreement between the predicted AE and observed AE rates, Area Under the Curve (AUC) 0.64 (95% CI 0.56-0.710). The incidence of observed AEs in the prospective cohort was 17.8% among the low-risk group, 23.0% in the medium-risk group, and 38.4% in the high risk group (p =.003). We externally validated a model for postoperative AEs following spine surgery (RAT-Spine). The results are presented as low-, moderate-, and high-risk designations.