Development of a preoperative predictive model for major complications following adult spinal deformity surgery

Abstract

<jats:sec><jats:title>OBJECTIVE</jats:title><jats:p>The operative management of patients with adult spinal deformity (ASD) has a high complication rate and it remains unknown whether baseline patient characteristics and surgical variables can predict early complications (intraoperative and perioperative [within 6 weeks]). The development of an accurate preoperative predictive model can aid in patient counseling, shared decision making, and improved surgical planning. The purpose of this study was to develop a model based on baseline demographic, radiographic, and surgical factors that can predict if patients will sustain an intraoperative or perioperative major complication.</jats:p></jats:sec><jats:sec><jats:title>METHODS</jats:title><jats:p>This study was a retrospective analysis of a prospective, multicenter ASD database. The inclusion criteria were age ≥ 18 years and the presence of ASD. In total, 45 variables were used in the initial training of the model including demographic data, comorbidities, modifiable surgical variables, baseline health-related quality of life, and coronal and sagittal radiographic parameters. Patients were grouped as either having at least 1 major intraoperative or perioperative complication (COMP group) or not (NOCOMP group). An ensemble of decision trees was constructed utilizing the C5.0 algorithm with 5 different bootstrapped models. Internal validation was accomplished via a 70/30 data split for training and testing each model, respectively. Overall accuracy, the area under the receiver operating characteristic (AUROC) curve, and predictor importance were calculated.</jats:p></jats:sec><jats:sec><jats:title>RESULTS</jats:title><jats:p>Five hundred fifty-seven patients were included: 409 (73.4%) in the NOCOMP group, and 148 (26.6%) in the COMP group. The overall model accuracy was 87.6% correct with an AUROC curve of 0.89 indicating a very good model fit. Twenty variables were determined to be the top predictors (importance ≥ 0.90 as determined by the model) and included (in decreasing importance): age, leg pain, Oswestry Disability Index, number of decompression levels, number of interbody fusion levels, Physical Component Summary of the SF-36, Scoliosis Research Society (SRS)–Schwab coronal curve type, Charlson Comorbidity Index, SRS activity, T-1 pelvic angle, American Society of Anesthesiologists grade, presence of osteoporosis, pelvic tilt, sagittal vertical axis, primary versus revision surgery, SRS pain, SRS total, use of bone morphogenetic protein, use of iliac crest graft, and pelvic incidence–lumbar lordosis mismatch.</jats:p></jats:sec><jats:sec><jats:title>CONCLUSIONS</jats:title><jats:p>A successful model (87% accuracy, 0.89 AUROC curve) was built predicting major intraoperative or perioperative complications following ASD surgery. This model can provide the foundation toward improved education and point-of-care decision making for patients undergoing ASD surgery.</jats:p></jats:sec>

Artificial Intelligence for Adult Spinal Deformity

“…Both surgeons and patients alike are interested in better tools for predicting outcomes, as it allows for more comprehensive preoperative discussions and surgical decision making. Predictive analytics has now been applied across a wide variety of topics within ASD surgery, including predicting intraoperative, 15 perioperative, 16,17 and postoperative complications and outcomes. [18][19][20][21][22][23][24][25] The majority of studies published on this topic share similar principles and methodologies in the development of their respective predictive models.…”

“…Models have also been created to assess length of stay (LOS) 16 and major early complications. 17 In the study by Scheer et al 17 a predictive model for early complications (intraoperative and within 6 weeks postoperative) was created using 45 variables from baseline demographic, radiographic, and surgical factors for 557 ASD patients. An ensemble of decision trees was trained with 5 different bootstrapped models and internal validation was accomplished using a 70:30 data split.…”

Potential of predictive computer models for preoperative patient selection to enhance overall quality-adjusted life years gained at 2-year follow-up: a simulation in 234 patients with adult spinal deformity

“…23 Predictive modeling is a statistical technique that represents a departure from the more traditional regression analyses; however, it affords greater flexibility by allowing patterns within the available data to create accurate, patient-specific predictions without the need to establish hypotheses or control groups a priori. 1 Through this established methodology, we have successfully developed models for accurate prediction of postoperative complications, 30 proximal junction failure and proximal junction kyphosis, 28 and length of hospital stay. 26 In the present study, the same modeling techniques were used in a large cohort of surgically treated patients with ASD to predict which patients will be likely to have better outcomes, as defined by meeting MCID thresholds for ODI and QALY improvements.…”

Applications of Machine Learning Using Electronic Medical Records in Spine Surgery

“…In fact, multiple studies have demonstrated the ability to build predictive models using EMR data for major perioperative complications in spine surgery, particularly surgical site infections. [53][54][55][56][57] Other models have also been created using EMR data to predict physical disability, return to work, major complications, readmission rates, walking ability, need for inpatient rehab following spine surgery, discharge, and disposition. [58][59][60][61] More specific algorithms have been created to predict preoperative factors impacting survival, discharge, and readmission rates in patients following spine surgery for spinal metastasis.…”