SPINE-RISK VE: Multimodal Predictive Model for Failed Back Surgery Syndrome in Venezuelan Surgical Patients

Overview

SPINE-RISK VE is a prospective multicenter cohort study designed to develop and internally validate a multimodal preoperative predictive model for Failed Back Surgery Syndrome (FBSS), now classified as Persistent Spinal Pain Syndrome Type 2 (PSPS-T2) per ICD-11 (code MG30.51), in Venezuelan adults patients undergoing elective lumbar spine surgery. The model integrates three variable domains obtainable from routine preoperative evaluation at zero additional cost to the patient: (1) inflammatory laboratory biomarkers (C-reactive protein \[CRP\], neutrophil-to-lymphocyte ratio \[NLR\], albumin, glycated hemoglobin \[HbA1c\], erythrocyte sedimentation rate \[ESR\]); (2) preoperative lumbar magnetic resonance imaging (MRI) findings (Modic changes, Pfirrmann disc degeneration grade, foraminal stenosis, number of surgical levels, spondylolisthesis); and (3) validated psychosocial instruments (Patient Health Questionnaire-9 \[PHQ-9\], Pain Catastrophizing Scale \[PCS\], smoking status, benzodiazepine use, prior lumbar surgery). Analysis proceeds in two phases: Phase 1 applies multivariable logistic regression with Least Absolute Shrinkage and Selection Operator (LASSO) variable selection to generate a printable clinical nomogram; Phase 2 applies a random forest machine learning algorithm with 10-fold cross-validation. Model reporting follows Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis plus Artificial Intelligence (TRIPOD+AI) guidelines. SPINE-RISK VE aims to produce the first validated multimodal predictive model for PSPS-T2/FBSS was developed in a Latin American surgical cohort, providing neurosurgeons with an evidence-based preoperative risk stratification tool applicable without Additional technological infrastructure.

Failed Back Surgery Syndrome (FBSS), formally reclassified as Persistent Spinal Pain Syndrome Type 2 (PSPS-T2) in ICD-11 (code MG30.51), affects 10-40% of patients undergoing lumbar spine surgery and constitutes one of the most complex therapeutic challenges in contemporary neurosurgery. Despite the identification of individual risk factors in the literature, no validated multimodal predictive model integrating laboratory biomarkers, lumbar magnetic resonance imaging (MRI) morphology, and psychosocial variables exist for Latin American surgical populations. The best available predictive model to date achieved Area Under the Receiver Operating Characteristic Curve (AUC) of 0.715 for decompression and 0.701 for fusion using only electronic health record variables, without laboratory biomarkers or MRI-derived predictors, and without validation in any Latin American cohort. SPINE-RISK VE addresses this gap through a prospective multicenter cohort design enrolling 100-150 adults with Elective lumbar surgery indication at three Venezuelan referral centers. PREDICTOR DOMAINS: Domain 1 - Inflammatory biomarkers: C-reactive protein (CRP greater than 3 mg/L), neutrophil-to-lymphocyte ratio (NLR greater than 3.0), serum albumin (less than 3.5 g/dL), glycated hemoglobin (HbA1c greater than 7%), and erythrocyte sedimentation rate (ESR). All obtainable from standard preoperative Laboratory panels. Domain 2 - Lumbar MRI findings: Modic changes (Types I-III), disc degeneration grade (Pfirrmann scale I-V), foraminal stenosis, number of surgical levels, and Spondylolisthesis grade (Meyerding I-IV). All from already-requested preoperative imaging. Domain 3 - Psychosocial factors: depression (Patient Health Questionnaire-9 \[PHQ-9\] cutoff of 10 or greater), pain catastrophizing (Pain Catastrophizing Scale \[PCS\] cutoff of 30 or greater, active smoking, preoperative benzodiazepine use, and prior lumbar surgery history. PRIMARY OUTCOME: PSPS-T2/FBSS incidence at 12 months, defined as the Numeric Rating Scale (NRS) of 4 or greater AND Oswestry Disability Index (ODI) of 40% or greater at postoperative follow-up, consistent with International Association for the Study of Pain (IASP) criteria. ANALYTICAL PLAN: Phase 1: Multivariable logistic regression with Least Absolute Shrinkage and Selection Operator (LASSO) regularization to identify independent predictors and generate a Printable clinical nomogram. Software: R Version 4.x (glmnet, rms packages). Phase 2: Random forest (500 trees, 10-fold cross-validation) compared against Extreme Gradient Boosting (XGBoost) and logistic regression. Performance metrics: AUC-ROC (target 0.80 or greater), sensitivity, specificity, calibration (Hosmer-Lemeshow test, Brier score). Model interpretability via SHapley Additive exPlanations (SHAP) values. REPORTING: Transparent Reporting of a multivariable prediction model for an individual Prognosis Or Diagnosis plus Artificial Intelligence (TRIPOD+AI) 2024 and Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. EXPECTED OUTPUTS: (1) Printable preoperative nomogram applicable without additional technological infrastructure; (2) exportable machine learning (ML) model with AUC target of 0.80 or greater; (3) first structured lumbar surgery database with 12-month Follow-up generated in Venezuela.