Poor predictive performance of early outcome models following lumbar disc herniation surgery

Abstract

Aims This study aimed to assess the preoperative predictors of individual outcomes after surgery for lumbar disc herniation (LDH). Methods We conducted a retrospective cohort study using patients treated from 1 January 2009 to 31 December 2019 at a single hospital. We included patients aged between 18 and 65 years who underwent surgery for the first time for a specific level of herniation, where surgery consisted of lumbar discectomy or sequestrectomy. Fusion or central decompression procedures were excluded. Data were collected from electronic patient registries and paper questionnaires, when available. The study outcomes included postoperative emergency department visits, readmission, length of stay, postoperative Oswestry Disability Index (ODI) score after four weeks, additional surgery due to reherniation within 60 days, and postoperative infection. Logistic and ordinal regression analyses were used to develop a predictive model. Results The data included 882 patients, with a mean age of 41 years, and 56% of whom were male. The model for the length of stay showed the best performance parameters with a moderate fit (pseudo-R2 0.134, area under the curve (AUC) 0.682) if BMI as a variable was omitted. The results with BMI as a variable were similar (pseudo-R2 0.144, AUC 0.681). Emergency operation (χ = 17.4, p < 0.001) and surgery on multiple intervertebral discs (χ = 8.3, p = 0.001, or for multiple levels: 9.29, 95% CI 1.96 to 44.02) were the most efficient variables in prediction in these models, respectively. The model for postoperative ODI score showed similar performance (pseudo-R2 0.170, AUC 0.651); the most important variable was sex (χ = 6.5, p = 0.011). Conclusion Length of stay in hospital after LDH surgery was best explained by emergency surgery and surgery on multiple intervertebral discs. The models created were moderate at best, and future research with larger datasets is needed.