A predictive model for outcome after conservative decompression surgery for lumbar spinal stenosis

Abstract

This study was designed to develop predictive models for surgical outcome based on information available prior to lumbar stenosis surgery. Forty patients underwent decompressive laminarthrectomy. Preop and 1-year postop evaluation included Waddell's nonorganic signs, CT scan, Waddell disability index, Oswestry low back pain disability questionnaire, low back outcome score (LBOS), visual analog scale (VAS) for pain intensity, and trunk strength testing. Statistical comparisons of data used adjusted error rates within families of predictors. Mathematical models were developed to predict outcome success using stepwise logistic regression and decision-tree methodologies (chi-squared automatic interaction detection, or CHAID). Successful outcome was defined as improvement in at least three of four criteria: VAS, LBOS, and reductions in claudication and leg pain. Exact logistic regression analysis resulted in a three-predictor model. This model was more accurate in predicting unsuccessful outcome (negative predictive value 75.0%) than in successful outcome (positive predictive value 69.6%). A CHAID model correctly classified 90.1% of successful outcomes (positive predictive value 85.7%, negative predictive value 100%). The use of conservative surgical decompression for lumbar stenosis can be recommended, as it demonstrated a success rate similar to that of more invasive techniques. Given its physiologic and biomechanical advantages, it can be recommended as the surgical method of choice in this indication. Underlying subclinical vascular factors may be involved in the complaints of spinal stenosis patients. Those factors should be investigated more thoroughly, as they may account for some of the failures of surgical relief. The CHAID decision tree appears to be a novel and useful tool for predicting the results of spinal stenosis surgery

Fig. 1

The chi-squared automatic interaction detection (CHAID) model for predicting surgical success in the subgroup of 32 patients, with complete data