Challenges in the Analysis of Longitudinal Pain Data: Practical Lessons from a Randomized Trial of Annular Closure in Lumbar Disc Surgery

Abstract

Purpose. To analyze leg pain severity data from a randomized controlled trial (RCT) of lumbar disc surgery using integrated approaches that adjust pain scores collected at scheduled follow-up visits for confounding clinical events occurring between visits. Methods. Data were derived from an RCT of a bone-anchored annular closure device (ACD) following lumbar discectomy versus lumbar discectomy alone (Control) in patients with large postsurgical annular defects. Leg pain was recorded on a 0 to 100 scale at 6 weeks, 3 months, 6 months, 1 year, and 2 years of follow-up. Patients with pain reduction ≥20 points relative to baseline were considered responders. Unadjusted analyses utilized pain scores reported at follow-up visits. Since symptomatic reherniation signifies clinical failure of lumbar discectomy, integrated analyses adjusted pain scores following a symptomatic reherniation by baseline observation carried forward for continuous data or classification as nonresponders for categorical data. Results. Among 550 patients (272 ACD, 278 Control), symptomatic reherniation occurred in 10.3% of ACD patients and in 21.9% of controls (p < 0.001) through 2 years. There was no difference in leg pain scores at the 2-year visit between ACD and controls (12 versus 14; p = 0.33) in unadjusted analyses, but statistically significant differences favoring ACD (19 versus 29; p < 0.001) in integrated analyses. Unadjusted nonresponder rates were 6.0% with ACD and 6.7% with controls (p = 0.89), but 15.7% and 27.8% (p = 0.001) in integrated analyses. The probability of nonresponse was 16.4% with ACD and 18.3% with controls (p = 0.51) in unadjusted analysis, and 23.7% and 31.2% (p = 0.04) in integrated analyses. Conclusion. In an RCT of lumbar disc surgery, an integrated analysis of pain severity that adjusted for the confounding effects of clinical failures occurring between follow-up visits resulted in different conclusions compared to an unadjusted analysis of pain scores reported at follow-up visits only.