Objective: To evaluate four models based on potential predictors for achieving a response to pregabalin treatment for neuropathic pain (NeP).
Methods: In total, 46 pain studies were screened, of which 27 NeP studies met the criteria for inclusion in this analysis. Data were pooled from these 27 placebo-controlled randomized trials to assess if baseline characteristics (including mean pain and pain-related sleep interference [PRSI] scores), early clinical response during weeks 1-3 of treatment (change from baseline in pain and PRSI scores), and presence of treatment-emergent adverse events (AEs) were predictive of therapeutic response. Therapeutic response was defined as a ≥30% reduction from baseline in either pain and/or PRSI scores at week 5 with supplemental analyses to predict pain outcomes at weeks 8 and 12. Predictors of Patient Global Impression of Change (PGIC) were also evaluated. Four models were assessed: Random Forest, Logistic Regression, Naïve Bayes, and Partial Least Squares.
Results: The number of pregabalin-treated subjects in the training/test datasets, respectively, were 2818/1407 (30% pain analysis), 2812/1405 (30% sleep analysis), and 2693/1345 (PGIC analysis). All four models demonstrated consistent results, and the most important predictors of treatment outcomes at week 5 and pain outcomes at weeks 8 and 12 were the reduction in pain score and sleep score in the first 1-3 weeks. The presence or absence of the most common AEs in the first 1-3 weeks was not correlated with any treatment outcome.
Conclusions: Subjects with an early response to pregabalin are more likely to experience an end-of-treatment response.
Keywords: Baseline characteristics; Neuropathic pain; Predictive modeling; Pregabalin.