Predicting Responses to Pregabalin for Painful Diabetic Peripheral Neuropathy Based on Trajectory-Focused Patient Profiles Derived from the First 4 Weeks of Treatment

Abstract

Introduction: Prediction of final clinical outcomes based on early weeks of treatment can enable more effective patient care for chronic pain. Our goal was to predict, with at least 90% accuracy, 12- to 13-week outcomes for pregabalin-treated painful diabetic peripheral neuropathy (pDPN) patients based on 4 weeks of pain and pain-related sleep interference data.

Methods: We utilized active treatment data from six placebo-controlled randomized controlled trials (n = 939) designed to evaluate efficacy of pregabalin for reducing pain in patients with pDPN. We implemented a three-step, trajectory-focused analytics approach based upon patient responses collected during the first 4 weeks using monotonicity, path length, frequency domain (FD), and k-nearest neighbor (kNN) methods. The first two steps were based on combinations of baseline pain, pain at 4 weeks, weekly monotonicity and path length during the first 4 weeks, and assignment of patients to one of four responder groups (based on presence/absence of 50% or 30% reduction from baseline pain at 4 and at 12/13 weeks). The third step included agreement between prediction of logistic regression of daily FD amplitudes and assignment made from kNN analyses.

Results: Step 1 correctly assigned 520/939 patients from the six studies to a responder group using a 3-metric combination approach based on unique assignment to a 50% responder group. Step 2 (applied to the remaining 419 patients) predicted an additional 121 patients, using a blend of 50% and 30% responder thresholds. Step 3 (using a combination of FD and kNN analyses) predicted 204 of the remaining 298 patients using the 50% responder threshold. Our approach correctly predicted 90.0% of all patients.

Conclusion: By correctly predicting 12- to 13-week responder outcomes with 90% accuracy based on responses from the first month of treatment, we demonstrated the value of trajectory measures in predicting pDPN patient response to pregabalin.

Trial registration: www.clinicaltrials.gov identifiers, NCT00156078/NCT00159679/NCT00143156/NCT00553475.

Funding: Pfizer. Plain language summary available for this article.

Keywords: Frequency domain; K-nearest neighbor (kNN); Monotonicity; Painful diabetic peripheral neuropathy (pDPN); Pregabalin; Trajectory prediction.

Fig. 1

Flow chart of steps for prediction of responder status at 12 or 13 weeks. FD frequency domain, kNN k-nearest neighbor, Pain₀ pain at week 0, Pain₄ pain at week 4, PMono₄ monotonicity of weekly pain from week 0 to week 4, PPL₄ path length of weekly pain from week 0 to week 4, 30 PRS₄ pain responder status at 30% at week 4, PRSI₀ pain-related sleep interference at week 0, PRSIPL₄ path length of weekly sleep interference from week 0 to week 4, 30 PRSIRS₄ pain-related sleep interference responder status of 30% at week 4, 50 PRSIRS₄ pain-related sleep interference responder status of 50% at week 4. The specific actions associated with the steps in the flow chart are shown below: (1) Collect two data elements at baseline and weekly until week 4 so that four data points exist for each patient: Pain on 0–10 NRS and pain-related sleep interference on 0–10 NRS. (2) Calculate monotonicity for the first 4 weeks (see Supplemental File 1). (3) Calculate path length for the first 4 weeks (see Supplemental File 2). (4) Generate the following four combinations of three of the data elements generated in the prior three actions: (a) 4-week monotonicity, 4-week path length, pain score at week 4 (PMono₄-PPL₄-Pain₄), (b) 4-week monotonicity, 4-week path length, pain score at baseline (PMono₄-PPL₄-Pain₀), (c) 4-week monotonicity, pain score at week 4, pain score at baseline (PMono₄-Pain₄-Pain₀), and (d) 4-week path length, pain score at week 4, pain score at baseline (PPL₄-Pain₄-Pain₀). (5) Check four patterns and see if the pattern aligns with those that are uniquely associated with one of the four responder groups (responders at both week 4 and the final week, non-responders at both week 4 and the final week, responders at week 4 but non-responders at the final week, non-responders at week 4 but responders at the final week). (6) If the pattern aligns, predict patient outcome at the final week (Step 1). (7) If the pattern does not align, move to Step 2a and check whether the pattern aligns with those uniquely associated with one of the four responder groups when the 30% threshold for being a responder in the final week is used. (8) If the pattern aligns with those uniquely associated with one of the four responder groups, then predict the patient’s outcome at the final week (Step 2a). (9) If the pattern does not align, move to Step 2b and check whether  the pattern aligns with those uniquely associated with one of the four responder groups when the 30% threshold is used for being a responder in the final week and in week 4. (10) If the pattern aligns with those uniquely associated with one of the four responder groups, then predict the patient’s outcome at the final week (Step 2b). (11) If the pattern does not align, move to Step 3 and implement the kNN analysis (see Supplemental File 4) by considering the following seven data elements for describing each patient: (a) pain-related sleep interference at baseline, (b) pain score at baseline, (c) 4-week path length of pain-related sleep interference, (d) 4-week path length of pain, (e) pain-related sleep interference responder status at week 4 (30% threshold), (f) pain-related sleep interference responder status at week 4 (50% threshold), (g) pain responder status at week 4 (30% threshold). (12) Identify if there are one or more nearest neighbors; if there is only one neighbor with the same vector values, then use it to predict the patient’s outcome and if there is more than one neighbor with the same value, the majority is selected for the prediction (see Supplemental File 4 for examples). (13) Before selecting the final choice of outcome for Step 3, also implement the FD analysis. For the FD analysis, use 28 days of daily pain score data and follow the steps outlined in Supplemental File 3. (14) Compare the outcomes predicted by the FD analysis with the outcome predicted by the kNN analysis. If the both the FD and kNN analyses assign the patient to the same responder group, select that responder group for the outcome. (15) If the responder group assignment differs between the FD analysis and the kNN analysis, use the responder group based on the one assigned by the FD analysis if the patient was a responder at week 4; use the responder group based on the one assigned by the kNN analysis if the patient was a non-responder at week 4. (16) If daily data are not available, use the kNN analysis alone for Step 3