Background and Objective: Spinal pain is highly prevalent, yet clinically accessible tools for pain quantification remain limited. This study evaluated whether machine learning (ML) algorithms using StimaWELL 120MTRS-derived scores alongside demographic variables can predict self-reported spinal pain intensity and compared their performance against linear mixed model (LMM)-based associations. Methods: Seventy-eight adults with spinal pain (44 female, 34 male; mean age 43.08 ± 11.50 years) were assessed under two electrode configurations: opposing electrode (OE) and adjacent electrode (AE). Self-reported pain was measured using the Numeric Pain Rating Scale (NPRS). Repeated measures correlation, intraclass correlation coefficients (ICC), LMM, and eight ML algorithms were applied to evaluate predictive capacity. Results: Repeated measures correlations between StimaWELL scores and NPRS were negligible and non-significant (OE: r = −0.003, p = 0.970; AE: r = 0.015, p = 0.876), with ICC values indicating poor agreement (OE: 0.364; AE: 0.452). LMM revealed no statistically significant fixed effects for either configuration. Despite this, ML algorithms achieved moderate predictive performance, with Extra Trees and CatBoost yielding the highest tolerance-based accuracy (up to 73–74% for AE-based models). The AE StimaWELL score was the dominant predictor across feature importance analyses, followed by body weight and height. Conclusions: ML algorithms extracted the predictive signal from StimaWELL 120MTRS data where LMM found none, highlighting nonlinear relationships between device-derived scores and pain intensity. AE configuration demonstrated superior predictive performance. These findings suggest that StimaWELL 120MTRS may provide complementary quantitative information in spinal pain assessment, pending external validation in larger, diverse populations.
No takes yet. Share an insight, caveat, or question.
Okuyucu et al. (2026) studied this question.