A machine-learning approach was able to classify patients with painful diabetic polyneuropathy (DPN) into irritable and nonirritable nociceptor phenotypes. These findings were published in Diabetologia.
Researchers from the University of Sheffield in the United Kingdom recruited 43 patients with painful DPN from the Sheffield Teaching Hospital for this study. Patients underwent magnetic resonance imaging (MRI), nerve conduction assessment, quantitative sensory assessment, and sensory phenotyping. The researchers used these data to classify patient groups using a support vector machine classifier.
Patients who had irritable (n=10) and nonirritable (n=33) nociceptor phenotypes were a mean age of 56.9±12.9 years and 58.4±11.2 years, respectively; 90% and 60.6% were male, respectively; and pain had persisted for 8.9±5.6 and 8.3±7 years, respectively.
Compared between groups, patients with the irritable phenotype had increased dynamic mechanical allodynia (P <.001), paradoxical heat sensation (P =.001), mechanical pain sensitivity (P =.003), and pressure pain thresholds (P <.05) and reduced cold pain threshold (P =.05).
Patient groups differed significantly in right thalamus volume (mean, 6475±701.3 mm3 vs 5874.4±626.3 mm3; P =.01), anterior cingulate cortex thickness (mean, 2.38±0.2 mm vs 2.57±0.2 mm; P =.02), and somatosensory cortex surface area (mean, 578.2±64.3 mm2 vs 535.3±54.1 mm2; P =.04) for irritable and nonirritable phenotypes, respectively.
During functional MRI, connectivity of the thalamus among patients with irritable phenotype increased with the insular cortex (b=0.2; T=3.11; P =.03) and decreased with the somatosensory cortex (b=-0.22; T=-4.98; P =.03). The functional connectivity between thalamus and insular cortex correlated with Neuropathy Total Symptom Severity-6 score (r =0.41; P =.01) and between thalamus and somatosensory cortex with Toronto Clinical Neuropathy Score (r =-0.35; P =.03).
Using these data, the machine-learning model grouped patients into the nonirritable subgroup with an accuracy of 0.92 (95% CI, 0.08), sensitivity of 90%, positive predictive value of 100%, and negative predictive value of 67%.
This study may have been limited by its skewed gender proportion and imbalanced cohorts.
These data indicated there were phenotype-specific alterations to pain-processing brain regions among patients with painful DPN. Additional studies are needed to determine whether these brain features and model may have use in patient stratification.
Disclosure: An author declared affiliations with industry. Please refer to the original article for a full list of disclosures.
Teh K, Wilkinson ID, Heiberg-Gibbons F, et al. Somatosensory network functional connectivity differentiates clinical pain phenotypes in diabetic neuropathy. Diabetologia. Published online March 25, 2021. doi:10.1007/s00125-021-05416-4