Background & Aims
Quantitative sensory testing (QST) is used to quantitively measure sensory nerve function via patient response to the application of various sensory stimuli. QST has been used in neuropathic pain clinical trials to confirm altered sensory function for eligibility, identify phenotypical subgroups, and changes in sensory function with intervention. Sensory modalities that are typically tested include touch, vibration, temperature, and pain.1
The Bedside Sensory Testing Kit (BSTK) is a low-cost, simplified alternative to traditional QST. Sensory tests assessed mechanical allodynia (MA), cold allodynia (CA), punctate hyperalgesia (PH), temporal summation (TS), and low-threshold mechanoreceptive function (LTM).2,3 Initial cluster analysis of BSTK responses revealed 2 phenotypes: sensory gain (heightened response on MA and CA) and sensory loss (dampened response on all tests).4 The current study sought to develop an algorithm for prospectively phenotyping patients based on BSTK results.
Methods
Data was aggregated from three clinical trials in painful diabetic neuropathy: Study One (N = 333), Study Two (N= 161) and Study Three (N = 138). Subjects in all three studies were administered the BSTK at prior to randomization. A cluster analysis was performed on the Study One training data to obtain the subgroup assignments. Several classification algorithms were employed with varying complexity to create the scoring algorithm via logistic regression and Linear Discriminant Analysis (LDA) using all 5 sensory tests. Several more complicated machine learning approaches such as Naive Bayes and classification trees with Random Forests were also implemented to identify the simplest algorithm based on some or all of the 5 sensory tests. These scoring algorithms were then used to predict subgroups on the Study Two and Study Three data.
Results
Exploratory factor analysis found a single factor solution that loaded highest onto MA and CA for all three studies. The scoring algorithm accurately and consistently assigned patients to the two clusters (i.e., sensory gain, sensory loss). All models were able to predict assignment consistently with accuracies close to 0.9 or greater.
The algorithm was able to be reduced to MA and CA with negligible impact on accuracy. The algorithm with MA only had slightly higher accuracy than the algorithm with CA only. The LDA and Logistic scoring algorithms trained on MA predicted subgroup assignments with 0.909 accuracy. The Random Forest and Naïve Bayes scoring algorithm predicted subgroup assignments with 0.932 accuracy. The scoring algorithm that was trained solely on CA achieved an accuracy of 0.882 on all four models. Models that contained MA or CA performed better than models with neither MA nor CA. MA and CA combined predicted assignment with a .92 accuracy; MA alone with .91 accuracy.
Conclusions
The current findings indicate that assessing mechanical allodynia alone is sufficient for assigning patients to phenotypes based on baseline sensory function in clinical trials. Decreasing the number of tests needed for accurate phenotyping reduces site and patient burden, as well as costs for materials and training. It is unclear whether these findings would have cross-indication applicability or be valid when using sensory testing as an outcome measure; however, reducing sensory testing to a single test of mechanical allodynia may facilitate baseline phenotyping in clinical trials in diabetic peripheral neuropathy.
References
1.Backonja MM et al. Value of quantitative sensory testing in neurological and pain disorders: NeuPSIG consensus. Pain. 2013; 154(9): 1807-1819. doi: 10.1016/j.pain.2013.05.047. Erratum in: Pain. 2014; 155(1): 205.
2.Osgood E, Eaton T, Trudeau J, Gammaitoni, A, Katz NP. Development of a bedside sensory testing kit for postherpetic neuralgia. The Journal of Pain. 2012; 13(4, S10): A139. https://doi.org/10.1016/j.jpain.2012.01.048
3.Osgood E et al. Development of a bedside pain assessment kit for the classification of patients with osteoarthritis. Rheumatol Int. 2015; 35(6): 1005-13. doi: 10.1007/s00296-014-3191-z.
4.Evans K, Erpelding N, Lanier R, Elder H, Katz NP. Development and validation of the Analgesic Solutions Bedside Sensory Testing Kit for painful diabetic neuropathy. Poster presented at 2018 NIH Pain Consortium Symposium, Bethesda, MD.