Machine learning analysis of primary hyperhidrosis for classification of hyperhidrosis type and prediction of compensatory hyperhidrosis.

Background: Although sympathectomy is highly effective for improving symptom, compensatory hyperhidrosis (CH) is a major issue. In this study, characteristics of primary hyperhidrosis were investigated in terms of the heart rate variability (HRV) parameters. Classification of hyperhidrosis type and prediction of CH after sympathicotomy were also determined using machine learning analysis.

Methods: From March 2017 to December 2021, 128 subjects who underwent HRV tests before sympathicotomy were analyzed. T2 and T3 bilateral endoscopic sympathicotomy were routinely performed in patients with craniofacial and palmar hyperhidrosis, respectively. Data collected age, sex, body mass index (BMI), hyperhidrosis type, symptom improvement after sympathicotomy, the degrees of CH after sympathicotomy, and preoperative HRV findings. The independent risk factors associated with the degree of CH after sympathicotomy were investigated. Machine learning analysis was used to determine classification of hyperhidrosis type and prediction of the degree of CH.

Results: Preoperatively, patients with palmar hyperhidrosis presented with significantly larger standard deviation of normal-to-normal (SDNN), root mean square of successive differences (RMSSD), total power (TP), and low frequency (LF) than patients with craniofacial hyperhidrosis after controlling for age and sex (P=0.030, P=0.004, P=0.041, and P=0.022, respectively). More sympathetic nervous predominance was found in craniofacial type (P=0.019). Low degree of CH had significantly greater RMSSD (P=0.047), and high degree of CH showed more sympathetic nervous predominance (P=0.006). Multivariate analysis showed the type and expansion of sympathicotomy were significant factors for CH (P=0.001 and P=0.028, respectively). The neural network (NN) algorithm outperformed and showed a 0.961 accuracy, 0.961 F1 score, 0.961 precision, 0.961 recall, and 0.972 area under the curve (AUC) for classification of hyperhidrosis type. The random forest (RF) model outperformed showed a 0.852 accuracy, 0.853 F1 score, 0.856 precision, 0.852 recall, and 0.914 AUC for prediction of the degree of CH.

Conclusions: The present study showed the machine learning algorithm can classify types and predict CH after sympathicotomy for primary hyperhidrosis with considerable accuracy. Further large-scale studies are needed to validate the findings and provide management guidelines for primary hyperhidrosis.