Risk assessment tool for diabetic neuropathy.

Abstract

Peripheral neuropathy is one of the serious complications of diabetes. Symptoms such as tingling and loss of touch sensation are commonly associated with the early stages of neuropathy causing numbness in the feet. Early detection of this condition is necessary in order to prevent the progression of the disease. Out of many detection techniques vibration perception is becoming the gold standard for neuropathy assessment. Devices like tuning fork, Biothesiometer and Neurothesiometer use this technology but require an operator to record and manually interpret the results. The results are user-dependent and are not consistent. To overcome these limitations, a platform-based device “VibraScan” was developed that can be self-operated and results displayed on a user interface. The development of the device is based on studying the effect of the vibration on the human subject by identifying the receptors responsible for sensation. The requirement of generating vibration was achieved by selecting a specific actuator that creates vibration perpendicular to the contact surface. The battery operated VibraScan is wirelessly controlled by software to generate vibration for determining the vibration perception threshold (VPT). Care has been taken while developing the user interface for human safety with the vibration intensity. The device can be operated without any assistance and results are automatically interpreted in terms of severity level indicated similar to the traffic-light classification. In order to provide consistent results with the existing devices a study was undertaken between Neurothesiometer and VibraScan with 20 healthy subjects. The results were compared using Bland-Altman plot and a close agreement was found between the two measurements. VibraScan accurately measures VPT based on the perceived vibration threshold, however, it does not predict any risk associated with neuropathy. In order to supplement this device with the progression of neuropathy a risk assessment tool was developed for automated prediction of neuropathy based on the clinical history of patients. The smart tool is based on the research related to the risk factors of diabetic neuropathy which was studied and analysed using summarised patient data. Box-Cox regression was used with the response variable (VPT) and a set of clinical variables as potential predictors. Significant predictors were: age, height, weight, urine albumin to creatinine ratio (ACR), HbA1c, cholesterol and duration of diabetes. Ordinary Least Squares Regression was then used with logarithmic (VPT) and the significant predictor set (Box-Cox transformed) to obtain additional fit estimates. With the aim to improving the precision of VPT prediction, a simulated patient data set (n = 4158) was also generated using the mean and the covariance of the original patient variables, but with reduced standard errors. For clinical or patient use, providing direct knowledge of VPT was considered less helpful than providing a simple risk category corresponding to a range of VPT values. To achieve this, the continuous scale VPT was recoded into three categories based on the following clinical thresholds in volts (V): low risk (0 to 20.99 V), medium risk (21 to 30.99 V) and high risk (≥ 31 V). Ordinal Logistic Regression was then used with this categorical outcome variable to confirm the original predictor set. Having established the effectiveness of this “classical” baseline, attention turned to Neural Network modelling. This showed that a carefully tuned Neural Network based Proportional Odds Model (NNPOM) could achieve a classification success >70%, somewhat higher than that obtained with the classical modelling. A version of this model was implemented in the VibraScan risk assessment tool. Integrating VibraScan and the risk assessment software has created a comprehensive diagnostic tool for diabetic neuropathy.