The amputation prevention initiative

Prediction of Diabetes Mellitus Progression Using Supervised Machine Learning

Diabetic peripheral neuropathy (DN) is a serious complication of diabetes mellitus (DM) that can lead to foot ulceration and eventual amputation if not treated properly. Therefore, detecting DN early is important. This study presents an approach for diagnosing various stages of the progression of DM in lower extremities using machine learning to classify individuals with prediabetes (PD; n = 19), diabetes without (D; n = 62), and diabetes with peripheral neuropathy (DN; n = 29) based on dynamic pressure distribution collected using pressure-measuring insoles. Dynamic plantar pressure measurements were recorded bilaterally (60 Hz) for several steps during the support phase of walking while participants walked at self-selected speeds over a straight path. Pressure data were grouped and divided into three plantar regions: rearfoot, midfoot, and forefoot. For each region, peak plantar pressure, peak pressure gradient, and pressure-time integral were calculated. A variety of supervised machine learning algorithms were used to assess the performance of models trained using different combinations of pressure and non-pressure features to predict diagnoses. The effects of choosing various subsets of these features on the model's accuracy were also considered. The best performing models produced accuracies between 94-100%, showing the proposed approach can be used to augment current diagnostic methods.

Choosing a Vibratory Test to Pair With Semmes Weinstein Monofilament Testing for Evaluating Lower Extremity Sensation in Patients With Diabetes: A Comparison of Three Vibratory Methodologies

BACKGROUND

Numerous guidelines recommend pairing Semmes-Weinstein monofilament (SWM) testing with a secondary clinical test when screening for diabetic peripheral neuropathy, yet time is very limited in clinical practice. This study compared the time to complete and the diagnostic agreement of three vibratory sensation tests.

METHODS

Sixty-five individuals (42% male; aged: 61 ± 12 years) were recruited. A single investigator administered the following tests bilaterally: 10-site SWM, traditional tuning fork (TTF), electronic tuning fork (ETF), and vibration perception threshold (VPT) via biothesiometer. Times to physically administer the tests were compared with a one-way repeated measures ANOVA. Cochran's Q test was used to compare the varied tests' diagnostic agreement.

RESULTS

The ANOVA indicated there were significant (P < .001, partial eta squared = .442) differences in time to complete the varied tests. Sidak post hoc comparisons indicated the VPT (21.2 ± 14.3) testing took an intermediate time to complete, while the ETF (9.7 ± 6.5) and TTF (10.1 ± 7.5) tests took the least amount of time, and the SWM (28.6 ± 8.4) test took the longest time. There were also numerous significant differences (P ≤ .001) between the different tests in regards to neuropathy diagnoses.

CONCLUSIONS

Tuning fork methods required 11 seconds less to administer than VPT testing. Although that may seem trivial, it is worth noting peripheral neuropathy screening often fails to occur in the precious few minutes clinicians are allotted per patient. Considering ETF's intrinsic control of stimulus amplitude and its ease of use with an embedded timer, the ETF is recommended over the TTF. Clinicians should also be mindful that different tests yield different diagnostic conclusions.