Commonly used methods of estimation of insulin resistance in diabetic children seems to be biased in children younger than 12 years or various ethnicity. Based on this observation and our group of 315 patients, we have developed two artificial neural networks (ANN) and one MARSplines model of insulin resistance utilizing different sets and definitions of clinical parameters. Although validation of MARSplines model showed overfitting, ANNs and the reference model performed similarly in training and hold-out test set. Predictions made by neural networks remained within ±20% error band in 75% of cases in both the training and test set, and didn’t show the bias of reference model. Developed ANNs surpassed reference model and therefore could be applied for qualification of children with type 1 diabetes complicated with marked insulin resistance or obesity into clinical trials utilizing additional medications.
Please download the Windows exectuable file and run on your desktop. First of all you will have to choose which output models you would like to use. This is important as different models require different input and therefore specific fields will be locked or unlocked. After you will complete clinical details press evaluate button to get the results. Please note that this software works online and downloaded executable is just a client protocol for scoring system.
This supplementary software is a part of article Stawiski K , Pietrzak I , Młynarski W , Fendler W , Szadkowska A . NIRCa: An artificial neural network-based insulin resistance calculator. Pediatr Diabetes. 2017;0:1-5. https://doi.org/10.1111/pedi.12551
Developer: Konrad Stawiski, M.D. (email@example.com; Department of Biostatistics and Translational Medicine, Medical University of Lodz, Poland.)