Healing profiles in patients with a chronic diabetic foot ulcer: an exploratory study with machine learning

Diabetic foot ulcers (DFU) are one of the most frequent and debilitating complications of diabetes. DFU wound healing is a highly complex process, resulting in significant medical, economic and social challenges. Therefore, early identification of patients with a high-risk profile would be important to adequate treatment and more successful health outcomes. This study explores risk assessment profiles for DFU healing and healing prognosis, using machine learning predictive approaches and decision tree algorithms. Patients were evaluated at baseline (T0; N = 158) and 2 months later (T1; N = 108) on sociodemographic, clinical, biochemical and psychological variables. The performance evaluation of the models comprised F1-score, accuracy, precision and recall. Only profiles with F1-score >0.7 were selected for analysis. According to the two profiles generated for DFU healing, the most important predictive factors were illness representations on T1 IPQ-B (IPQ-B ≤ 9.5 and < 10.5) and the DFU duration (≤ 13 weeks). The two predictive models for DFU healing prognosis suggest that biochemical factors are the best predictors of a favorable healing prognosis, namely IL-6, microRNA-146a-5p and PECAM-1 at T0 and angiopoietin-2 at T1. Illness perception at T0 (IPQ-B ≤ 39.5) also emerged as a relevant predictor for healing prognosis. The results emphasize the importance of DFU duration, illness perception and biochemical markers as predictors of healing in chronic DFUs. Future research is needed to confirm and test the obtained predictive models.
This study was conducted at CIPsi, School of Psychology, University of Minho, supported by Fundação para a Ciência e a Tecnologia (FCT) through the Portuguese State Budget (UID/01662/2020), and by COMPETE 2020 - Operational Programme for Competitiveness and Internationalisation (POCI-01-0145-FEDER-028163) under the project PTDC/PSI-GER/28163/2017 assigned to the first author (M. Graça Pereira); COMPETE 2020 and FCT, under projects POCI-01-0145-FEDER-007440, UIDB/04539/2020, UIDP/04539/2020 and LA/P/0058/2020 assigned to the last author (Eugénia Carvalho); and the Ph.D. grants 2020.04990.BD (Jéssica Da Silva) and SFRH/BD/144199/2019 (Diana Santos).