Machine Learning for Prognostication: Harnessing Chronic Wound Registry Data for Predicting Wound Healing Outcomes

Chronic wounds have been a pressing concern worldwide, and the most significant issue is unnecessary amputations. It is made worse by delays in receiving suitable treatment which can be avoided. To address this, this paper introduces a pivotal algorithm that utilizes Gradient Boosting classifier to forecast the likelihood of wound healing within specific time frames. The wound healing predictions are made at three distinct intervals: 1 month, 3 months, and 6 months following the initial visit to the healthcare facility by the patient. The algorithm relies exclusively on the data gathered during the patient's initial visit and a systematically assembled chronic wound registry. The algorithm yields an area under the receiver operating characteristic curve (AUC) of 0.80, 0.74, and 0.70 for the prediction at the 3 time-points using the baseline features. Additionally, this paper also yields an additional result of 0.76 and 0.73 for the prediction at the three-months and six-months using the baseline plus one-month data. The proposed prediction model uses mainly tabular data which are readily available to clinicians after the initial visit. This helps the clinicians to have a more accurate diagnosis of the wound healing time, therefore allowing the clinicians to intervene early to prevent unnecessary amputations that are very costly to the patients and to provide the appropriate treatment to prevent the loss of lives through chronic wound injuries.