Interpolating and Forecasting Wound Trajectory using Machine Learning Approaches

Chronic wound treatment and management heavily rely on accurate patient outcome predictions. As chronic wounds become increasingly prevalent in many US states, it’s crucial to accurately predict and classify wound progress for timely and effective clinical intervention. The trajectory and outcomes of wounds inform multiple aspects of care, from estimating healing times to understanding disparities in treatment quality and outcomes. For accurate predictions, robust time-series forecast models are required, necessitating comprehensive and consistent wound data. Yet, much of this data is frequently marred by gaps and inaccuracies. In this work, we combined two approaches to overcome this challenge: filling in missing clinical data and forecasting wound healing trajectories. We employed various interpolation techniques, time series, and machine-learning models, including Holt-Winters, ARIMA, Prophet, LSTM, and BiLSTM. Our study assessed 14,571 wounds from 6,171 patients over a five-year span, post-data imputation and interpolation. We introduced a novel approach to firstly impute missing data using XGBoost and then choose the best interpolation method for absent wound data, ensuring precise forecasting without distorting the dataset’s demographic makeup. Data for this research was sourced from four wound centers in Indiana and eight research studies, focusing on a diverse range of wound-related demographics, clinical, and omics data. This comprehensive approach enhances wound trajectory predictions, providing clinicians with valuable insights to formulate personalized treatment plans for both individual patients and demographic groups. This will aid clinicians in better understanding wound healing trends and crafting targeted treatment strategies.