Can Big Data Provide Predictive Analytics for Wound Trajectories?

Background: Accurately forecasting wound closure trajectories is difficult due to the incomplete knowledge of wound dynamics. However, the ability to predict wound trajectories would greatly benefit wound care practitioners in identifying high risk wounds while also determining treatment efficacy.

Purpose: Here, we demonstrate that models based on big data and modern predictive analytical techniques can adapt to patient specific variations and be more suited for inference in a clinical setting.

Methods: A predictive model was created that more accurately predicts time-to-heal than an existing model based on PUSH scores (Gardner, 2011). The model to predict the trajectory of wound area progression is based on an exponential function using linear mixed-effects model (LMM) methodology. LMMs incorporate inter-subject variances and adapt their predictions to individual subjects/patients. As such, LMMs model the dependent variable as a combination of weighted fixed (population) effects and random (subject-specific) effects.

Results: The proposed model predicted wound trajectory with a goodness-of-fit (r2) value of 0.40. Also, it was observed that the median error in predicting time-to-heal for wounds with size greater than 1, 5 and 8 cm2 for the PUSH-based model was 30, 50 and 55 days, whereas for the proposed model it was 20, 17 and 23 days, respectively. Therefore, the proposed LMM based exponential model is much more robust in predicting time-to-heal as compared to the PUSH-based model that was previously published.

Conclusions: The application of predictive algorithms on wound trajectories can be applied to digital wound care management to see if wounds are healing as expected, or alternatively, if treatment strategy needs to be changed. Our future work would focus on incorporating more data and more clinically relevant features to improve model performance.