Utilizing Precision Medicine to Estimate Timing for Surgical Closure of Traumatic Extremity Wounds.

Abstract

Both the frequency and high complication rates associated with extremity wounds in recent military conflicts have highlighted the need for clinical decision support tools (CDST) to decrease time to wound closure and wound failure rates. Machine learning was used to estimate both successful wound closure (based on penultimate debridement biomarker data) and the necessary number of surgical debridements (based on presentation biomarkers) in 73 service members treated according to military guidelines based on clinical data and the local/systemic level of 32 cytokines. Models were trained to estimate successful closure including an additional 8 of 80 civilian patients with similar injury patterns. Previous analysis has demonstrated the potential to reduce the number of operative debridements by 2, with resulting decreases in ICU and hospital LOS, while decreasing the rate of wound failure. Analysis showed similar cytokine responses when civilians followed a military-like treatment schedule with surgical debridements every 24 to 72 hours. A model estimating successful closure had AUC of 0.89. Model performance in civilians degraded when these had a debridement interval > 72 hours (73 of the 80 civilians). A separate model estimating the number of debridements required to achieve successful closure had a multiclass AUC of 0.81. CDSTs can be developed using biologically compatible civilian and military populations as cytokine response is highly influenced by surgical treatment. Our CDSTs may help identify who may require serial debridements versus early closure, and precisely when traumatic wounds should optimally be closed.