Propensity analysis and adjusted outcomes by cohorts of timing of DSWI diagnosis
Given the significant differences in baseline characteristics between the early and late diagnosis groups, five baseline variables were used to create a logistic regression model for selection to an early or late diagnosis of DSWI. Figure 3 shows the results of the logistic regression model predicting selection to the early or late diagnosis of DSWI group. Those more likely to be in the early diagnosis group were those with positive wound cultures (odds ratio 0.06, 95% CI 0.01-0.69, p=0.24) and those more likely to be in the late diagnosis group were females (OR 8.75, 95% CI 2.0-38.4, p=0.004) and those requiring urgent DSWI procedures (OR 9.25, 95% CI 1.86-45.9, p=0.007). The area under the receiver operator curve (c-statistic) was 0.84 and the Hosmer-Lemeshow goodness-of-fit test was not statistically significant (p=0.62), suggesting good model discrimination and limited collinearity and interactions (Supplemental Figure 1) .
Propensity score adjusted multivariable Cox proportional hazard regression outcomes for mortality are shown in Figure 4 . Early diagnosis of DSWI and initial attempted medical management were both strongly associated with increased mortality (hazard ratio 7.48, 95% CI 1.38-40.4, p=0.019 and hazard ratio 7.76, 95% CI 1.67-35.9, p=0.009, respectively). This was independent of initial operation (flap or negative pressure wound therapy) or whether any flap was eventually performed.