101 Risk prediction of implant loss following implant-based breast reconstruction Individual risk factors After adjusting for confounders, four individual risk factors were significantly associated to implant loss. These risk factors included BMI (adjusted OR: 1.050 per 1 kg/m increment, CI: 1.023-1.077), active smoking status (adjusted OR: 2.081, CI: 1.513-2.862), previous radiotherapy (adjusted OR: 1.811, CI: 1.064-3.081), prepectoral placement (adjusted OR: 1.911, CI: 1.346-2.713) and volume of the permanent implant (adjusted OR: 1.306, CI: 1.109-1.539, calculated per 100 cubic centimeters increase of volume). All factors and their correlation with implant loss before and after adjusting for confounders are summarized in Table 2. Risk prediction model After multivariable backward stepwise logistic regression analysis, the following four significant risk factors retained in the risk prediction model: BMI, active smoking status, previous radiotherapy and prepectoral placement. Within the training cohort, 974 out of 4208 reconstructions (23.1%) were excluded because of missing data on one or more risk factors. As volume of the permanent implant was not applicable to the total study population (about two third were TEs), it was only included in the subgroup analysis. The included factors and corresponding ORs, 95% CIs, β regression coefficients and P values are summarized in Table 3. An accurate risk prediction could be calculated for each individual patient using the following formulas. Log odds = -3.8715 + (BMI*0.0439) + (active smoking status*0.7823) + (previous radiotherapy*0.5213) + (prepectoral placement*0.5566). BMI is filled in as a continuous variable, active smoking status = 1, previous radiotherapy = 1 and prepectoral placement = 1. The predicted probability = (elog odds)/(1+elog odds)*100%. For example, the predicted probability of implant loss in a patient with a BMI of 25 kg/m2, active smoking status, no previous radiotherapy and no prepectoral placement is 12.01%, as calculated with the formula as: -3.8715 + (25*0.0439) + (1*0.7823) + (0*0.5213) + (0*0.5566) = -1.9917 (log odds). (e-1.9917)/(1+e1.9917) *100% = 12.01%. In addition, to facilitate an easy risk calculation, BMI was dichotomized to calculate the predicted probabilities of implant loss for each number of risk factors. BMI ranged from 18.6 to 30 and 30 to 34.6 (excluding percentiles 0-2.5 and 97.5-100). The predicted probabilities were compared to the observed implant loss rates in the training and validation cohort (Table 4). Risk prediction model validation The model was applied to the validation cohort, 273 out of 1052 reconstructions (26.0%) were excluded because of missing data on one or more risk factors. The predicted probabilities of ten percentiles were compared to the corresponding observed probabilities. Each tenth contained 74 to 83 subjects and the ratio between the observed and predicted probability ranged from 0.215 to 1.050, with 7
RkJQdWJsaXNoZXIy MTk4NDMw