107 Risk prediction of implant loss following implant-based breast reconstruction a mean ratio of 0.753 across all tenths. The tenths, with CIs for observed probability, are visualized in a calibration plot (Figure 2). As the reference line lies within the CIs of seven of the tenths, a good agreement can be concluded. Additional analyses A subgroup of TE and DTI reconstructions was separately analyzed. A risk prediction model for implant loss after TE reconstructions was created in the training cohort. This model included the following risk factors: BMI (β: 0.046, OR: 1.047, CI: 1.012-1.082), active smoking status (β: 0.453, OR: 1.573, CI: 1.035-2.389) and prepectoral placement (β: 0.763, OR: 2.145, CI: 1.337-3.440). The model was applied to the validation cohort and divided into quintiles because of a low sample size. Each quintile contained 113 to 120 subjects. The ratio between the observed and predicted probability ranged from 0.301 to 1.055, with a mean ratio of 0.811. This comparison is visualized in Figure 3. The reference line lies within the CIs of four out of five quintiles. Figure 3. Validation of the TE risk prediction model using a calibration plot Subsequently, a risk prediction model was created for DTI reconstructions in the training cohort. This model included the following risk factors: prophylactic mastectomy (β: 0.867, OR: 2.380, CI: 1.378-4.109), active smoking status (β: 1.495, 7
RkJQdWJsaXNoZXIy MTk4NDMw