90 Chapter 6 DISCUSSION This study aimed to validate a multicenter risk model for implant loss after implantbased breast reconstructions, using the DBIR database. Although the observed implant loss rate increased when the risk factors accumulated, the calibration plot showed that the predicted probability of implant loss based on the previous risk model and the observed probability in the current nationwide cohort do not match. This implies that the previous created risk model is not generalizable to the reconstructive population at large. It is crucial that any developed model is generalizable and predicts well in ‘comparable but different’ patients outside the development set.15 In the current validation cohort, an implant loss rate of 8.1% was found after implant-based breast reconstruction, which is slightly lower than the 11.8% implant loss rate found in the original cohort. The previous risk model consists of four risk factors: obesity, active smoking status, a nipple sparing procedure and a DTI approach. BMI, smoking status and a DTI approach could directly be extracted from the DBIR data. However, a nipple sparing procedure was not an exact variable in the DBIR database and could only be derived from the incision type. Furthermore, substantial differences in baseline characteristics were observed between the validation cohort and previous multicenter cohort. Next to ASA score, indication Figure 2. Calibration plot Ratio between the predicted probability on implant loss based on the previous risk model and the observed probability in the current cohort.
RkJQdWJsaXNoZXIy MTk4NDMw