99 Risk prediction of implant loss following implant-based breast reconstruction data, median and interquartile ranges (IQR) were given, other data are reported using frequencies and percentages. The dataset was randomly divided into a training cohort and a validation cohort, involving 80 and 20 percent of the data, respectively. The training cohort was used to identify risk factors and create a risk prediction model for implant loss. Univariable logistic regression analysis were performed to determine the association between potential risk factors and implant loss, providing odds ratios (OR) with 95% confidence intervals (CIs) and P values. Individual breasts were used as the unit of analysis and cases with missing data on risk factors were excluded from the analyses. Multivariable logistic regression analysis was used to adjust for all possible confounding variables. Finally, univariate risk factors with a P value below 0.157, based on the Akaike information criterion,(12) were inserted in a multivariable logistic regression model. Backward stepwise selection was performed to create the risk prediction model. Risk factors with a P value of <0.05 retained in the model. The risk prediction model was tested in the validation cohort. First, predicted probabilities of implant loss for each subject were calculated using β regression coefficients. Second, the subjects were divided into ten groups by using ten percentiles of the predicted probabilities. For each group, the observed probability of implant loss was calculated with corresponding 95% confidence interval. Finally, the probabilities were visualized in a calibration plot with predicted probability on x-axis and observed probability on y-axis. IBM SPSS statistics (version 26) was used for standard statistical analysis. Additional analyses Additional analyses were performed to investigate whether risk prediction models were more accurate for TE and permanent implants separately. Therefore, subgroups were created, stratified for TE and DTI, after which the analyses were repeated. Probability range for each number of risk factors In order to simplify the use of the risk model, the continuous variable (BMI) was dichotomized in BMI <30 and ≥30 kg/m2, excluding percentiles 0-2.5 and 97.5-100, resulting in a total of four dichotomous variables. The predicted probability range for each number of risk factors (zero to four) was computed using β regression coefficients. Furthermore, the observed implant loss rate was extracted from the training and validation cohort for each number of risk factors and visualized in a table. 7
RkJQdWJsaXNoZXIy MTk4NDMw