Page 105 - Strategies for non-invasive managementof high-grade cervical intraepithelial neoplasia - prognostic biomarkers and immunotherapy Margot Maria Koeneman

A prediction model for CIN2 prognosis
Results
Population characteristics
A total of 129 women were eligible for inclusion in the study. Spontaneous regression within 24 months occurred in 91 of these women (71%). None of the women showed disease progression beyond CIN 3. Most (106) women completed the 24 month follow-up. Of the 23 women that did not complete the follow-up, those with a test result of PAP 1 after 6 and/or 12 months were included in the regression group (n = 21), while the others were included in the non-regression group (n = 2). Table 1 shows the population characteristics with regard to the selected predictors in the model.
5
Table 1. Patient characteristics with regard to the selected predictors in the model
Mean age at diagnosis, mean (range)
Smoking, no. (%)
PAP 1 or PAP 2 (ASCUS) before CIN 2 diagnosis, no. (%) Concomitant CIN 1 in same biopsy, no. (%)
More than one biopsy containing CIN 2, no. (%)
Model development and internal validation
Missing (%)
0
22 (17%) 0
0
0
Total
36 (17–74) 61 (47%) 21 (16%) 46 (36%) 9 (7%)
Missing values were only present for smoking as a predictor. After imputation, all 129 women were available for development of the prediction model. After entering the five potential predictor variables in the model, the following predictors met our selection criteria: smoking, a result of PAP 1 or PAP 2 (ASCUS) before the CIN 2 diagnosis, concomitant CIN 1 in the same biopsy, and more than one biopsy containing CIN 2. All outcomes are dichotomous variables. These four variables were combined into one model. Table 2 shows the original prediction model that estimates the spontaneous regression of CIN 2 within 24 months. Internal validation of the model by bootstrapping produced a shrinkage factor of 0.95, which was used to adjust the regression coefficients, as shown in Table 2. To estimate the individual probability of regression of CIN 2 within 24 months, the following predictive equation can be used: P(regression) = 100% × 1/(1+exp(- (0.928 - (0.677 * smoking) + (1.237 * PAP1 or PAP2 before CIN2 diagnosis) + (0.881 * concomitant CIN1 in same biopsy) - (0.752 * more than one biopsy containing CIN 2)))). For example, a smoking patient with a pre-biopsy PAP 3a (cytological LSIL with indications for HPV) and a concomitant CIN 1 in the same biopsy but only one CIN 2 lesion has a predicted probability of regression of 73.7%. With a diagnosis of a pure CIN 2 lesion, the probability of regression would drop to 56.2%.
103