Page 109 - Strategies for non-invasive managementof high-grade cervical intraepithelial neoplasia - prognostic biomarkers and immunotherapy Margot Maria Koeneman
P. 109

the prediction model, it seems sensible to take age into account when counseling a patient for conservative treatment. The largest benefit of conservative treatment will be for younger patients with a pregnancy wish in the future, by omitting surgical treatment and the associated risk of preterm birth. As such, conservative treatment is primarily advised for younger patients. The current prediction model aids the decision-making in these patients based on four other parameters.
The discriminative ability of the prediction model can be considered reasonable-to-good. The
AUC can range from 50 to 100%, indicating no discriminative capacity to perfect discriminative
capacity. Generally, an AUC of 70% and higher is considered indicative of good discriminative
ability. The AUC of our model is 69.2%, which almost reaches 70% and therefore indicates a reasonable-to-good discriminative ability. The model is most suitable for prediction of disease 5 regression in patients with high predicted probabilities of disease regression. Although a cut-
off value of 70% identifies 80% of patients who will show disease regression correctly, the 20% of patients will still experience disease persistence and could be considered ‘under treated’. However, several studies have shown that disease progression within the follow up period of our study is very rare, making watchful waiting a safe option for all patients.[6, 23, 24] We therefore advocate the use of this model to reassure patients that observation is an appropriate treatment option. As such, the model could function as a tool for a more individualized approach to CIN 2 management.
Strengths of the study include the population size and robust statistical methods. Another strength of the study is the internal validation procedure, as an alternative to external validation when a validation population is lacking, Large populations of untreated women with CIN 2 are currently scarce, as conservative management of CIN 2 has only recently been included in treatment guidelines. In these situations, it is not uncommon to apply only internal validation to a newly developed prediction model[29, 30]. Prospective external validation is of course desirable and encouraged. An important limitation to this study is the low reproducibility of CIN 2 diagnoses [31]. All biopsies were assessed according to the WHO guidelines. The biopsies were not re-analyzed for this study, as this does not necessarily lead to a more reliable diagnosis and is not performed in routine clinical practice. Another potential limitation of the study is the high observed regression rate in our population (71%). The sample size calculation in this study was based on the regression rate of CIN 2 in a randomized controlled trial. However, studies on spontaneous regression of CIN 2 show a wide range of regression rates over different follow-up periods, with several studies reporting high regression rates of up to 71% [32, 33]. Our high regression rate may therefore be the upper range of normal, or may be (partially) due to a selection bias, as a wait-and-see policy may have been largely applied to patients with a high perceived chance of disease regression. Another explanation could be that some patients were over-diagnosed, as a result of the low reproducibility of CIN 2 diagnoses. Prospective validation of the model should clarify this issue. Additionally, a prospective study should assess the usefulness and clinical impact of the model, in order to assess whether its implementation leads to the desired effect of less overtreatment of CIN 2, without under-treatment.
A prediction model for CIN2 prognosis
 107


























































































   107   108   109   110   111