ABSTRACT
AIMS: In France, guidelines for cervical cancer screening recommend that women between the ages of 25-65 have a smear test performed once every three years. However, some women are screened significantly more frequently. In this study, we used a data-driven approach as opposed to a traditional hypothesis-driven approach to characterise the population of women who are screened more frequently than advised. METHODS: Data came from an organised cervical cancer screening programme of a French department in the Alps (Isère). We retrospectively selected women aged between 25 to 65 years old who had at least two smear tests during the follow up period (2011-2015). We used a data-driven clustering approach to compare the population of over-screened women with other populations. We then performed a descriptive analysis of the over-screened population using univariate (Chi2 test) and multivariate (logistic regression) methods. RESULTS: A total of 10,000 patients were randomly chosen from a population of 54,073. In our univariate analysis, women in the over-screened population were significantly younger, participated less in organised screening, were more likely to be followed by a gynaecologist and had more smear test results showing inflammation than the other populations. Patient location (urban v.s. rural area) was not significant for this population. The multivariate analysis confirmed these results. CONCLUSION: This data-driven approach based on an unsupervised learning method enables us to more accurately characterise the over-screened population. These data invite to improve communication with the youngest women and the gynecologists to recall the benefit of an interval between two normal smears complying with the recommendations. This approach could help to improve the prevention and have a real impact on this Public Health issue.
