Population-based screening programs are typically evaluated using a participation rate, which is calculated as the proportion of the eligible target population that completes a screening test during a defined round. But before a person can be counted as a non-participant, program organizers must first determine whether they are even eligible. Individuals who have already undergone a colonoscopy, who are under active surveillance for colorectal disease, or who have been diagnosed with certain serious illnesses should be medically excluded from the denominator. If these exclusions are under-ascertained, the denominator remains artificially inflated, and the reported participation rate is mechanically deflated. Conversely, departments that manage to identify more medically ineligible individuals will report higher participation rates, even if the actual number of people screened is identical. It is this mechanical link between exclusion ascertainment and participation estimates that the French research team, led by Akoï Koïvogui of the Université Sorbonne Paris Nord and colleagues, set out to quantify.

The challenge is that the tools available for identifying medically ineligible individuals vary dramatically from one part of France to another. Screening programs can draw on three principal external data sources: primary health insurance data, which capture reimbursements for procedures and treatments; hospital morbidity data, which record hospitalizations and diagnoses; and cancer registry data, which provide detailed information on diagnosed malignancies. Not every department has enjoyed the same access to these provider data, and not every department has used them with the same consistency. Between 2010 and 2023, only 21 of the 60 departments studied obtained data from a medical data provider institution, and they did so for an average of just 6.8 years within the study window. This patchwork of data access creates what the authors describe as heterogeneity in exclusion practices, and it raised a critical methodological question: how much of the variation in reported participation rates across France reflects genuine differences in screening uptake, and how much reflects differences in the ability to refine the eligible population?

To answer this question, the researchers conducted a retrospective ecological study using aggregated screening-round data from Santé publique France, the national public health agency, combined with information from a 2021 national survey documenting the exclusion practices of departmental screening management structures. Their analytical strategy had two components. First, they used mixed-effects models to assess the association between the duration of access to each of the three external data sources and the crude medical exclusion rate. Second, because participation is mechanically influenced by exclusion ascertainment, they analyzed a complementary outcome: the screened fraction of the total target population, which is immune to changes in the denominator. This dual approach allowed them to separate statistical artifacts from real changes in screening behavior.

The findings were striking. Longer use of primary health insurance data was associated with higher medical exclusion rates, with a regression coefficient of 0.35 and a 95 percent confidence interval of 0.24 to 0.47. Longer use of hospital morbidity data showed a similar direction of effect, with a coefficient of 0.14 and a confidence interval of 0.04 to 0.24. Cancer registry data, by contrast, showed no independent association with exclusion rates, suggesting that registries alone contribute little to the identification of medically ineligible individuals in the French context. In practical terms, departments with sustained access to insurance and hospital data were systematically better at refining their screening denominators than departments without such access.

The consequences for screening indicators followed directly. Departments with provider-data access reported higher participation rates than departments without it, but part of this difference turned out to reflect more complete refinement of the eligible population rather than greater screening uptake. When the team examined the screened fraction of the target population as a complementary outcome, the difference between departments was considerably more limited. In other words, some of the apparent gap in screening performance between French departments may have been an illusion produced by unequal data infrastructure rather than a real difference in how many people actually completed their screening tests.

To push the analysis further, the researchers turned to model-based counterfactual projections, using Monte Carlo methods to estimate how screening indicators might have evolved under two hypothetical scenarios of more sustained data access. The first scenario assumed uninterrupted use of provider data until 2023 in the departments that had used them at any point between 2010 and 2023. The second, more ambitious scenario assumed continuous provider-data use from 2010 onward in all departments. Under the second scenario, the projections suggested that more sustained access to provider data would have increased the reported medical exclusion rate by 6 percentage points and the reported participation rate by 2 percentage points. These are not trivial shifts. A two-point rise in reported participation, achieved purely through better denominator refinement, could meaningfully alter how departments are ranked, how resources are allocated, and how the program as a whole is judged against European targets.

The study’s implications extend well beyond France’s borders. Organized screening programs throughout Europe, and indeed around the world, depend on accurate denominators to measure performance, and many face comparable fragmentation in the availability of medico-administrative data. The French experience illustrates a subtle but important form of measurement bias: when the infrastructure needed to exclude medically ineligible individuals is unequally distributed, the resulting indicators are not comparable across jurisdictions, and programs serving populations with equal screening behavior may appear to perform very differently. This matters for policy planning, because participation rates are often used to identify underperforming regions, justify interventions, and evaluate the effect of program changes. If part of the observed variation is an artifact, interventions may be targeted at the wrong places, and genuine improvements may go unrecognized.

The authors argue that equitable access to provider data and the adoption of standardized exclusion algorithms based on the Système national des données de santé, France’s national health data system, would substantially improve program evaluation. The SNDS, which links insurance claims, hospital discharge records and other medico-administrative sources at the national level, offers a route to uniform exclusion ascertainment that would remove the geographic lottery currently embedded in the program’s statistics. The study was conducted as part of the continuous evaluation process of the colorectal cancer screening program and received approval under the framework of the French National Commission for Informatique and Civil Liberties, with all data anonymized before analysis and processing steps compliant with the European General Data Protection Regulation. The research did not receive specific funding from any public, commercial or non-profit agency, and the authors declare no competing interests.

What emerges from this work is a clear message for the growing movement toward data-driven public health: the quality of the data infrastructure is not a technical afterthought but a determinant of what programs can honestly see. As countries invest in screening for colorectal and other cancers, ensuring that every region has the same ability to identify who should and should not be counted may be as important as any improvement in invitation strategies or test logistics. Until then, the apparent shortfall in France’s screening participation—and possibly in other countries’—may partly be a story about data pipelines rather than about patients.

Subject of Research: Counterfactual analysis of how heterogeneous access to medical exclusion data sources affects participation estimates in France's population-based colorectal cancer screening program.

Article Title: Impact of medical exclusion data sources on population-based colorectal cancer screening round participation in France: a counterfactual analysis

Article References: Koïvogui, A., Balamou, C., Benamouzig, R., Mullaert, J., & Duclos, C. (2026). Impact of medical exclusion data sources on population-based colorectal cancer screening round participation in France: a counterfactual analysis. BMC Health Services Research. https://doi.org/10.1186/s12913-026-15499-7

Image Credits: AI Generated

DOI: 10.1186/s12913-026-15499-7

Keywords: colorectal cancer screening, France, participation rate, medical exclusion, counterfactual analysis, SNDS, health services research, population screening, Monte Carlo method, screening program evaluation, public health, cancer screening