Comparing health insurance and survey data in estimating prevalence of chronic diseases

Abstract Background Population prevalence of chronic conditions can be estimated from national health surveys and from administrative data sources such as insurance records. This study evaluated the agreement between the Belgian Health Interview Survey (BHIS) and the Belgian compulsory health insurance data (BCHI) in ascertaining chronic hypertension, hypercholesterolemia and diabetes in Belgium. Methods The most recent cycle of BHIS (2018) provided the self-reported prevalence of diabetes, hypertension, and hypercholesterolemia among a representative sample of Belgian adults. For BCHI, the chronic conditions were attributed for every individual in the BHIS reviewing the medication prescription records identified using the ATC/DDD system. These two data sources were linked through unique identifiers by STATBEL. Disease prevalence, measures of agreement, and measures of concordance were estimated. Logistic regression was performed to determine the factors affecting agreement between BHIS and BCHI’s disease classifications. Results Data linkage was done for 9,753 individuals aged 15 years and older. From the sample, BHIS and BCHI respectively identified 5.9% and 5.6% diabetes cases, 18% and 24% of hypertension cases, and 18% and 16% of hypercholesterolemia cases. The kappa coefficient between BCHI and self-reported diabetes, hypertension, and hypercholesterolemia was 0.79, 0.59, and 0.49, respectively. Gender, age, and subjective health significantly affected the agreement in chronic condition classification between BHIS and BCHI. Conclusions Data on reimbursed drugs is a potential alternative method in the surveillance of chronic diabetes. This procedure could be used in estimating disease prevalence but further validation is needed to evaluate its applicability and bias in other chronic conditions. Key messages • BCHI is a possible alternative data source for the surveillance of diabetes in the population. • BCHI overestimated hypertension and underestimated hypercholesterolemia prevalence.


Background:
Population prevalence of chronic conditions can be estimated from national health surveys and from administrative data sources such as insurance records. This study evaluated the agreement between the Belgian Health Interview Survey (BHIS) and the Belgian compulsory health insurance data (BCHI) in ascertaining chronic hypertension, hypercholesterolemia and diabetes in Belgium.

Methods:
The most recent cycle of BHIS (2018) provided the selfreported prevalence of diabetes, hypertension, and hypercholesterolemia among a representative sample of Belgian adults. For BCHI, the chronic conditions were attributed for every individual in the BHIS reviewing the medication prescription records identified using the ATC/DDD system. These two data sources were linked through unique identifiers by STATBEL. Disease prevalence, measures of agreement, and measures of concordance were estimated. Logistic regression was performed to determine the factors affecting agreement between BHIS and BCHI's disease classifications. Results: Data linkage was done for 9,753 individuals aged 15 years and older. From the sample, BHIS and BCHI respectively identified 5.9% and 5.6% diabetes cases, 18% and 24% of hypertension cases, and 18% and 16% of hypercholesterolemia cases. The kappa coefficient between BCHI and self-reported diabetes, hypertension, and hypercholesterolemia was 0.79, 0.59, and 0.49, respectively. Gender, age, and subjective health significantly affected the agreement in chronic condition classification between BHIS and BCHI.

Conclusions:
Data on reimbursed drugs is a potential alternative method in the surveillance of chronic diabetes. This procedure could be used in estimating disease prevalence but further validation is needed to evaluate its applicability and bias in other chronic conditions.

Background:
Considering the growing prevalence of chronic disease and diabetes mellitus (DM) in Belgium, alongside population aging, insight into the economic burden of DM is essential for decision makers. To the best of our knowledge, there is no research on the subject in Belgium. Thus, our aim was to estimate the direct and indirect costs associated to DM in Belgium between 2013 and 2017.

Methods:
On a first phase, we performed a retrospective observational study, calculating the direct (i.e., ambulatory care, hospitalizations and medications) and indirect (work absenteeism, by multiplying mean daily wage and days absent from work) costs in the Belgian population with DM in 2013-2017. Data was retrieved from the Belgian Intermutualistic Agency (which manages compulsory health insurance) database and the Belgian Health Interview Survey database, namely DM prevalence, healthcare costs, days absent from work and sociodemographic and health factors. Subsequently, negative binomial regression models were used to assess the association of mean yearly costs to DM and adjustments for age, education level, physical activity, sugared drink consumption and bodymass index were included. Mean incremental costs were estimated through recycled predictions, considering the observed DM prevalence in Belgium in the study period and a counterfactual scenario with null prevalence.

Results:
We found a direct mean yearly incremental cost of E2 477 per DM patient, in Belgium, associated with age, low educational level and low physical activity. In the total Belgian population, the total yearly incremental healthcare cost of DM was E1.5 billion. Indirect yearly incremental cost of DM resulted to be not significantly different from the population without DM. Conclusions: DM has a major economic burden in Belgium, one that is expected to continue to rise in the future, alongside population aging. These results are essential for health planning and resource allocation. Key messages: DM has a major economic burden in Belgium, especially when it comes to direct health expenditures with ambulatory care, hospitalizations and medications.
Considering the growing prevalence of DM and population aging, these results are essential for health planning and resource allocation.