OBJECTIVES: Socially deprived areas are characterized by poorer health status and health outcomes and higher mortality compared to more affluent areas. Additionally, there is “misplaced demand” of patients seeking medical care for social problems. This can lead to increased inadequate use of health care services, resulting in increased healthcare cost. One approach to addressing these problems is to create an integrated range of care services. A much-discussed approach to solving these problems are care models that create a cross-sectoral and integrated range of services. These models are seen as having the potential to improve access and quality of care by integrating medical and social care. However, the evidence on the success of integrated care models is heterogeneous and previous approaches are characterized by methodological limitations that restrict the robustness of the results.

METHODS: Our analyses are based on administrative data of three German sickness funds for the years 2015-2019 containing 556.145 observations. We focus on the outcome dimensions healthcare utilization and costs for the empirical setting of an integrated care model in a socially deprived area in Germany. We conduct difference-in-differences analyses of propensity score matched groups. The propensity score is estimated using a Superlearner machine learning approach.

RESULTS: The results based on 10.000 randomly selected observations show that the methodological approach works. Sensitivity analyses demonstrate the robustness of the methodology. With regard to the effects of the integrated care model, no significant changes in health care utilization can be detected. With regard to costs, a slight increase can be observed.

CONCLUSIONS: This study addresses existing methodological limitations. By means of using machine learning methods, the study enhances evaluation methods of integrated care models. In the next step, the model is applied to the full sample.