Care management is seen as a promising approach to address the complex care needs of patients with multimorbidity. Predictive modeling based on insurance claims data is an emerging concept to identify patients likely to benefit from care management interventions. We aimed to identify and explore patterns of multimorbidity in primary care patients with high predicted risk of future hospitalizations in order to develop a primary care-based care management intervention. We conducted a retrospective cohort study to assess insurance claims data of 6026 patients from 10 primary care practices in Germany. We stratified the population by the predicted likelihood of hospitalization (LOH) using a diagnostic cost group-based case-finding software. Co-occurrence of chronic conditions in multimorbid patients with an upper-quartile LOH score was explored by extraction of mutually exclusive patterns. Predictive modeling identified multimorbid elderly patients with a high number of co-occurring chronic conditions (mean number 7.8 [SD 3.1]). Assessing co-occurrence of highly prevalent chronic conditions in 1407 multimorbid patients with upper-quartile LOH revealed 471 mutually exclusive patterns with low single frequencies. The observed prevalence significantly exceeded expected prevalence for patterns with causal comorbidity. Additionally, chronic pain (related to osteoarthritis) or depression could be identified as discordant co-occurring conditions in 80% (12/15) of the most common multimorbidity patterns. High-risk primary care patients suffer from heterogeneous individual patterns of co-occurring chronic conditions. Care management interventions will have to account for discordant co-occurring conditions such as osteoarthritis and depression.