Based on Electronic Health Record data collected in the North Karelia region in Finland from 2012 to 2016, we built a non-causal augmented naïve Bayesian (ANB) network model to examine the directional relationship between 16 risk factors and the costs of care for a random cohort of 363 AUD patients. Jouffe’s proprietary likelihood matching algorithm and van der Weele’s disjunctive confounder criteria (DCC) were used to calculate the direct effects of the variables, and sensitivity analysis with tornado diagrams and analysis maximizing/minimizing the total cost of care were conducted.

The highest direct effect on the total cost of care was observed for a number of chronic conditions, indicating on average more than a €26,000 increase in the 5-year mean cost for individuals with multiple ICD-10 diagnoses compared to individuals with less than two chronic conditions. Remission had a decreasing effect on the total cost accumulation during the 5-year follow-up period; the percentage of the lowest cost quartile (42.9% vs. 23.9%) increased among remitters, and that of the highest cost quartile (10.71% vs. 26.27%) decreased compared with current drinkers.

The ANB model with application of DCC identified that remission has a favorable causal effect on the total cost accumulation. A high number of chronic conditions was the main contributor to excess cost of care, indicating that comorbidity is an essential mediator of cost accumulation in AUD patients.