OBJECTIVES: Cost-effectiveness of using a machine learning-based risk identification tool for Alzheimer's disease (AD) was evaluated from the perspective of Finnish public payer. The aim was to find out how accurate the real-world data-based AD risk prediction model should be when it is used on unselected persons over 65 years of age in connection with the use of social and healthcare.

METHODS: The cost-effectiveness of using the AD risk prediction model was evaluated by means of a state transition model with different accuracy (50–90%) and usage time scenarios (1–5 years before current practice diagnosis) and compared to a situation where the risk prediction model was not used. Based on the risk prediction model, enhanced follow-up was started for those at high risk of developing AD, thanks to which AD could be diagnosed in time in more cases and a new innovative drug treatment could be started, potentially slowing down the course of AD. The time horizon of the model was 15 years, and a 3%/year discounting was used. Probabilistic and one-way sensitivity analyses were done.

RESULTS: Using the risk prediction model produced more quality-adjusted life years at lower costs in almost all scenarios. Also, in those scenarios where the costs were higher with the use of the risk prediction model, risk prediction was highly likely to be cost-effective. Cost-effectiveness improved because of the better accuracy of the risk prediction model. The biggest uncertainty drivers were the costs of using the risk prediction model and the success in follow-up for making earlier diagnoses.

CONCLUSIONS: The use of the AD risk prediction model has a very high probability of being cost-effective in an unselected population of users of social and health services over 65 years of age. Future studies should investigate the actual benefits of earlier diagnosis after risk prediction.