Objectives: The purpose of this study was to assess the feasibility of the adoption of a machine learning (ML) algorithm in support of the investment decisions regarding high cost medical devices based on available clinical and epidemiological evidence. Methods: Following a literature search, the set of epidemiological and clinical need predictors was established. Both the data from The Central Statistical Office and The National Health Fund were used. An evolutionary algorithm (EA) model was developed to obtain the prediction of the need for CT scanners across local counties in Poland (hypothetical scenario). The comparison between the historical allocation and the scenario developed by the EA model based on epidemiological and clinical need predictors was established. Only counties with available CT scanners were included in the study. Results: In total, over 4 million CT scan procedures performed across 130 counties in Poland between 2015 and 2019 were used to develop the EA model. There were 39 cases of agreement between historical data and hypothetical scenarios. In 58 cases, the EA model indicated the need for a lower number of CT scanners than the historical data. A greater number of CT procedures required compared with historical use was predicted for 22 counties. The remaining 11 cases were inconclusive. Conclusions: Machine learning techniques might be successfully applied to support the optimal allocation of limited healthcare resources. Firstly, they enable automatization of health policy making utilising historical, epidemiological, and clinical data. Secondly, they introduce flexibility and transparency thanks to the adoption of ML to investment decisions in the healthcare sector as well.
Keywords: CT; access to healthcare; health economics; health policy; machine learning; retrospective database analysis.