OBJECTIVES: Innovative medicines are defined as the medicines that contain an active substance or combination of active substances that have not been authorised before. Conducting health technology assessment (HTA) in silico may inform manufacturers of portfolio prioritisation and evidence generation strategy. This study aims to identify whether artificial intelligence algorithms may predict reimbursement decision for innovative drugs in Scotland.

METHODS: All appraisals for innovative medicines in Scottish Medicines Consortium (SMC) from 2016 to 2020 were screened to extract decision outcomes (accepted, accepted for restricted use and not recommended) and 24 explanatory variables that may drive the decisions. Univariable analysis was conducted to identify the statistically important variables based on the P value ≤ 0.1. Six machine learning classifiers including decision tree, multivariable logistic regression model, random forest, support-vector machine, Xgboost and K-nearest neighbours were used to build prediction models with the identified important variables.

RESULTS: A total of 111 appraisals were identified, among which, 47 were accepted, 48 accepted for restricted use and 16 not recommended. 14 of 24 explanatory variables were selected through the univariable analysis and used in the prediction models. The prediction models shown that indication restriction by manufacturer, uncertainty of economic evidence, validation of primary outcome and acceptance of comparator were the most important drivers of SMC decision-makers. Four of six models had good prediction performance with the accuracy and F1-score over 0.9. The model with best prediction performance was decision tree with accuracy and F1-score of 0.96.

CONCLUSIONS: This pilot study shown that artificial intelligence algorithms may be used to predict reimbursement decisions and support portfolio management and evidence generation. More research is requested to improve prediction accuracy and expand to other HTA bodies.