OBJECTIVES: The Institute for Clinical and Economic Review’s role in Value Assessment has important implications for company planning, resourcing, and setting potential global launch impact expectations relating to price and value among stakeholders. Whilst ICER outlines priority topics for assessment on their website, these are infrequently posted, and companies therefore need to informally pre-empt potential assessment. A more formalized approach to predicting topic selection can help companies' strategic planning. This research developed a method to predict the likelihood that a therapy will be selected for a future ICER Assessment.

METHODS: Drawing on data from multiple sources, including historical ICER assessments, FDA Orange and Purple Books, and indication-specific data, we selected a Neural Network and K- Means Clustering dual approach, which provided an effective technique to predict ICER assessments. Data classification/prediction falls into one of three categories: ICER: Intervention, ICER: Comparator, or None (not assessed). To test the predictive ability of the neural network approach, 50% of the data were randomly chosen to "train" the neural network model. That test model was applied to the remaining 50% of the data to test predictive accuracy.

RESULTS: The results of the model testing via the Neural Networks approach demonstrates 86% predictive accuracy in classifying a therapy into one of the three categories (ICER: Intervention, ICER: Comparator, or None), and 72% prediction accuracy upon validation testing.

CONCLUSIONS: A determination on likelihood of assessment can help teams allocate the appropriate cross-functional staff to proactively prepare for a possible ICER engagement. Next analytics steps will involve continuing to evaluate model performance with new FDA approvals/ICER assessments and incorporate additional predictors.