Real-World Hospital Data Suitability for Multi-Country Health Technology Assessment in Early-Stage Breast Cancer: A Feasibility Study in Six Oncovalue Centres

OBJECTIVES: Real-world hospital data have the great potential to produce rich amounts of information for health technology assessment [HTA] to inform regulatory, reimbursement, and clinical decision-making. For example, next-generation AI tools such as NLP offer the opportunity to use both structured and unstructured data. However, hospital data run the risk of being only partially complete and fit-for-purpose, leading to biases and uncertainties. Hence, the objective of this ongoing study is to assess the quality and suitability of real-world hospital data in generating real-world evidence for multi-country HTA in the clinical context of early breast cancer.

METHODS: Data suitability was assessed in six clinical centres from five European countries involved in ONCOVALUE that extracted real-world hospital data on high-risk, early-stage breast cancer patients who received neoadjuvant treatment between 2015 and 2023. Relevant data on patient characteristics, treatments, effects, costs and quality-of-life (QoL) were selected according to ESMO-GROW. Semi-structured interviews were conducted with hospital data experts to evaluate data suitability using a technical questionnaire developed following the ISPOR SUITABILITY framework.

RESULTS: Breast cancer-specific confounders and treatment effect modifiers were retrieved with varying degrees of completeness across centres and measurement timings, including ER-/PR-/HER2-status (63%-87%), tumour histology (75%-99%), tumour grade (37%-87%), and performance status (75%-99%). Well-structured and near-complete data were extracted on demographics, neoadjuvant treatment (e.g. medications, dosage) and survival. Differing QoL questionnaires were collected across 3/6 of centres. Adverse events were registered mostly unstructured. Common data modelling commenced in 4/6 of centres.

CONCLUSIONS: Our preliminary results demonstrate considerable heterogeneity in terms of availability, structure and completeness of real-world hospital data both between hospitals and variables. NLP tools may help to derive rich data from unstructured clinical notes and reports. Our results can guide the appropriate choice of statistical methodology and data harmonisation as well as data quality improvement along the hospital workflows.