Real-world evaluation of performance and practical utility of multi-biomarker risk report for cardiovascular and other chronic diseases

Introduction Identifying individuals at high risk for cardiovascular disease (CVD) and other chronic diseases is the key to most effectively use targeted preventative interventions to reduce individual disease risks and the burden on the healthcare system. In current clinical practice, risk estimation is based on calculators like SCORE2 require information from multiple sources (physical measurements and questions). Estimating risk of other diseases (e.g. diabetes or liver disease) requires separate calculators that require non-overlapping inputs. Thus multi-disease risk prediction is rarely deployed at scale. Purpose To improve the accuracy and efficacy of primary prevention in clinical practice, we developed and tested a risk report that is based on biomarkers from a single blood sample, age and sex, and estimates risks for 8 chronic diseases. Methods To identify disease risks, we utilise nuclear magnetic resonance (NMR) spectroscopy to measure 39 clinically validated metabolic biomarkers from a single blood sample. We trained multi-biomarker risk scores (including age and sex) for CVD, myocardial infarction (MI), type two diabetes, chronic kidney disease, and 2 liver and 2 lung diseases using over 60,000 samples from UK Biobank and tested score performance on a separate testing set as well as independent data from two other large biobanks. We designed a report that shows both an individual’s absolute risk and relative risk compared to others of their sex and age for all 8 diseases. To enable large-scale usage in healthcare, we implemented an automated protocol for blood sample measurements, generating and returning the risk reports, and obtained regulatory approval. During an implementation pilot in Finnish clinics we assessed the quality of returned results and collected feedback from healthcare professionals. Results For all diseases, the performance of multi-biomarker scores was similar to or better than the selected existing risk tool. E.g., for MI, multi-biomarker score resulted an AUC of 0.72 (0.71-0.73, 95% confidence interval) while the national risk calculator for MI, FINRISK 2.0, showed AUC of 0.73 (0.72-0.74). A total of 185 patients participated in the implementation pilot. The results of the quality assessment of biomarker concentrations and disease risks demonstrated a good concordance with the expected values and rates from population specific research cohorts. According to feedback obtained from the healthcare professionals, the risk report was deemed clear and understandable, as well as easy to communicate with patients. These reports have now been adopted in routine use for nation-wide occupational health screens with a Finnish private healthcare provider. Conclusions We presented a real-world application for preventative health care to assess disease risk for 8 common chronic diseases in one go and demonstrated its performance and feasibility for large-scale risk assessment in a clinical setting.