All Times ET
Keywords: clinical risk prediction, R Shiny, visualization, validation
Clinical risk prediction models are commonly developed in a post-hoc and passive fashion, capitalizing on convenient data from completed clinical trials or retrospective cohorts. Impacts of the models often ends at the publication rather than with the patients. Fortunately, the field of clinical risk prediction is rapidly improving in today’s progressive data science era. Based on collective experience over the past decade by the Prostate Biopsy Collaborative Group (PBCG), this paper proposes the following four data science-driven strategies for improving clinical risk prediction to the benefit of clinical practice and research:#1 Actively design prospective data collection, monitoring, analysis and validation of risk tools using the same standards as for clinical trials, #2 Post risk tools and model formulas online to maximize doctor-patient decision-making and multiple external validation, #3 Dynamically update risk tools and tailor to individual clinical centers to adapt to changing demographic and clinical landscapes, #4 Accommodate systematic missing data patterns across heterogeneous cohorts for model training and provide flexible online tools that allow missing information. The data science, statistical and informatics methods behind these strategies are illustrated using the PBCG experience over the past decade, concluding with recommendations for application in all fields of clinical risk prediction.