Harmonizing heterogeneous endpoints in COVID-19 trials without loss of information - an essential step to facilitate decision making

Background: Many trials are now underway to inform decision-makers on potential effects of treatments for COVID-19. To provide sufficient information for all involved decision-makers (clinicians, public health authorities, drug regulatory agencies) a multiplicity of endpoints must be considered. It is a challenge to generate detailed high quality evidence from data while ensuring fast availability and evaluation of the results. Methods: We reviewed all interventional COVID-19 trials on Remdesivir, Lopinavir/ritonavir and Hydroxychloroquine registered in the National Library of Medicine (NLM) at the National Institutes of Health (NIH) and summarized the endpoints used to assess treatment effects. We propose a multistate model that harmonizes heterogeneous endpoints and differing lengths of follow-up within and between trials. Results: There are currently, March 27, 2020, 23 registered interventional trials investigating the potential benefits of Remdesivir, Lopinavir/ritonavir and Hydroxychloroquine. The endpoints are highly heterogeneous. Follow-up for the primary endpoints ranges from four to 168 days. A detailed precisely defined endpoint has been proposed by the global network REMAP-CAP, which is specialized on community-acquired pneumonia. Their seven-category endpoint accounts for major clinical events informative for all decision-makers. Moreover, the Core Outcome Measures in Effectiveness Trials (COMET) Initiative is currently working on a core outcome set. We propose a multistate model that accommodates analysis of these recommended endpoints. The model allows for a detailed investigation of treatment effects for various endpoints over the course of time thereby harmonizing differing endpoints and lengths of follow-up. Conclusion: Multistate model analysis is a powerful tool to study clinically heterogeneous endpoints (mortality, discharge) as well as endpoints influencing hospital capacities (duration of hospitalization and ventilation) simultaneously over time. Our proposed model extracts all information available in the data and is - by harmonizing endpoints within and between trials - a step towards faster decision making. All ongoing clinical trials, especially those with severe cases, should accommodate primary analysis with a stacked probability plot of the major events mechanical ventilation, discharge alive and death.