Report on final analysis of results

TASK 35 Clinical Analysis of study results and future recommendations M4M12 UMCU Julius Clinical AVA RocheThere are multiple research questions included within this project Appropriate statistical methods will be selected for each question depending upon the nature of the data distribution as well as the effect that is to be characterised Cohort 3 provides a reference group towards Cohorts 1 and 2 for Device Signs and symptoms diary vs Signs and symptoms diary alone comparisons Cohorts 1 and 2 also allow for withinsubject comparisons of the machine learning models with and without the inclusion of the Device data into the models Comparisons against Cohort 3 allow for the actual impact of recommendations coming from Device Signs and symptoms diary vs Signs and symptoms diary alone Comparisons against Cohort 3 also allow for the assessment of the Device impact on compliance to the signs and symptoms diary data entry Comparisons within Cohorts 1 and 2 allow for assessing the impact of the Device on the predictiveness of the machine learning with all other things being equal The study will contain two epochs as relates to the machinelearning approach for the recommendation algorithms The first epoch is the learning phase during which the ML models will be developed ML models for Cohorts 1 and 2 can consider baseline data signs and symptoms diary data and Device data ML models for Cohort 3 can consider baseline data and signs and symptoms diary data but do not have Device data The second epoch is the confirming phase during which the operational characteristics of the recommendation algorithms will be formally assessed During the confirming phase it is expected that all subjects will be provided realtime recommendations The algorithm may continue to adapt during this epoch but realtime recommendations should not be turned off during this period unless is it is determined to be ineffective and the study discontinued During the learning phase there may be periods during which the subjects provide data but the signs and symptoms diary does not yet indicate a recommendation The recommendation feature could be turned on and off multiple times during the learning period The status of this functionality will be clearly indicated to the subjects at all timesThe performance of the recommendation algorithm during the confirming phase of the study will be assessed using sensitivity specificity and AUC The serology status from end of followup for each subject still alive at that time will be used as the gold standard For subjects not alive at end of followup the status of available viral load data will be used positive at any time and for those without viral load data physician diagnosis will be accepted provided there was a supportive lung CT scan used All other subjects will be summarized for transparency but excluded from the estimation for sensitivity specificity and AUC The sensitivity specificity and AUC will be presented once a sufficient number of subjects have reached end of followup as well as for each month prior using the end of followup status to assess the evolution of the recommendation algorithm over time These statistics will be provided within Cohort 1 with and without Device data Cohort 2 with and without Device data and Cohort 3 for comparative purposes The withinsubject comparisons within Cohort 1 and Cohort 2 will be used to assess the added predictive value of the Device The differences between Cohort 1Cohort 2 vs Cohort 3 will be used to assess the impact on realtime recommendations whether to visit their healthcare provider and the impact of the Device on signs and symptoms diary compliance The calculation of differences will reweight the subjects so that Cohort 1Cohort 2 has a similar risk profile as Cohort 3Additional statistical metrics will be provided to support assessment of the clinical outcomes