Rapid Identification of High-Risk Participants in a COVID-19 Research Study: How Evidation Enables Precision Recruitment at Scale

March 7, 2022

Today, Evidation is sharing a recent method development paper published as a preprint on MedRxiv demonstrating precision recruitment in the context of COVID-19. In it, study authors, led by Aziz Mezlini, share how Evidation recruitment into a COVID-19 detection study yielded up to 7 times higher rates of infection than observed in actual vaccine trials using traditional recruitment methods. 

Evidation researchers, along with researchers from the Biomedical Advanced Research and Development Authority (BARDA) were able to 1) directly recruit subjects for a COVID-19 study via a mobile application they were already using, and 2) create an AI risk model that allows targeted recruiting of persons most likely to become infected during the study,  while also recruiting a representative cohort. All variables used in the risk model were either publicly available (such as geographic hot spots) or directly permissioned by research participants for that use.

Studies developing screening tools, diagnostics, and preventative treatments, such as vaccines, must observe a sufficient number of infection events in a short time. If researchers can identify those most likely to get sick beforehand, it becomes feasible to observe more of these events faster and with fewer participants, which will have profound implications for accelerating research and bringing life-saving innovations to individuals faster.

A Built-In Candidate Pool

Evidation is a health reward platform that encourages its members to develop healthy habits such as walking, meditating, and logging meals. It also incentivizes members to participate in research by completing surveys and sharing data from commercially available wearable sensors. The app has been used since 2017 to monitor annual waves of influenza infections and has more than four million users to date.  

“With Evidation, recruitment can be done very quickly and in a decentralized way,” noted Mezlini. “Participants don’t have to travel to a medical center. Testing kits can be mailed and surveys can be taken as frequently as needed using a mobile phone. This capability can be essential to a rapid response to crises such as COVID-19 and to speedy enrollment of the necessary number of participants for a trial.”

The researchers sent an initial risk survey to members of Evidation, asking them about possible exposure to COVID-19 through their jobs. They then sent a second survey to 94,700 members to identify who had received a diagnosis of COVID-19 since their initially negative response. Of the 66,040 users who responded to the second survey, 514 (0.8%) had developed COVID-19 in the interim.

Modeling an Infection

The team then created a machine learning model to identify the factors that best predicted COVID-19 infection among the 514 infected people, taking into account the local prevalence of infection according to the Centers for Disease Control and Prevention (CDC). Risk scores mostly depended on the number of potentially risky contacts (household size and residential situation), location (living in a city with numerous COVID-19 cases at the time of recruitment), and working in a risky occupation (such as healthcare jobs).

The model was then used to calculate an up-to-date infection risk score for each of the 128,629 people who had responded to the first survey, and those with the highest risk scores were targeted for recruitment. In all, 840 people were selected based on risk score and balanced for age, sex, and ethnicity. 

After adjustment for the different study periods, the incidence rates of COVID-19 based on the model were up 7 times higher than the actual incidence rate seen in the control arms of three vaccine trials. The incidence rate was also much higher than that observed in another Evidation-generated COVID-19 study that did not use targeted enrollment. 

Incidence of COVID-19 in targeted modeling vs. previous vaccine trials. Incidence attained through precision recruitment (average between men and women = 5.47 IRR) is 7x higher than what observed in the Moderna control arm (0.781) and 4x higher than NIH-funded ILLNESS study, run by Evidation without using precision recruitment. 


“When analyzing the behavioral variables the algorithm considered important, we found some expected risk-increasing factors such as practicing less social distancing and going out to bars and indoor gyms,” Mezlini stated. 

Risk factors for COVID-19 in targeted modeling. Importance scores and coefficients from univariate logistic regression of most relevant variables for the model


“One surprising finding came up when we asked participants about their perceived risk of COVID-19,” noted Mezlini. “The algorithm deemed that participants answering that they were at ‘no risk’ of contracting COVID-19, were actually at higher risk than participants answering they were at ‘low’, ‘moderate’, or ‘high risk’.”


This precision recruitment approach could be used to reduce the sample size of future studies by 4 to 7 times, or shorten their duration by the same factor. “Beyond reducing the cost and time required for clinical trials, this approach also could be valuable in situations of emergency such as during a pandemic,” added Mezlini.

-----

To learn more about Evidation's offerings or to schedule a call with our experts, click here.