key: cord-0759541-gvkiqn6l
authors: George, Stephanie M.; Chen, Haiying; Miller, Michael E.; Rejeski, W. Jack; Stowe, Cynthia L.; Webb, Christopher; Kraus, William E.; Musi, Nicolas; Jakicic, John M.
title: Rapid Report on Using Data to Make Standardized Decisions about Enrollment during the COVID-19 Pandemic: Perspectives from the MoTrPAC Study
date: 2021-05-03
journal: Ann Epidemiol
DOI: 10.1016/j.annepidem.2021.04.016
sha: 2a78489fdf2f9e44ee59806d9dbe8e8c44bca3d6
doc_id: 759541
cord_uid: gvkiqn6l

nan

The emergence and persistence of Coronavirus Disease 2019 in the United States has presented challenges for ongoing, large clinical and observational research studies that involve inperson components and measurements. To facilitate continuous operations, a need has arisen to develop systems to use COVID-19 incidence data to make standardized decisions about starting and restarting enrollment across multiple clinical sites. However, incidence rates based on raw county-level data can be dramatically higher than actual values. We report here on an approach to a) centrally monitor COVID-19 incidence rates using available, rolling epidemiological data, b) identify solutions for addressing data limitations during monitoring when data are available from multiple imperfect data sources each with important differences in methodology, scope, and quality, and c) use the results for informing temporary suspension or restarting of individual clinic activities.

The Molecular Transducers of Physical Activity Consortium (MoTrPAC), the largest investment in physical activity research to date by the National Institutes of Health, was established to elucidate how exercise improves health and ameliorates diseases by building a map of the molecular responses to acute and chronic exercise. 1 Suspension or restarting of MoTrPAC recruitment and in-person clinic activities is based on thresholds of 14-day county-level incidence rates. When the 14-day average incidence rate exceeds 10 cases per 100,000 population, COVID-19 testing is required before all in-person study visits; when incidence rates meet or exceed 30 cases per 100,000 population, new recruitment is halted. We developed dynamic reports utilizing a publicly available dataset updated daily and released by the New York Times through GitHub (link) to rapidly monitor the daily incidence rates of new COVID-19 cases for counties with potential participants that may be recruited by each clinical site. Depending on the clinical site and state, the targeted recruitment catchment areas ranged from one to eight counties. During the process, it became evident that there were periodic data dumps due to electronic lab reporting on backlogs of old cases. Thus, we adopted an autoregressive integrated moving-average (ARIMA) modeling approach for active monitoring of the new COVID-19 cases. ARIMA modeling has been widely used for analyzing univariate time series. 2 It can identify additive outliers (AO) in the response series that are not accounted for by the estimated model.

Briefly, each Friday after the New York Times updates their data, the process involves sending to each site county-specific reports that contain both the daily new cases with AOs flagged and corresponding 14-day averages for incidence rates. Depending on where the 14-day averages fall (< 10/100k, ≥ 10/100k and < 30/100K, or ≥ 30/100k), the sites may choose not to investigate the potential outliers and just follow usual rules for clinical operations, or the sites may choose to investigate the reason for the spikes with the local health department. If spikes are found to be related to backlog of cases, evidence of data dumps with written justification is sent back to the Data Coordinating Center. If some of the cases in the dump occurred within the recent 14-day span (i.e., sometimes sites can only ascertain that the dump contained cases from a date range), all of the cases are included in the 14-day average. Only if all of the cases in the data dump are at least two weeks old, they are removed from the 14-day average and the report for the clinical site is revised. Based on the new 14-day average incidence rate (over same time period), the sites then follow the usual rules for clinical operations and determine if their operations for the upcoming week need to change from the previous week. The study protocol was changed to incorporate the above procedure.

In this paper, we use the report for Friday 10/23 for the clinic site in Bexar County, Texas as an example to illustrate this process. Figure 1 shows the time series of reported daily new cases for Bexar County/San Antonio through Thursday, 10/22/2020. For illustrative purposes, 10 most statistically significant AOs across the series identified by an ARIMA (0,1,1) model were indicated as * on the graph. We focused on the outliers from 10/9-10/22, the 14-day data that were used to generate the current rate estimate to guide clinical operations for the upcoming week. The site reviewed the outliers and verified with the Bexar County/San Antonio Metropolitan Health District that a backlog of 2,583 cases was added to Bexar's cumulative case count on October 18 in addition to the cases that were identified on that date. Further investigation revealed that all the backlog cases occurred more than 14 days prior, thus they were excluded from the 14-day average and the rate was revised to 7.5 cases per 100,000 population. Figure  2 shows that if this correction were not made, the 14-day average would be skewed to 16.7 cases per 100,000 population and the clinical operation for the upcoming week would be impacted due to the crossing of the threshold of 10 cases per 100,000 population. This would also affect the subsequent Friday report on 10/30.

The COVID-19 pandemic has brought unforeseen challenges to conducting in-person research, and there are still many unknowns ahead. The method we have applied can aid consortia in systematically identifying outliers due to delays in uploading of COVID-19 tests, not actual evidence of increase in community spread. A larger backlog than the one observed in Bexar County, or a similar backlog in a county with a higher rate, could raise the rate enough to bring recruitment from that county to a halt in our study. Our weekly centralized reporting on the data allow for clinical sites to be more aware of the rise and fall in rates and allow them to follow the pre-specified procedures related to the ongoing recruitment and enrollment of individuals to the study. We view this process as an exploratory aid to help the clinical sites manage operations. During a pandemic, in-person research necessitates comprehensive public health mitigation measures, including use of personal protective equipment, to keep participants and staff safe. Our work highlights the additional value of building novel analytic approaches for monitoring and understanding community spread during current and future infectious disease outbreaks.

Molecular Transducers of Physical Activity Consortium (MoTrPAC): Mapping the Dynamic Responses to Exercise

Time Series Analysis: Forecasting and

This work from the MoTrPAC Study is supported by NIH grants, U01AR071133, U01AR071130, U01AR071124-01, U01AR071128, U01AR071150, U01AR071160, U01AR071158 (Clinical Centers), U24AR071113 (Consortium Coordinating Center).

All authors contributed equally. 

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.