Considerations for Improving Reporting and Analysis of Date-Based COVID-19 Surveillance Data by Public Health Agencies

Abstract

More than a year after the first domestic COVID-19 cases, the United States does not have national standards for COVID-19 surveillance data analysis and public reporting. This has led to dramatic variations in surveillance practices among public health agencies, which analyze and present newly confirmed cases by a wide variety of dates.

The choice of which date to use should be guided by a balance between interpretability and epidemiological relevance. Report date is easily interpretable, generally representative of outbreak trends, and available in surveillance data sets. These features make it a preferred date for public reporting and visualization of surveillance data, although it is not appropriate for epidemiological analyses of outbreak dynamics. Symptom onset date is better suited for such analyses because of its clinical and epidemiological relevance. However, using symptom onset for public reporting of new confirmed cases can cause confusion because reporting lags result in an artificial decline in recent cases.

We hope this discussion is a starting point toward a more standardized approach to date-based surveillance. Such standardization could improve public comprehension, policymaking, and outbreak response. (Am J Public Health. 2021;111(12):2127–2132. https://doi.org/10.2105/AJPH.2021.306520)

The COVID-19 pandemic has placed pressure on public health agencies to produce and report surveillance data at an unprecedented speed and granularity.¹ While the Centers for Disease Control and Prevention publishes guidelines for collecting COVID-19 surveillance information, there is limited information on standard practices for analysis and public reporting of date-based surveillance data.^2,3 This has led to dramatic variations in reporting practices among health departments. For example, 30% of health departments use report date for visualizing new COVID-19 cases in epidemic (epi) curves, 22% use test date, 12% use symptom onset date, 16% display multiple dates such as report date and symptom onset date, and 20% do not define what dates are used or do not show epi curves.

The choices that health departments make regarding date-based analysis and reporting of COVID-19 cases have important consequences for public comprehension and trust, policymaking, and outbreak response. For example, until July 2020, all epi curves included in the Georgia Department of Public Health (GDPH) Daily COVID-19 Status Reports showed new cases by symptom onset date. Although this approach was in keeping with standard epidemiological practice,⁴ it resulted in an apparent downward trend in recent cases because of incomplete reporting of cases whose symptoms started recently.⁵ This led to public confusion and incorrect conclusions about Georgia’s progress in reducing infections in the early months of the epidemic.^6,7 Discrepancies in dates used across different reporting platforms have caused further confusion.⁶

This article discusses considerations for reporting and analysis of date-based surveillance data, using a longitudinal COVID-19 surveillance data set from GDPH as an example. This data set included 862 153 confirmed COVID-19 cases as of April 11, 2021.⁵ We hope this discussion will contribute to the development of more standardized approaches for analysis and presentation of date-based COVID-19 surveillance data across jurisdictions. Such standardization could improve public comprehension, policymaking, outbreak response, and communication.

WHICH DATES TO USE FOR SURVEILLANCE AND REPORTING

Confirmed COVID-19 cases follow a basic timeline from infection date to symptom onset date (for symptomatic individuals), test date, and, finally, report date. Figure 1 presents this timeline, along with median lags between each date from GDPH COVID-19 surveillance data and published reports. These dates are commonly used for epidemiological analysis and public reporting of routine COVID-19 surveillance data. The strengths and weaknesses of each are discussed in the following paragraphs.

FIGURE 1—

Timeline of Confirmed COVID-19 Cases for Date-Based Surveillance: Georgia

Note. DPH = Department of Public Health; IQR = interquartile range. Dates follow a timeline of infection date on the left to report date on the right, with median and IQR differences depicted between each date. The incubation period is from Lauer et al.¹⁵ All other dates are derived from Georgia DPH surveillance data as of April 11, 2021. Dates on the left are more epidemiologically relevant in that they are reflective of the occurrence and patterns of disease in individuals and populations. However, they are potentially less available and interpretable for public reporting because of large reporting lags. Dates on the right are less epidemiologically relevant but more appropriate for reporting and public presentation of surveillance data because of higher availability in surveillance data sets and minimal reporting lags.

Report Date

Report date (i.e., the date a confirmed case was reported to the health department) is generally available in surveillance data sets. For example, it is available for more than 99% of confirmed COVID-19 cases in Georgia. Report date is easily interpretable for decision-makers and the lay public because there is no need to account for reporting lags (Figure 2a). In addition, report date is usually representative of outbreak trends, especially when a running average is presented. These features make it a preferred choice for public reporting and visualization of new confirmed COVID-19 cases via epi curves, maps, and tables. This is the date that is used by the World Health Organization, Centers for Disease Control and Prevention, and many state and local health departments.^8,9 Report date is also a suitable choice for summary metrics (e.g., 14-day community transmission rates) where reporting lags associated with other dates could otherwise complicate interpretation.

FIGURE 2—

Comparisons of Epidemic Curves for Confirmed COVID-19 Cases by (a) Report Date, (b) Test Date, (c) Symptom Onset Date, and (d) Estimated Infection Date: Georgia, December 1, 2020–April 11, 2021

Note. Missing symptom onset dates were imputed using generalized additive models, and infection dates were estimated as symptom onset dates minus a median incubation period of 5.1 days. Reporting lags when < 75% of cases were likely to have been reported are shown with gray boxes on the right of each plot.

However, report date is the least epidemiologically relevant date because it is not directly related to the occurrence of disease in individuals or patterns of disease in populations. Epidemiological analyses of outbreak dynamics (e.g., contact tracing, investigations of transmission chains, cluster or hotspot analyses, estimation of the effective reproductive number) require precise characterization of temporal relationships between cases. Using report date for such analyses could introduce systematic or random error.¹⁰ Variations in reporting practices can also introduce artificial trends in surveillance data, such as a weekend drop or spikes from resolved reporting backlogs.¹¹

Test Date

Test date (i.e., the day a positive molecular test was first performed for a confirmed case) may be less biased by the previously mentioned variations in reporting practices and is more clinically relevant. Test date is available in electronic laboratory records, so it is timelier than dates requiring patient interviews. More than 98% of confirmed cases in Georgia include test date (Figure 1). For these reasons, some health departments report confirmed COVID-19 cases by test date. However, positive COVID-19 cases in Georgia have a median lag of 3 days between test date and report date, and median lags were as long as 5 days during parts of the outbreak. These lags may cause confusion when test date is used for public presentation of surveillance data because incomplete reporting results in an artificial decline in recent cases. Test date is not a recommended date for most analyses of transmission dynamics because it may not directly align with the clinical course of infections or patterns of disease in populations. In addition, many testing centers have irregular operating dates, and health care utilization often changes depending on the day of the week.¹² This could introduce artificial “weekend effects” or other irregularities in epi curves that use test date (Figure 2b).

Symptom Onset Date

Symptom onset date (i.e., the date an individual first began experiencing COVID-19‒like symptoms before a positive diagnostic test) is usually reportable on COVID-19 case report forms or collected during case interviews.¹³ Because symptomatic individuals are most infectious at or just before their symptoms begin, symptom onset date approximates the time period when individuals were most likely to infect close contacts.¹⁴ This makes it an appropriate date for epidemiological analyses related to transmission dynamics. Because symptoms generally appear a median of 5 days after infection, symptom onset date can be used to estimate infection date for contact tracing, outbreak response, or other epidemiological analyses of infection dynamics when the date of infection is unknown.^15,16

Symptom onset date is not ideal for public reporting because the inherent lag from onset to report creates an artificial downward trend in recent cases (Figure 2c). Symptom onset date also must be collected via case report forms or contact-tracing interviews, which can cause further delays in reporting. The public and decision-makers may misinterpret reporting lags as true downward trends. This can negatively affect public trust and potentially lead to misinformed policy decisions.^6,7 Unlike previous pandemics such as HIV and severe acute respiratory syndrome coronavirus 1 (SARS-CoV-1), a high proportion of SARS-CoV-2 infections are asymptomatic.¹⁷ Reporting all COVID-19 cases by symptom onset date obscures this fact and may cause confusion. In addition, public digestion of COVID-19 epidemiological data is dramatically higher than in previous pandemics.^1,18 This makes it more imperative to clearly present data and avoid potential ambiguities from dates with long reporting delays.

Limitations of using symptom onset date include underreporting, potential recall bias, and inapplicability to asymptomatic or presymptomatic individuals. This can lead to issues with reliability and a high degree of absence from surveillance data sets. In Georgia, for example, symptom onset date is only available for 55% of confirmed COVID-19 cases. Despite these limitations, various methods can be used to impute symptom onset date and assess bias. GDPH and other health departments derive symptom onset date for missing observations by using a decision tree of other available dates, and other approaches include predictive regression models and multiple imputation.^5,19

Infection Date

Infection date (i.e., when an individual was exposed to and infected with SARS-CoV-2) is the most epidemiologically relevant date because of its direct relationship to disease patterns in populations. When available, it is an ideal date for contact tracing,¹⁴ outbreak investigations,²⁰ and spatiotemporal analyses of transmission dynamics.^21,22 However, individuals often do not know the date they were infected, and it is not collected as part of most COVID-19 surveillance systems. This reduces its applicability to analyses that use routine surveillance data. It can be estimated by using symptom onset date, but many estimation methods homogenize substantial heterogeneities in incubation periods.^23,24 Long lags between infection and report dates also complicate the interpretation of epi curves that use infection date (Figure 2d). In Georgia, for example, most cases are reported more than 10 days after infection (Figure 1). Therefore, infection date is not recommended for use in public presentation of surveillance data.

PUBLIC HEALTH IMPLICATIONS

More than a year after the first domestic COVID-19 cases, the United States still does not have national standards for jurisdiction-level analysis and reporting of date-based COVID-19 surveillance data. A more standardized approach could increase public confidence and enable better harmonization of indicators and reporting methods across jurisdictions.²⁵ This is increasingly important in the current context when strengthened surveillance and reporting are required to inform reopening decisions, maintain public confidence, and rapidly detect and respond to recurrent outbreaks.

The choice of which date to use for public reporting or epidemiological analysis should be guided by a balance between interpretability and epidemiological relevance. In general, report date is preferable for use in public reporting, while symptom onset date is the best choice for many epidemiological analyses of transmission dynamics when infection date is not available. Some health department dashboards, such as that of GDPH, now give users an option to view epi curves by date of report and by date of symptom onset. This may reduce public confusion, but additional efforts are needed to identify best practices for reporting and visualization of surveillance data.^5,18