Existing case surveillance definitions for COVID-19, as shown in Table 1, were generally sensitive in our study conducted among household contacts of infected persons, a population with proven SARS-CoV-2 exposure.

However, they tended to have low specificity and poorly estimated disease prevalence. By systematically screening novel definitions that optimized sensitivity, specificity, and PPV, we improved community prevalence estimation and overall accuracy of individual screening, which could be useful if diagnostic testing is limited. In particular, we affirmed loss or dysfunction of taste or smell as a uniquely discerning characteristic central to constructing an effective, concise case surveillance definition when applied across all age groups (i.e., derived compound combination 3).

An appropriate discriminatory balance between sensitivity and specificity for a newly emerging pathogen depends on the objectives of the surveillance activity.²⁰ Highly sensitive case definitions capture a larger proportion of true COVID-19 cases, which is ideal when diagnostic testing is widely available and results are timely. Highly sensitive definitions, however, generally rule in a larger number of non-cases (i.e., FP symptom screens).²¹ In addition to testing resources, the public health system’s tolerance for false-positive screens is, of course, dependent on human resources. This is especially apparent when intensive interventions involve extensive contact tracing, isolation and quarantine. At high community COVID-19 prevalence, these intensive mitigation efforts may benefit from evidence-based prioritization. By example, CSTE combination 2 had a FP symptom screening rate (77/136; 57%) eight times that for derived compound combination 3 (10/136; 7%) in our cohort. At the population level, such differences could expose shortcomings in resources for core interventions, such as universal contact tracing. The COVID-19 response has repeatedly been strained in these requisite areas.^22-24 Novel symptom screening criteria that more tightly couple sensitivity and specificity (i.e., diagnostic accuracy), such as the derived compound combination 3, could help to prioritize interventions when strategically deployed. This principle may also apply when evaluating novel vaccines or therapeutics in large clinical trials involving thousands of participants, where feasibility constraints often dictate the use of symptom-prioritized testing to confirm outcomes. Still, highly sensitive symptom rules, such as CSTE combination 2, are preferred for COVID-19 when resources are unlimited.

For using syndromic surveillance systems to estimate community burden, the highly sensitive existing case definitions overestimated true burden. Conversely, highly specific case definitions, such as ILI, may detect changes in disease trends over time but underestimate true burden.²⁰ ILI underestimated disease prevalence by more than 80% in this study population. Current laboratory-based surveillance grossly under-ascertains incidence,²⁵ especially where diagnostic testing is not easily accessible or widespread. Retailoring community-based syndromic surveillance systems already in place²⁶ (i.e., altering the symptoms included or applying a correction factor based on results such as ours) would more accurately reflect true burden.

For most symptoms and their combinations, overall performance, most notably sensitivity, differed between children and adults. These findings are consistent with prior observations whereby children generally show fewer and milder symptoms of COVID-19 compared with adults,²⁷ and COVID-19 syndromes vary across ages.¹³ The small number of children with COVID-19 in this cohort limits the conclusion of specific recommendations, but further examination into the utility of age-specific case definitions is warranted in considering policies for schools and other child congregate settings, and for deriving accurate burden estimates from syndromic surveillance.

While the number of individuals in this study is relatively small, our investigation is among the largest and most well-characterized of its kind to date. We collected extensive symptom data, which yielded a comprehensive assessment of multiple symptom combinations. We also employed inclusion criteria that were not based on disease status or symptom status, and a reference category based on standardized laboratory testing. Nonetheless, we acknowledge this study’s limitations. These analyses were not intended to produce definitive symptom combinations to be applied to the general public, however they may be used to guide the development of future candidate case definitions. One key consideration for future validation efforts is that enrollment started immediately after the precipitous decline in laboratory-confirmed influenza virus infections in the United States in mid-March 2020.²⁸ Thus, our estimates of diagnostic performance may differ during the viral respiratory season. In addition, COVID-19 prevalence was higher for our study population (i.e., contacts of laboratory-confirmed household members) compared to the entire community, thereby limiting the generalizability of predictive values (although sensitivity and specificity remain unaffected by disease prevalence). Screening criteria applied to persons seeking medical care may also perform differently, as those individuals probably tend to have more severe illness.

Additionally, we showed that existing COVID-19 case definitions are highly sensitive and do well to screen in persons for testing and individual-level public health interventions like community isolation. In the first such endeavor for evaluating and deriving novel COVID-19 case surveillance definitions in a community setting among SARS-CoV-2—exposed individuals with largely mild illness, we evaluated novel symptom combinations for COVID-19 using methodology previously applied to tuberculosis in low resource settings.¹⁹ These derived combinations and CSTE definition 2 better estimated community disease burden and used taste and/or smell dysfunction as a primary component. The latter is supported by prior studies.^5,8-10 Because most SARS-CoV-2 infections are mild²⁹ and core public health functions may need prioritization when testing and other resources are limited, case definitions that accurately determine COVID-19 status in the general public may assist continued interruption of community transmission.³⁰ When timely diagnostic testing is readily available, however, using less sensitive screening tools could inappropriately miss cases and lead to further community transmission.

Our study population, which includes persons across the age spectrum and enrolled participants independent of disease and symptom status, may better reflect the diagnostic performance in the general population than previously published research. Accurate clinical case definitions are likely to also apply to large clinical trials for candidate vaccines and therapeutics where serial confirmatory SARS-CoV-2 testing for any new symptom is impractical. It is important that our results be validated against the growing body of larger ambulatory surveillance databases in diverse communities and in other countries; in particular, our methodology should be assessed in the context of the annual influenza season, at varying community COVID-19 prevalence, and across the age spectrum. Such studies ideally can be accompanied by cost-effectiveness modeling of intervention strategies.