Methods
Data Source
The case investigation form (CIF; Appendix 1, https://wwwnc.cdc.gov/EID/article/27/9/20-4577-App1.pdf) is a supplemental questionnaire designed by CDC in January 2020 to collect detailed demographic and epidemiologic information about a convenience sample of US COVID-19 case-patients reported by participating states. This purposive nonprobability sample was selected at the state level from persons identified through care-seeking, surveillance, or contact tracing as having COVID-19; infection with severe acute respiratory coronavirus 2 (SARS-CoV-2) was confirmed by reverse transcription PCR. CDC provided guidance for selection of case-patients across a range of ages and symptom severities (i.e., hospitalized and nonhospitalized), but states individually controlled sampling. The CIF was completed by state or local health department personnel or by CDC staff through case-patient or proxy interviews, along with medical record reviews (when relevant).
Case-patient demographic information included age, sex, race, ethnicity, and occupation. Workplace settings were classified according to 2012 census industry codes (Appendix 2, https://wwwnc.cdc.gov/EID/article/27/9/20-4577-App2.pdf). Clinical information included underlying conditions, symptoms, symptom onset date, dates of medical visits, and outcome (death or survival). For hospitalized case-patients, information was requested about whether the patient had been admitted to an intensive care unit, whether oxygen was received, admission and discharge dates, diagnosis, and location. Questions about exposure included whether in the 14 days before illness onset the case-patient had known exposure to a case-patient with laboratory-confirmed COVID-19 (COVID-19 contact) and, if so, the relationship and setting of the exposure. Case-patients were also asked about their exposure risks (activities and possible exposures in the 14 days before illness onset) including travel; friends, acquaintances, co-workers, or family members with fever or respiratory symptoms; close contact with (e.g., caring for, speaking with, or touching) any ill persons; attendance at a mass gathering (e.g., religious event, concert, sports event); public transportation use; attendance or work at a school or daycare; school or daycare attendance by household members; close contact with a contact of a laboratory-confirmed case-patient; close contact with someone with fever, acute respiratory illness, or both who had traveled internationally in the previous 14 days; and time in a healthcare setting as an employee, patient, or visitor.
The CIF also collected data on the case-patient's household members, defined as anyone who stayed overnight in the same residence as the case-patient during the 14 days before the case-patient's illness onset until the date of interview. Case-patients were asked for household members' age, sex, relationship to the case-patient, and whether each person had "experienced fever or respiratory symptoms (e.g., cough, sore throat, etc.) within 14 days before or after the COVID-19 patient's illness"; if yes, date of illness onset was collected. When the CIF was designed in January 2020, the most commonly reported COVID-19 signs and symptoms were fever and respiratory symptoms, and guidance for mitigation measures within households had not been widely distributed.
Analysis of Exposures
We compared exposures between those reporting known close contact with a COVID-19 case-patient in the 14 days before illness onset and those reporting no known contact. Categorical variables were compared by using χ 2 or Fisher exact tests, as appropriate. Continuous variables were compared by using t tests for normally distributed data and Wilcoxon rank sum tests otherwise. p<0.05 was considered significant. Analyses were conducted in SAS version 9.4 (https://www.sas.com) and R (https://www.r-project.org).
Analysis of Presumed Household Transmission
We separately assessed presumed household transmission by using information about household members provided by the interviewed COVID-19 case-patient (CIF subject). In the absence of SARS-CoV-2 testing data for all household members, we used reported signs and symptoms (i.e., fever or respiratory symptoms) as a proxy for symptomatic COVID-19 infection (i.e., household transmission). We analyzed households of ≥2 members (including the CIF subject) if the CIF subject had experienced ≥1 symptom (to enable identification of the first ill person [index case-patient] in the household), and symptom status was provided for ≥1 other household member. We required that the earliest symptom onset date in the household be ≥1 calendar day before symptom onset in subsequent case-patients (to limit effect of co-exposures outside the home) and that the earliest onset date in the household be ≥3 days (our median serial interval) before the interview (to allow time for symptoms to develop in exposed household members). We considered presumed household transmission to have occurred if ≥1 household member, in addition to the CIF subject, was reported as having fever or respiratory symptoms. The person with the earliest symptom onset date in a household was considered the index case-patient, regardless whether SARS-CoV-2 testing had been performed. Any members of a given household not identified as the index case-patient are hereafter referred to as household contacts.
We calculated the overall household attack rate for symptoms as the number of symptomatic household contacts divided by the total number of household contacts with reported symptom status, with Wilson score 95% CI, and the serial interval as the time from symptom onset in the index case-patient to first symptom onset in a household contact. We investigated age and sex of the index case-patients and their contacts, household size, and relationship of the contact to the index case-patient as possible correlates of contact symptom status by using generalized estimating equation logistic regression with households as the cluster and individual symptom status as the outcome; we used an exchangeable correlation matrix and robust SEs. We excluded household contacts missing symptom status from this analysis. We examined models for collinearity and reduced if necessary. We did not include hospitalization status of the index case-patient in models because of collinearity with index case-patient age. We dichotomized contact age (<18 or ≥18 years) to avoid collinearity with familial relationship and index case-patient age.
To explore the validity of using reported symptom status to estimate household symptomatic attack rates, we calculated sensitivity and specificity by using a subset of households for which complete reverse transcription PCR and serologic testing data were available.[8] We conducted a sensitivity analysis by reclassifying data according to a range of plausible misclassification rates (Appendix 2).
Emerging Infectious Diseases. 2021;27(9):2323-2332. © 2021 Centers for Disease Control and Prevention (CDC)
