Hospitalized and nonhospitalized patients were identified from laboratory-confirmed COVID-19 cases reported to the Colorado Electronic Disease Reporting System (CEDRS) as of April 5, 2020. Based on data available in CEDRS, patients were considered eligible if they had known hospitalization status; had self-reported illness onset during March 9–31, 2020; and resided in 1 of the 9 counties (Adams, Arapahoe, Boulder, Denver, Douglas, El Paso, Jefferson, Larimer, and Weld) that account for ≈80% of Colorado’s population. March 9 was selected because it was the date on which testing for SARS-CoV-2 became more widely available in Colorado and was no longer restricted to suspected cases requiring hospitalization or having an epidemiologic link to a confirmed case, though travel to an area with ongoing community transmission was required for testing early in this period. To obtain interviews from at least 300 patients (200 nonhospitalized and 100 hospitalized), we used stratified, simple random sampling to select 600 patients (using a 2:1 ratio) from 1,738 COVID-19 cases meeting inclusion criteria.

At least 3 attempts were made to contact each selected patient on at least 2 separate days, at different times of the day, during April 10–30, 2020. For contacted patients who consented, a trained public health official administered a standardized questionnaire by telephone to obtain demographic information, verify hospitalization status and date of illness onset, and determine whether the patient had experienced any of 30 symptoms during their illness. For patients whose hospitalization status differed between CEDRS data and interview, we confirmed status using electronic medical records. For all deceased patients, minors, and persons unable to be interviewed (e.g., those with dementia), a proxy (i.e., relative or caregiver) was interviewed. Patients were asked what their first and subsequent symptoms were, and for each reported symptom, when it occurred relative to onset of illness and how long it lasted. No follow-up contact was made once the questionnaire was completed.

Data were entered into a Research Electronic Data Capture database (13,14). Frequencies and percentages were calculated and stratified by hospitalization status. We calculated odds ratios (ORs), 95% CIs, and p values to identify COVID-19 symptoms associated with hospitalization. Multivariable logistic regression was conducted to construct a model examining association of all symptoms with hospitalization status, while adjusting for demographic variables associated with hospitalization for COVID-19 (i.e., male sex, age >65 years, and Hispanic ethnicity) (Appendix Table). A reduced multivariable model was constructed by using purposeful selection to identify a subset of symptoms from the full model that had statistically significant association (15). In multivariable models, anosmia and ageusia were combined because of a high degree of collinearity; no other significant collinearity was identified.

Median and interquartile ranges (IQRs) were calculated for duration and timing of individual symptoms in relation to overall onset of illness. To account for patients who died and the large proportion of patients who were still symptomatic at the time of interview, we used survival analysis to calculate estimated median illness duration compared by hospitalization status. For participants still experiencing symptoms at interview, individual symptom duration was truncated to the date of interview because a low proportion (<10%) of patients reported individual symptoms still occurring at that time. Symptoms were categorized by organ system based on codes from the International Classification of Diseases, 10th Revision, Clinical Modification. Statistical analyses were conducted by using SAS 9.4 (SAS Institute, https://www.sas.com) and R version 3.6.3 software (https://r-project.org) (16). Significance was defined as α = 0.05, and all testing was 2-sided.

This investigation received a nonresearch determination as a public health response from human subjects advisors at the Centers for Disease Control and Prevention. The investigation was considered a public health response to a notifiable disease by the Colorado Department of Public Health and Environment.

Of 600 randomly selected case-patients, 364 (61%) completed the interview, 46 (8%) were ineligible (because onset date was before March 9 or they were asymptomatic), 57 (10%) declined to participate, and 133 (22%) were unreachable. Median age of the 364 participating patients was 50 years (range 2 months–94 years); 187 (51%) were male, 288 (79%) identified as white, and 75 (21%) identified as Hispanic. Almost all patients (345 [95%]) reported having health insurance; 128 (35%) patients were hospitalized, and 18 (5%) died. Compared with nonhospitalized patients, hospitalized patients were older and more likely to be Black; they were also more likely to be male (Table 1). Compared with patients who declined to participate or were unreachable, investigation participants resided proportionately in the same counties and had similar hospitalization rates (35% vs. 31%) and case-fatality ratios (5% vs. 8%), but they were older than nonparticipating patients (median age 50 vs. 43 years).

Among 364 participating patients, interviews were conducted with 322 (88%) patients and proxies for 42 (12%) patients. Patients who were interviewed directly reported a higher median number of symptoms (13 [IQR 9–16]) than proxies reported (6 [IQR 4–10]). Proxies were interviewed more frequently for hospitalized patients than for nonhospitalized patients (Table 1) and more often for participants >65 years of age (30/85 [35%]) than for those <65 years (12/279 [4%]). Median number of days from illness onset to interview (33 days) did not differ by hospitalization status.

Based on International Classification of Diseases, Tenth Revision, Clinical Modification, categorization of symptoms, general systemic symptoms (i.e., fever, chills, myalgia, headache, or anorexia) were commonly reported among both hospitalized (122 [95%]) and nonhospitalized (234 [99%]) patients (Table 2). Symptoms associated with potential lower respiratory tract infection (cough, dyspnea, wheezing, or chest pain) were reported by 116 (91%) of hospitalized patients and 213 (90%) of nonhospitalized patients. Cognition and perception symptoms (altered mental status, anosmia, or ageusia) were reported by 87 (68%) hospitalized and 165 (70%) nonhospitalized patients. More nonhospitalized patients (158 [67%]) reported upper respiratory tract infection symptoms (i.e., rhinorrhea, nasal congestion, or sore throat) than were reported by hospitalized patients (60 [47%]). Gastrointestinal symptoms (i.e., nausea, vomiting, diarrhea, or abdominal pain) were reported by 58% of participants, regardless of hospitalization status.

The most frequently reported symptoms were similar for hospitalized and nonhospitalized participants (Table 2). Among 128 hospitalized patients, the most commonly reported symptoms were fever (108 [84%]), fatigue (106 [83%]), cough (93 [73%]), and dyspnea (92 [72%]). Among 236 nonhospitalized patients, the most commonly reported symptoms were fatigue (213 [90%]), fever (196 [83%]), cough (196 [83%]), and myalgia (175 [74%]). Ageusia was reported by 149 (63%) nonhospitalized and 63 (49%) hospitalized patients, and anosmia by 131 (56%) nonhospitalized and 45 (35%) hospitalized patients. A total of 123 (96%) hospitalized patients and 229 (97%) nonhospitalized patients reported fever, cough, or dyspnea. Of the 12 participants not reporting these symptoms, the most commonly reported symptoms were fatigue (7 patients), anosmia (6 patients), ageusia (6 patients), anorexia (6 patients), and diarrhea (5 patients).

Participants who reported altered mental status and vomiting had at least twice the odds of being hospitalized (Table 2). Patients reporting wheezing and dyspnea also had higher odds of hospitalization. In contrast, patients who reported lymphadenopathy, anosmia, rhinorrhea, myalgia, headache, sore throat, or nasal congestion had less than half the odds of hospitalization. Patients reporting fatigue, dry cough, and ageusia also had lower odds of hospitalization.

When we controlled for all reported symptoms and characteristics included in the reduced multivariable logistic regression model, we found that participants who reported vomiting (OR 2.46 [95% CI 1.2–5.06]), dyspnea (OR 2.32 [95% CI 1.26–4.37]), altered mental status (OR 2.12 [95% CI 1.18–3.83]), dehydration (OR 1.88 [95% CI 1.1–3.26]), and wheezing (OR 1.88 [95% CI 1.03–3.43]) had higher odds of hospitalization, as did participants who were male (OR 2.13 [95% CI 1.27–3.62]) or >65 years of age (OR 3.93 [95% CI 2.16–7.27]) (Figure 1). Patients reporting rhinorrhea (OR 0.43 [95% CI 0.24–0.74]), headache (OR 0.47 [95% CI 0.27–0.82]), sore throat (OR 0.5 [95% CI 0.28–0.87]), and anosmia or ageusia (OR 0.57 [95% CI 0.33–0.96]) had lower odds of hospitalization.

The most common initial symptoms for hospitalized and nonhospitalized patients were cough (25% for hospitalized and 21% for nonhospitalized patients) and fever (25% for hospitalized and 20% for nonhospitalized patients) (Table 3). No participants reported conjunctivitis, rash, or lymphadenopathy as an initial symptom of their illness. Patients reporting sore throat as their initial symptom had lower odds of being hospitalized (OR 0.28 [95% CI 0.11–0.74]); no other initial symptom was associated with hospitalization status.

Little variation was observed between hospitalized and nonhospitalized patients in terms of symptom progression (Figure 2). Upper respiratory symptoms and general systemic symptoms were reported early in the course of disease; many patients reported these types of symptoms within 1 day of illness onset. Symptoms related to cognition, perception, and lower respiratory tract (except cough) were generally reported to occur 2–4 days after illness onset. Gastrointestinal symptoms were reported to occur ≈3–6 days after illness onset, and rash generally appeared last.

Among 346 surviving patients, 134 (39%) were still symptomatic at time of interview. The estimated median duration of illness was 18 days longer in hospitalized patients (36 days; p<0.01) than in patients who were not hospitalized (18 days; p<0.01) (Appendix Figure). The median duration of most individual symptoms was <10 days; notable exceptions were fatigue for both hospitalized (14 days [IQR 9–27 days]) and nonhospitalized (12 days [IQR 7–15 days]) participants and, among hospitalized patients, anosmia (14 days [IQR 7–24 days]), ageusia (14 days [IQR 8–21 days]), arthralgia (13 days [IQR 7–17 days]), anorexia (12 days [IQR 7–17 days]), wheezing (12 days [IQR 5–16 days]), and myalgia (11 days [IQR 7–15 days]) (Table 2). The median durations of chills, myalgia, sweats, diarrhea, arthralgia, dehydration, sore throat, abdominal pain, vomiting, and hemoptysis for hospitalized patients were >2 times those of nonhospitalized patients.