Geographic characteristics

Table 1 presents descriptive statistics of geographic classifications, demographic and SDOH characteristics, and COVID-19 IR for all 3,142 U.S. counties. There is no missing data. Among the 3,142 counties, 36.7% are designated as urban areas, accounting for 85.1% of the U.S. population, while the remaining 63.3% are classified as rural, representing 14.9% of the population. The 3,142 counties are divided across four Census regions: South, West, Northeast, and Midwest, each containing up to three Census divisions. The South has the highest number of counties (n = 1,422, 45.3%) followed by the Midwest (n = 1,055, 33.6%). Within these regions, the South-South Atlantic division has the largest number of urban counties (n = 354, 30.7%), followed by the Midwest-East North Central (n = 158, 13.7%) and South-West South Central (n = 156, 13.5%) divisions. The South-West South Central division also contains one of the largest numbers of rural counties (n = 314, 15.8%), alongside the Midwest-West North Central division, which has the highest concentration of rural counties (n = 511, 25.7%). These data demonstrate marked regional differences in the number of counties and their urban–rural makeup, which frame subsequent analyses of COVID-19 IR patterns across the US (map of U.S. Census Regions and Divisions).

Demographic characteristics

Demographically, Black or African American and Hispanic or Latino/a residents constitute an average of 9.0% (SD = 14.4%) and 9.6% (SD = 13.9%), respectively, of residents across all U.S. counties. The average median age across counties is 41.6 years (SD = 5.5). The mean of the median household income across counties is $55,022 (SD = $14,656). In urban counties, 12.0% (SD = 14.2%) of residents are Black or African American; the mean of the median age is 40.1 years (SD = 4.9); and the average median household income is $62,463 (SD = $17,319). In rural counties, the percentage of Black or African American residents is lower at 7.3% (SD = 14.3%); the mean median age is slightly older at 42.4 years (SD = 5.6); and the average median household income is lower $50,700 (SD = $10,706) compared to urban communities. The proportion of Hispanic or Latino/a residents is consistent between urban and rural areas.

Social determinants of health (SDOH) index

The five-factor principal component solution explained 72.6% of variability in the 17 items of the SDOH index. The first five eigenvalues were greater than 1, aligning with the number of hypothesized constructs. Supplemental Table S1 provides the Promax-rotated factor pattern, confirming that each item loaded meaningfully onto its respective component with minimal cross-loadings. The five sub-indices include Healthcare Access, Food Access, Housing and Transportation, Resource Access, and Economic Security.

The results of the confirmatory factor analysis (CFA) supported the hypothesized five-factor structure, demonstrating an adequate model fit. The Comparative Fit Index (CFI) was 0.803, the Tucker-Lewis Index (TLI) was 0.754, and the Root Mean Square Error of Approximation (RMSEA) was 0.090. Our results indicate moderate fit (conventional cutoffs for good fit: CFI and TLI ≥ 0.90, RMSEA ≤ 0.08) (Table S1). The factor loading patterns were consistent with the PCA results, with minimal cross-loading among factors. The normalized residual matrix indicated that absolute residual values remained below 0.2, suggesting the model adequately accounted for the observed data. These findings support the theoretical structure of the SDOH index, reinforcing the conceptual distinction among the five interrelated sub-domains.

For this analysis, we focus on the overall SDOH index, which demonstrated high internal reliability (Cronbach’s alpha = 0.78) and significant concurrent validity (0.37, p-value < 0.0001) with the Well-Being Index (WBI) (Supplementary Table S2). The overall SDOH index represents all 3,142 U.S. counties with a mean of 49.9 (SD = 4.5) and ranges from 34.9 to 76.3, with a higher score indicating greater resources. Urban counties had a slightly higher mean SDOH score of 51.5 (SD = 5.4) as compared to rural counties, with a mean score of 49.0 (SD = 3.6) (Table 1).

There were 315 counties that fell within the lowest SDOH percentile (< 10th) group, with a larger proportion of rural (n = 220, 11.1%) compared to urban (n = 95, 8.2%) designations. In the ≥ 10th to < 50th percentile category, which includes 1,256 counties, rural counties again make up a substantial proportion (n = 900, 45.3%) compared to urban counties (n = 356, 30.8%). Counties in the ≥ 50th to < 90th percentile group are more evenly distributed, with 1,256 counties split between 466 urban (40.4%) and 790 rural (39.7%) counties. The highest SDOH percentile group (≥ 90th), reflecting greatest access to resources, housing stability, and economic security, includes 315 counties. This group is disproportionately urban, encompassing 20.5% of all urban counties compared with only 3.9% of rural counties in the U.S.

Figure 1 illustrates the geographic distribution of the four SDOH percentile categories across the United States. Counties in the highest SDOH group (≥ 90th percentile) are generally concentrated around densely populated urban centers and along the coasts of New England and California, with some rural, less-densely populated areas in the western US. Counties in the lowest SDOH category (< 10th percentile) are predominantly located in rural, less populated areas in the Southeast Appalachian region, the northern states of Michigan and North and South Dakota, and the majority of states along the southern U.S. border.

Geographic distribution of SDOH groups across all U.S. counties (n = 3,142). Note: White areas represent unpopulated lands. Map created using urbnmapr (version: 0.0.0.9002), https://github.com/UrbanInstitute/urbnmapr.

COVID-19 incidence rates (IR)

The mean COVID-19 incidence rate (IR) across U.S. counties from March 15 – November 2, 2020, was 2,897 cases per 100,000 population (SD = 1,789), with a median of 2,627 cases per 100,000. The IR ranged widely, from 0 to 17,952 cases per 100,000, reflecting substantial variability across counties and over time (Table 1).

During Period 1 (March 15–June 14), the mean COVID-19 IR was 388 cases per 100,000 (SD = 691), with a median of 178 cases per 100,000 and a range from 0 to 12,903 per 100,000. This period had the lowest IR, indicating a relatively slower spread of the virus. In Period 2 (June 15–August 14), IR increased substantially to a mean of 805 cases per 100,000 (SD = 804), with a median of 538 cases per 100,000 and a range of 0 to 9,072 cases per 100,000. By Period 3 (August 15–November 2), the mean IR had risen sharply to 1,705 cases per 100,000 (SD = 1,275), with a median of 1,423 cases per 100,000 and a range from 0 to 14,592 cases per 100,000. Period 3 represented the highest COVID-19 rates, reflecting wider transmission. Throughout all three periods, some counties reported no COVID-19 cases, underscoring the uneven geographic spread of the virus, particularly during the early months of the pandemic (Table 1).

Growth mixture modeling (GMM)

Model evaluation and selection followed the criteria outlined in the Methods section. Five-knot GMMs provided a good balance between model fit and stability. Across the three periods, fit indices and class-size thresholds supported selection of five, four and four latent classes, respectively (Supplementary Table S3; Tables S4a–S4c; Figures S1a–S1c). For Period 1, a six-class solution produced the lowest SABIC, AIC, and BIC values, but one class contained fewer than 5% of counties; therefore, the five-class model was selected to maintain stability and interpretability. For Periods 2 and 3, although the five-class models yielded slightly lower SABIC, AIC, and BIC values, the four-class models were retained because small classes in the five-class fell below the 5% county threshold (one class in Period 2 and two classes in Period 3). Entropy ranged from 0.69 to 0.79 across the three periods, exceeding the generally acceptable threshold (≥ 0.70) for class separation in large-sample trajectory studies. The fit indices were concordant overall, supporting the final class selection. After determining the number of classes, predicted marginal trajectories (Supplementary Figures S1a–S1c) were examined to confirm the distinctiveness of the retained classes.

As noted, across the three periods, GMM identified up to five distinct trajectory classes of county-level COVID-19 incidence rates (Fig. 2). Five classes were observed in Period 1, and four in Periods 2 and 3; however, for consistency, we retained five-class labeling across periods, aligning colors and trajectory shapes. The light green line (Class 1) represents counties with the lowest IR trajectories. It appears only in Period 1, conceptually merging with Class 2 thereafter. The dark green line, labeled Class 2 in Period 1 and Class 1/2 in Periods 2 and 3, represents counties with consistently low and flat IRs. The yellow trajectory (Class 3) reflects initially low IRs followed by a sharp rise near the end of each period. The orange line (Class 4) displays a steady increase with shape variations: an inverted U-shape in Periods 1 and 2 and a U-shape in Period 3 with a late rise. The purple trajectory (Class 5) captures the highest IRs, showing an inverted U-shape in Periods 1 and 2 and a plateau at high rates in Period 3. This harmonized structure enables direct visual comparison of IR dynamics across time.

COVID-19 incidence rate trajectory classes for U.S. counties across three distinct periods in 2020 (n = 3,142). Panels A, C, and E provide U.S. County incidence rates for each trajectory class by week for the period noted. Panels B, D, and F show the geographic distribution of trajectory classes across all U.S. counties for the period noted. Class percentages indicate the proportion of U.S. counties assigned to each class within each period. Incidence rates are represented as the number of new cases per 100,000 population. Maps created using urbnmapr (version: 0.0.0.9002), https://github.com/UrbanInstitute/urbnmapr.

Period 1: March 15 to June 14, 2020

During this 13-week period, five distinct trajectory classes were identified (Fig. 2A, Table 2), consistent with the model selection criteria described above. Posterior probabilities (Supplementary Table S4a) showed high probabilities of counties belonging to their assigned class, with values exceeding 0.77 for each class. The predicted marginal trajectories (Supplementary Figure S1a) confirmed that the five retained classes captured distinct COVID-19 IR patterns within this period.

Class 1 (light green; 18.6% of counties, 1.9% of the U.S. population) and Class 2 (dark green; 43.3% of counties, 45.8% of the population) had consistently low IRs, though Class 2 was slightly higher. Class 3 (yellow; 8.9% of counties, 3.4% of the population) was low until week 11, when the incidence rate rose. Class 4 (orange; 22.0% of counties, 33.6% of the population) rose gradually and then declined, while Class 5 (purple; 7.2% of counties, 15.3% of the population) peaked around week 8 (Fig. 2a, Table 2). Over 90% of Class 1 counties were rural, whereas Class 4 contained higher proportions of urban counties. Overall, most counties experienced no or low incidence during this period (Fig. 2B). Higher IR trajectories were concentrated in the Northeast Census Region, where approximately 45% and 60% of the counties in the New England and Middle Atlantic divisions, respectively, were classified in Class 4 or 5. In contrast, only 17.5% and 9.0% of counties in the Mountain and Pacific divisions, respectively, fell into these high-trajectory classes (Supplementary Table S5).

Period 2: June 15 to August 14, 2020

During this 9-week period, four distinct trajectory classes were identified (Fig. 2C, Table 2), consistent with the model-selection criteria described above. Posterior probabilities (Supplementary Table S4b) showed high certainty of class assignment, with values exceeding 0.77 for each class, and predicted marginal trajectories (Supplementary Figure S1b) confirmed that the retained classes captured distinct COVID-19 IR patterns within this period.

Class 1/2 (dark green; 15.3% of counties, 7.9% of the U.S. population) represents counties with consistently low and stable IRs throughout the period (Fig. 2C, Table 2). Class 3 (yellow; 8.2% of counties, 1.7% of the population) began with low IRs that rose after week 5 and included a predominantly rural composition. Class 4 (orange; 40.2% of counties, 40.8% of the population) showed gradual and steady increase in IR across the period, while Class 5 (purple; 36.3% of counties, 49.6% of the population) peaked around week 6 before declining. Together, these four classes capture the nationwide shift in disease dynamics that occurred after the initial wave (Supplementary Table S5, Fig. 2D). Over 85% of counties in the South and in the East North Central Division had high-IR trajectories (Classes 4 and 5), while 88% of New England shifted toward low IR-trajectories (Class 1/2). Counties in the Pacific division, however, were largely in higher-IR classes (68%), indicating a westward shift in disease spread during this period.

Period 3: August 15 to November 2, 2020

During this 11-week period, four distinct trajectory classes were identified (Fig. 2E, Table 2), consistent with the model-selection criteria described above. Posterior probabilities (Supplementary Table S4c) showed high certainty of class assignment, with values exceeding 0.80 for each class, and predicted marginal trajectories (Supplementary Figure S1c) confirmed that the retained classes captured distinct COVID-19 IR patterns within this period.

Class 1/2 (dark green; 4.5% of counties, 2.0% of the U.S. population) represents counties with persistently low and stable IRs throughout the period. Class 3 (yellow; 6.3% of counties, 0.9% of the population) began with low IRs that rose sharply after week 6. Class 4 (orange; 76.0% of counties, 93.2% of the population) showed moderate IRs early in the period followed by steady and substantial increases, while Class 5 (purple; 13.2% of counties, 3.9% of the population) reached the highest IRs overall, peaking late in the period before plateauing.

Geographically, the highest IR trajectories (Classes 4 and 5) were concentrated in the South and Midwest, where nearly all counties in the South Atlantic (95.9%), East South Central (95.6%), and West South Central (94.7%) divisions fell into these higher-IR groups (Supplementary Table S5, Fig. 2F). Class 4 included a mix of both rural and urban counties, reflecting the broad reach of this wave across community types. High-IR trajectories were also dominant across the Midwest, with more than 90% of counties in the East and West North Central divisions classified in Classes 4 or 5. In contrast, resurgence was observed in the Pacific and New England divisions, where approximately 70% and 40% of counties, respectively, shifted into Class 4, suggesting renewed spread in areas that had experienced earlier declines.

COVID-19 trajectories and disparities

In both Periods 1 and 2, counties with the highest IR had higher proportions of Black or African American residents (Supplementary Table S6 and Fig. 3). In Period 1, counties in Class 5 had the highest percentage of Black or African American residents, with the most significant representation in the lowest SDOH group: 40.7% (SD = 24.7%) compared to Class 1 at 4.4% (SD = 11.0%). In Period 2, Class 5 counties had the largest proportion of Black or African American residents, particularly in counties within the bottom 10th percentile of SDOH, with 30.5% (SD = 24.6%) compared to Class 1/2 with 2.3% (SD = 4.0%). Period 3 showed a slightly different trend, with the highest percentage of Black or African American residents in Class 4 counties within the bottom 10th SDOH percentile with 20.4% (SD = 23.5%), compared to Class 1/2 with 5.0% (SD = 9.3%). Class 5 included counties with the second highest Black or African American representation, and these residents were most likely to fall into the bottom 10th SDOH group with 10.8% (SD = 15.0%) compared to the highest with 3.9% (SD = 5.7%). These findings remained consistent regardless of urbanicity (data not shown).

Black or African American population and age in U.S. counties by covid-19 incidence rate trajectory group and SDOH group across three distinct time periods (n = 3,142).

Across all periods, older ages were consistently associated with lower IR (Supplementary Table S6, Fig. 3). In Periods 1 and 2, median age generally declined across trajectory groups as COVID-19 incidence rates increased, with younger populations observed in higher-IR classes (Fig. 3, Supplementary Table S6). In Period 1, counties in Class 5 had the lowest median age (40.2 years, SD = 4.3), with the youngest subgroup in the ≥ 90th percentile SDOH group (38.5 years, SD = 3.9). In contrast, counties in Class 1 had a median age of 44.5 years (SD = 6.4). In Period 2, this trend persisted. Class 5 counties had a median age of 40.1 years (SD = 5.2), compared to 44.8 years (SD = 5.9) in Class 1/2. Within Class 5, the lowest median age (37.6 years, SD = 4.8) was found in the highest SDOH group (≥ 90th percentile), further reinforcing the inverse relationship between age and SDOH observed in high-IR counties. In Period 3, the trend became less consistent, though younger median ages remained concentrated in higher-IR classes. Class 5 counties had a median age of 41.8 years (SD = 6.3), while Class 1/2 counties had a higher median age of 45.1 years (SD = 7.4).

Patterns for age and income did not vary significantly by urbanicity (data not shown). No clear pattern emerged for median income or Hispanic or Latino/a ethnicity with respect to IR across all periods and classes (Supplementary Table S6, Supplementary Figure S2).

Points represent county-level means, and error bars indicate 95% confidence intervals. Trajectory group and time period definitions are provided in Table 2.