Long COVID Prevalence among U.S. Adults: A State-level Ecological Analysis of the Contribution of COVID-19 Incidence, Severity of Acute Illness, COVID-19 Vaccination, and Chronic Conditions

This ecological analysis of U.S. state-level data reveals that while SARS-CoV-2 incidence, hospitalization rates, and multimorbidity vary across regions, only higher vaccine coverage and greater infection incidence are independently associated with Long COVID prevalence, highlighting vaccination as a critical factor in mitigating long-term impacts.

The Gist

Imagine the United States as a giant patchwork quilt made of 50 different states. Each patch has its own unique pattern, color, and texture. For a long time, we've known that some patches of this quilt have more people suffering from "Long COVID" (lingering symptoms after getting sick) than others, but we didn't quite understand why the patterns were so different.

This study is like a detective trying to figure out what makes some patches of the quilt "sicker" than others. The researchers looked at four main ingredients that might be mixing together to create these differences:

1. How many people got sick (SARS-CoV-2 incidence).
2. How many people ended up in the hospital (Severity).
3. How many people got vaccinated (Vaccine coverage).
4. How many people had other health issues like diabetes or obesity (Chronic conditions).

The Investigation: Mixing the Ingredients

The researchers gathered data from a massive national health survey (like a giant census of health habits) and combined it with official government records on infections and vaccines. They treated each state as a single data point, asking: "If we look at the whole state, do these four ingredients explain why Long COVID is high or low here?"

Think of it like baking 48 different cakes (one for each state). The researchers wanted to know: Is the cake sweeter because you used more sugar (more infections)? Is it denser because you used more flour (more chronic conditions)? Or is it lighter because you used a special leavening agent (vaccines)?

The Big Reveal

When they looked at the ingredients one by one, it seemed like a few things mattered. States with more people having chronic health issues seemed to have more Long COVID. States with higher vaccination rates seemed to have less.

But here is the twist: When they put all the ingredients into the mixing bowl at once to see how they interacted, the picture changed.

1. The Vaccine Shield: Vaccination remained a powerful force. States with higher vaccination rates had significantly less Long COVID. It's like having a strong umbrella; even if it rains (infection), you stay drier. The study found that boosting vaccination is a direct way to shrink the Long COVID problem.
2. The Infection Rain: The number of new infections still mattered, but only a little bit. More infections meant a slightly higher chance of Long COVID, but it wasn't the only driver.
3. The "Silent" Factors: Surprisingly, when everything was considered together, hospitalization rates and chronic health conditions (like having three or more diseases) stopped being the main stars of the show.

Why Did the Other Factors Disappear?

You might wonder, "But I thought having diabetes or going to the hospital made Long COVID worse!" And you're right—at the individual level, that's true. If you are sick and have diabetes, you are at higher risk.

However, at the state level, these factors are like shadows cast by the same light.

• States with many people having chronic conditions often also have lower vaccination rates.
• States with high hospitalization rates often also have high infection rates.

When the researchers adjusted for everything, it turned out that vaccination and infection rates were the "master switches." The other factors (like chronic disease) were important, but their effect was already being explained by how many people got infected and how many were vaccinated. It's like trying to figure out why a car is moving; you might say "the wheels are turning," but the real cause is the engine. In this case, the engine is vaccination and infection control.

The Takeaway for Everyone

The main lesson from this study is simple: To stop Long COVID, we need to stop the virus from spreading and keep people vaccinated.

While having other health issues makes the ride harder for individuals, the best way to fix the problem for the whole country is to focus on two things:

1. Keep the virus from spreading (lowering the "infection rain").
2. Keep the "umbrella" open (high vaccination rates).

The study concludes that if states focus their energy on these two levers, they can significantly reduce the burden of Long COVID for their communities. It's a reminder that public health is a team effort, and the most effective tools we have are prevention and protection.

Technical Summary

1. Problem Statement

Long COVID (symptoms persisting ≥3 months post-infection) is a significant public health burden in the United States, yet the drivers of its geographic variation across states remain poorly understood. While individual-level risk factors (e.g., severity of acute illness, vaccination status, comorbidities) are known, there is a lack of integrated, population-level data quantifying how state-level differences in these factors correlate with Long COVID prevalence. The study aims to fill this gap by determining whether state-level variations in SARS-CoV-2 incidence, acute illness severity (hospitalizations), vaccination coverage, and multimorbidity explain the observed heterogeneity in adult Long COVID prevalence.

2. Methodology

Study Design:
An ecological study using the U.S. state as the unit of analysis.

Data Sources & Variables:

• Outcome: State-level crude prevalence of self-reported Long COVID among adults (aged ≥18) derived from the 2023 Behavioral Risk Factor Surveillance System (BRFSS). Data collection extended into early 2024. The final dataset included 48 states (Kentucky and Pennsylvania were excluded due to insufficient data).
• Predictors (Analytic Window: Oct 1, 2022 – Sep 30, 2023):
  1. SARS-CoV-2 Incidence: Cumulative new cases per person-year (CDC Response Data).
  2. Acute Illness Severity: Cumulative adult COVID-19 hospitalization rates per 100,000 persons (National Healthcare Safety Network).
  3. Vaccination Coverage: Weighted average of monthly bivalent booster coverage among adults who completed the primary series (NIS-Adult COVID Module).
  4. Multimorbidity: Proportion of adults with ≥3 chronic conditions (e.g., diabetes, obesity, heart disease), derived from CDC Center for Forecasting and Outbreak Analytics (CFA) estimates.
• Covariates: Age and sex distribution variables (proportion of adults ≥65, proportion of males ≥65, proportion of females <65) to adjust for demographic differences.

Statistical Analysis:

• Uncertainty Propagation: To account for sampling error in BRFSS prevalence estimates, the authors generated 1,000 Monte Carlo draws from a normal distribution based on state means and standard errors.
• Modeling: Multivariable linear regression models were fitted to each of the 1,000 datasets.
• Pooling: Estimates were combined using Rubin's rules to produce pooled point estimates and total variances, incorporating both within- and between-simulation variability.
• Significance: Defined as $p < 0.05$.

3. Key Contributions

• First Systematic Ecological Analysis: This is the first study to systematically investigate the association between state-level COVID-19 metrics (incidence, severity, vaccination, comorbidities) and adult Long COVID prevalence across the entire U.S.
• Methodological Rigor: Unlike many ecological studies, this analysis explicitly propagated the uncertainty inherent in survey-based prevalence estimates (BRFSS) through the regression modeling using Monte Carlo simulations and Rubin's rules.
• Temporal Alignment: The study carefully aligned the analytic window (Oct 2022–Sep 2023) with the definition of Long COVID (symptoms lasting 3+ months) to ensure temporal relevance.

4. Results

Descriptive Statistics:

• Long COVID prevalence varied significantly by state, ranging from 4.8% (Maryland) to 9.7% (West Virginia), with a median of 6.6%.
• All predictor variables (incidence, hospitalization, vaccination, multimorbidity) showed marked interstate heterogeneity.

Univariable Regression:

• Vaccine Coverage: Significantly negatively associated with Long COVID prevalence ($\beta = -0.13, p < 0.001$).
• Multimorbidity: Significantly positively associated with Long COVID prevalence ($\beta = 0.13, p = 0.001$).
• Incidence & Hospitalization: Not significantly associated in univariable models ($p > 0.05$).

Multivariable Regression (Adjusted for Age/Sex):

• Vaccine Coverage: Remained a strong, significant negative predictor ($\beta = -0.11, p < 0.001$). Higher booster coverage correlated with lower Long COVID prevalence.
• SARS-CoV-2 Incidence: Became a significant positive predictor after adjustment ($\beta = 0.36, p = 0.013$). Higher incidence correlated with higher prevalence.
• Multimorbidity: Lost statistical significance ($\beta = 0.07, p = 0.07$), suggesting its effect may be mediated through vaccination coverage, severity, or demographics.
• Hospitalization Rates: Remained non-significant after full adjustment.

Uncertainty Analysis:

• Results from the Monte Carlo simulations (incorporating BRFSS standard errors) were consistent with the point-estimate models, confirming the robustness of the findings regarding vaccine coverage and incidence.

5. Significance and Implications

• Primary Drivers: State-level variation in Long COVID burden is most strongly driven by vaccination coverage (protective) and SARS-CoV-2 incidence (risk factor).
• Public Health Policy: The findings reinforce that promoting COVID-19 vaccination (specifically boosters) and preventing infection are critical strategies for mitigating the long-term burden of Long COVID at the population level.
• Ecological Nuance: The lack of association between hospitalization rates and Long COVID at the state level (despite individual-level evidence) highlights the complexity of ecological data. This may be due to the exclusion of institutionalized populations in BRFSS or the confounding effect of antiviral treatments in hospitalized patients.
• Limitations: As an ecological study, causal inference is limited. The study relies on self-reported data and aggregate metrics, which may not capture individual-level lag times or specific variant impacts.

Conclusion:
The study concludes that while chronic conditions and acute severity are individual risk factors, at the state level, the prevalence of Long COVID is primarily a function of how many people are getting infected and how many are vaccinated. Public health interventions should prioritize increasing vaccination coverage and reducing transmission to lower the national Long COVID burden.