Risk of Potentially Preventable Hospitalizations After SARS-CoV-2 Infection

Key Points Question Is infection with SARS-CoV-2 associated with an increased risk of potentially preventable hospitalization, and if so, how long does this association persist after infection? Findings In this cohort study of 1 132 220 US veterans enrolled in the Veterans Health Administration, veterans with SARS-CoV-2 had 3 times greater risk of potentially preventable hospitalization than matched comparators without SARS-CoV-2 within 30 days after infection and more than 40% greater risk at 1 year. Meaning These findings suggest that the persistently higher risk of potentially preventable hospitalization among veterans with SARS-CoV-2 infection may reflect difficulty meeting postinfection ambulatory health care needs in the broader context of the pandemic.


Introduction
2][3][4] They are also increasingly used as a measure of health system performance during public health emergencies. 5][8][9][10][11][12] Messaging early in the pandemic encouraging people to stay home when sick, as well as postinfection sequelae, may have compounded these risks among individuals with SARS-CoV-2. 13However, studies show potentially preventable hospitalizations declined in the US during the first year of the pandemic. 6,7However, with only contemporaneous pre-COVID-19 era comparison periods or groups, 6,7 these studies were not designed to detect the effect of SARS-CoV-2 on potentially preventable hospitalizations.
[16][17] Understanding the influence of SARS-CoV-2 infection on potentially preventable hospitalizations is necessary for evaluating whether patients' postinfection health care needs are being met.We examined the association, during varying follow-up periods, between SARS-CoV-2 and the risk of potentially preventable hospitalization in Veterans Health Administration (VHA) facilities, VHA-purchased community care, and Medicare fee-for-service care among a national cohort of veterans enrolled in the VHA, the largest integrated national health system in the US.We additionally characterized the association between SARS-CoV-2 and potentially preventable hospitalizations during 3 pandemic waves and by selected clinical and sociodemographic factors.

Methods
This cohort study was approved by the institutional review boards of the Ann Arbor, Michigan; Durham, North Carolina; Palo Alto, California; Portland, Oregon; and Puget Sound, Washington, VHA medical centers, which waived the informed consent requirement as this was a retrospective study of extant data in accordance with 45 CFR §46.This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Target Trial Approach and Study Participants
We emulated nested monthly sequential trials to assess the risk of potentially preventable hospitalization among VHA patients with SARS-CoV-2.Data from the Veterans Affairs (VA) COVID-19 Shared Data Resource (CSDR) 18 were used to identify SARS-CoV-2 diagnoses between March 1, 2020, and April 30, 2021, a period during which health care facility-based testing was common and self-testing was rarer.The CSDR integrates a variety of data sources to provide patient-level information on VHA facility-, non-VHA facility-, and community-detected infections. 18ce and ethnicity were self reported by participants and were categorized as American Indian or Alaska Native, Asian, Black, Hispanic, Native Hawaiian or Other Pacific Islander, White, and Emulating the balance achieved through randomization (eTable 1 in Supplement 1), we a priori identified potential confounders associated with SARS-CoV-2 and clinical, functional, and economic outcomes. 19We then used VA electronic health record data to identify and create matched monthly cohorts composed of similar veterans with and without (hereinafter termed comparators) SARS-CoV-2 during the same month as their matched veteran's documented infection (index date).

JAMA Network Open | Infectious Diseases
Matched cohorts were created using a combination of exact matching and propensity score matching with replacement.Veterans without a history of SARS-CoV-2 in a given month could be matched to more than 1 veteran with SARS-CoV-2 and were eligible for matching in subsequent months until infected.Our matching procedure included 5 exact-matched variables and 37 propensity score-matched variables, including baseline sociodemographic, clinical, and health care use variables (eTable 2 in Supplement 1). 20terans were included if they had used VHA primary care during the 2 years prior to the index date or were enrolled in a VHA Patient-Aligned Care Team service.15 days prior to their CSDR-documented infection were also excluded to avoid misclassification.For this analysis, we included up to 5 comparators nearest in propensity score to their matched veteran with SARS-CoV-2; 0.5% of the SARS-CoV-2 cohort had at least 1 but fewer than 5 matched comparators (eFigure 1 in Supplement 1 for a study flow diagram).

Measures
Our primary exposure was SARS-CoV-2 infection, described above.2][23] We identified potentially preventable hospitalizations using Agency for Healthcare Research & Quality prevention quality indicators (PQIs), which are defined based on the principal diagnosis of a hospital stay and have been validated for identifying hospitalizations for ACSCs. 1 Outcomes included PQI overall and acute and chronic composites (eTable 3 in Supplement 1). 24The overall composite consists of the chronic composite (eg, diabetes with short or long-term complications, chronic obstructive pulmonary disease) and acute composite (eg, bacterial pneumonia, urinary tract infection).In primary analyses, hospitalizations that occurred on the same date as the index date or death were counted as outcomes.
Baseline sociodemographic, clinical, and health care use variables (eTable 2 in Supplement 1) were used to match veterans with SARS-CoV-2 to comparators and control for potential confounding in subgroup analyses.Additional variables used to identify strata for subgroup analyses included COVID-19 pandemic wave of index date (March 1 to June 30, 2020, July 1 to November 31, 2020, and December 1, 2020, to April 30, 2021), 25 baseline Elixhauser rehospitalization risk score, 26 residence in a primary care professional shortage area, 27 and hospitalization at index date.Primary care professional shortage area data were obtained from the Health Resources & Services Administration. 27Variables for Medicare Advantage status at index date and long-term institutionalization (ie, inpatient stays >180 days) 3 during the 2 years prior to or 1 year after the index date were used to identify veterans for inclusion or exclusion in sensitivity analyses.

Statistical Analysis
Data were analyzed from May 10, 2023, to January 26, 2024.To assess exchangeability on observable data and descriptively examine outcomes between SARS-CoV-2 and comparator groups, we calculated frequencies and proportions, means and SDs, and standardized mean differences (SMDs). 28,29Cumulative incidence plots were used to assess overall trends in time to first potentially preventable hospitalization, with crossover infection among comparators and death treated as censoring events.
For our primary analysis, which we refer to herein as per-protocol 1 (PP1), we used extended Cox regression models 30 and a stratified baseline hazard specification based on matched group to estimate the mean hazard ratio (HR) of overall, acute, and chronic composite potentially preventable hospitalizations during 4 cumulative follow-up periods: 0 to 30, 0 to 90, 0 to 180, and 0 to 365 days.
We treated non-PQI-based (ie, nonpreventable) hospitalizations as a time-varying covariate, 31 defined as a binary variable equal to 1 during intervals when a veteran was hospitalized and otherwise 0. Veterans who had not yet experienced an outcome or another censoring event were right censored at the end of each follow-up period.Crossover infection among comparators and death among infected and comparators were treated as censoring events.We used Hubert-White robust SEs to account for clustering on matched group.
We conducted a series of sensitivity analyses for the overall composite outcome.First, potentially preventable hospitalizations that occurred on the index date were eliminated, as these cumulative intervals, we estimated models with discrete follow-up periods: 0 to 30, 31 to 90, 91 to 180, and 181 to 365 days.For this analysis, individuals no longer at risk for an outcome due to previous events, crossover infections, or death were excluded.Additionally, matched groups without at least 1 at-risk veteran with SARS-CoV-2 and 1 at-risk comparator were excluded.Last, we tested a second form of censoring, which we refer to herein as per-protocol 2 (PP2), in which we censored the entire matched group at the time of a crossover infection among comparators; deaths were censored individually as in PP1.
Using our PP1 approach, we conducted exploratory subgroup analysis based on age group (<65, 65-85, and Ն85 years) and sex (male or female).In addition, using our PP1 approach but without a stratified baseline hazard specification, we conducted subgroup analyses based on COVID-19 pandemic wave (March 1 to June 30, 2020, July 1 to November 30, 2020, and December 1, 2020, to April 30, 2021), baseline Elixhauser rehospitalization risk score tertile, baseline residence in a primary care professional shortage area (yes or no), and hospitalization at index date (yes or no).In subgroup models, we adjusted for variables with SMDs greater than 0.10, as some variables used for matching were less well balanced across strata.For our hospitalization-at-index subgroup analysis, outcome events that occurred on the same day as index date were excluded.
Statistical significance was determined a priori at α = .05(2-sided).All analyses were conducted in R, version 4.3.1 (R Project for Statistical Computing).

Potentially Preventable Hospitalizations
Overall, 3.10% of cohort members (3.81% of veterans with SARS-CoV-2 and 2.96% of comparators) had a potentially preventable hospitalization during 1-year follow-up (Table 2), with fewer than 1% of cohort members having more than 1.The cumulative incidence of potentially preventable hospitalizations grew more rapidly among veterans with SARS-CoV-2 than comparators during the first 90 days after the index date, with trends becoming similar with longer follow-up (Figure 1).c Race and ethnicity data from the VHA electronic health record are collected through self-identification either at enrollment or at a health care encounter.
d Scores are centered around 1. A value of 1 indicates that the veteran is expected to have annual health care costs that are the national average for VHA patients; a score higher than 1, the veteran has an expected annual health care cost that is higher than that of the average VHA patient; and a score lower than 1, the veteran has an expected annual health care cost that is lower than that of the average VHA patient.Overlapping score categories are mutually exclusive.
e Higher scores indicate higher risk for hospitalization and/or mortality within the next 90 days.
f Scores range from −2 to 22, with higher scores indicating higher comorbidity burden.
g Higher scores indicate higher risk for rehospitalization within 1 year.Overlapping score categories are mutually exclusive.

Exploratory Subgroup Results
For all subgroups we examined with 1 exception (0-30 day follow-up period for those not hospitalized at index date), risk of a potentially preventable hospitalization was greater among veterans with SARS-CoV-2 than among comparators in all follow-up periods, with some differences in the magnitudes of association (Figure 3 and eTable 6 in Supplement 1).For example, regarding pandemic waves, differences in risk between SARS-CoV-2 and comparator groups were larger during wave 1 (eg, 0-to 30-day AHR, 3.74 [95% CI, 3.17-4.40])compared with wave 2 (eg, 0-to 30-day AHR, 2.74 [95% CI, 2.46-3.05]).Differences in risk between SARS-CoV-2 and comparator groups were also larger among those who resided in primary care shortage areas (eg, 0-to 30-day AHR, 3.37 [95% CI, 2.99-3.79])compared with those not in a primary care shortage area (eg, 0-to 30-day AHR, 3.08 [95% CI, 2.87-3.30]).After excluding events that occurred on the same day as the index date, among those hospitalized at index, risk of a potentially preventable hospitalization was greater among veterans with SARS-CoV-2 than comparators (eg, 0-to 30-day AHR, 4.95 [95% CI, 4.35-5.62]).However, among those who were not hospitalized at the index date, differences in risk were not significant during the 0-to 30-day period.Veterans who were hospitalized at the index date had twice the overall incidence of potentially preventable hospitalization at 1-year follow-up than those who were not hospitalized at index (hospitalized: 4.69%, not hospitalized: 2.62%) (eTables 7-12 in Supplement 1 provide subgroup characteristics.)

Discussion
In this cohort stuy using an emulated target trial design in the largest integrated health system of the US, approximately 1 in every 25 veterans with SARS-CoV-2 from March 1, 2020, to April 30, 2021 (3.81%), had a potentially preventable hospitalization in the year that followed, compared with approximately 1 in every 33 comparators (2.96%).Both groups had more hospitalizations for exacerbations of chronic conditions (eg, diabetes, asthma, heart failure) than for acute conditions (ie, bacterial pneumonia, urinary tract infection).During the 30-day follow-up, the risk of a potentially preventable hospitalization was more than 3 times higher among veterans with SARS-CoV-2 than comparators.Differences in risk became attenuated with longer follow-up time.However, when assessed through 1 year, veterans with SARS-Cov-2 had a 44% increased risk of a potentially preventable hospitalization relative to comparators.Expanding on prior studies reporting a positive association between SARS-CoV-2 and all-cause hospitalization, 14,15,17 our results suggest that some postinfection hospitalizations need not occur.With day 0 includes potentially preventable hospitalizations on day 0 (index date), and without day 0 excludes these.There is substantial overlap in cumulative incidence and 95% CIs for comparators with and without day 0 hospitalizations.Cumulative incidence individually censors for death and crossover infection.
Although results are exploratory, subgroup analyses suggest that suboptimal access to ambulatory care after infection increases the risk of a potentially preventable hospitalization among individuals with SARS-CoV-2.For example, the risk of a potentially preventable hospitalization was more pronounced for infections occurring during the earliest (March to June 2020) compared with latest (December 1, 2020, to April 30, 2021) pandemic wave, which represent a period of sweeping disruptions to care delivery, [9][10][11][12] as well as in primary care professional shortage areas.However, additional research is needed to uncover causal mechanisms for these findings and to evaluate whether and for whom improved ambulatory care access may mitigate the risk of postinfection hospitalization.With a relaxing of telehealth regulations and increased payment parity during the COVID-19 pandemic, 32 it will also be important to understand whether shifts from in-person to telehealth care have affected veterans' risk of preventable hospitalization after SARS-CoV-2.In a recent analysis composed of a cohort similar to the one from our overall study, Hebert et al 33     shortage area indicates veteran residence.All subgroup analyses controlled for nonoutcome hospitalizations.In addition, the following were controlled for in each subgroup analysis due to covariate imbalance within subgroups: for age subanalyses, Medicare Advantage enrollment at index, Care Assessment of Need score category, smoking status, Nosos category, and Gagne score; for COVID-19 wave subanalyses, no additional covariates; for Elixhauser rehospitalization risk score tertile subanalyses, VHA primary care visits in the 24 months prior to index date, VHA specialty care visits in the 24 months prior to index date, and Nosos category; for PC shortage area subanalyses, no additional covariates; and for sex subanalyses, no additional covariates.
may have resulted in incidental and unrelated SARS-CoV-2 diagnoses.Second, because we did not have access to Medicare Advantage data, we excluded individual comparators with Medicare Advantage and matched groups in which the veteran with SARS-CoV-2 had Medicare Advantage (346 988 [30.65%]).Third, we excluded individual comparators who were institutionalized at baseline or follow-up and matched groups in which the veteran with SARS-CoV-2 was institutionalized at baseline or follow-up, as these individuals may have systematically different risk of SARS-CoV-2 and hospitalization (210 768 [18.62%]).Fourth, to understand risk during discrete vs

Figure 1 .
Figure 1.Cumulative Incidence of Potentially Preventable Hospitalizations During 1-Year Follow-Up Among Veterans With SARS-CoV-2 and Comparators4

Table 1 .
Sample Characteristics for SARS-CoV-2 and Matched Uninfected Comparators a

Table 1 .
Sample Characteristics for SARS-CoV-2 and Matched Uninfected Comparators a (continued) multiple races or ethnicities.Most cohort members resided in urban areas (70.03%), and 24.11% in primary care professional shortage areas.Cohort members were well-connected with the VHA, with a mean (SD) of 8.31 (10.27)VHA primary care contacts (ie, face-to-face visits, telephone interactions, clinician time managing patient care) during the 2 years prior to index date.

Table 1 .
Sample Characteristics for SARS-CoV-2 and Matched Uninfected Comparators a (continued) Pregnancy was a matching variable but was zero for all persons.State of residence was a matching variable and included 50 states and Washington, DC.Index month was an exact matching variable and spanned 14 months (not shown, absolute SMDs for both state of residence and index month <0.100).

Table 2 .
Incidence, Risk Difference, and Adjusted Hazards of Potentially Preventable Hospitalization Among Veterans With SARS-CoV-2 Relative to Comparators Abbreviation: AHR, adjusted hazard ratio.aExtended Cox regression models censor for deaths and crossover infections individually and are adjusted for nonpreventable hospitalizations as a time-varying covariate.
found that compared with matched comparators, veterans with SARS-CoV-2 experienced marked increases in outpatient visits during the first 30 days after infection, and that half of the additional outpatient visits were delivered via telehealth.Taken together, findings may suggest the importance of in-person care after SARS-CoV-2 for preventing adverse events such as hospitalization.However, more study is needed.Overall, our findings suggest that care disruptions, which may have originated from individual and systems-level responses to the pandemic, may have induced consequential gaps in the treatment of ACSCs among individuals with SARS-CoV-2, leading to potentially preventable hospitalizations.Alternatively, in some cases, potentially preventable hospitalization may not be simply a proxy measure for ambulatory care access and quality, but rather may represent sudden disruptions in health care that can quickly lead to exacerbations among those with SARS-CoV-2.
Extended Cox models censor for deaths and crossover infections individually and are adjusted for nonpreventable hospitalizations as a time-varying covariate.Risk of Potentially Preventable Hospitalizations After SARS-CoV-2 InfectionFigure 3. Subgroup Adjusted Hazard Ratios (AHRs) of Potentially Preventable Hospitalizations Among Veterans With SARS-CoV-2 Relative to Comparators