Factors Associated with COVID-19 Infection among a National Population of Individuals with Schizophrenia in the United States

Background: Schizophrenia is a serious mental illness and individuals with schizophrenia are a particularly vulnerable and often under-served population. Methods: This research sought to evaluate the factors associated with risk of COVID-19 infection among a representative population of individuals with schizophrenia in the United States. This study was a retrospective cohort analysis of medical and pharmacy claims among a population-based sample of 493,796 individuals residing in the United States with schizophrenia or schizoaffective disorder, between January 1, 2019 and June 30, 2020. A conrmed diagnosis of COVID-19 infection by September 30, 2020 was regressed on demographics, social determinants (measured at the county level, with each measure categorized into quintiles), comorbidity, and pre-pandemic (December 2019 – February 2020) healthcare utilization characteristics. Results: A total of 35,249 (7.1%) individuals were diagnosed with COVID-19. Elevated odds of COVID-19 infection were associated with age groups above 40 years (compared to 18-29 years) increasing consistently from 40-49 years (OR: 1.16) to 80+ years (OR:5.92), male sex (OR: 1.08), Medicaid (OR: 2.17) or Medicare (OR: 1.23) insurance, African American race (OR: 1.42 comparing the highest and lowest quintile), Hispanic ethnicity (OR: 1.23, comparing highest to lowest quintile), and higher Charlson Comorbidity Index, increasing from scores of 1-2 (OR: 1.84), 3-4 (OR: 2.53), to 5+ (OR: 2.76). Select psychiatric comorbidities (depressive disorder, adjustment disorder, bipolar disorder, anxiety, and sleep-wake disorder) were associated with elevated odds of infection, while alcohol use disorder and PTSD were associated with lower odds. A pre-pandemic psychiatry (OR:0.56) or community mental health center (OR:0.55) visit were associated with lower odds as was antipsychotic treatment: long-acting injectable antipsychotic (OR: 0.72) and oral antipsychotic (OR: 0.62). Conclusions: Individuals with a diagnosis of schizophrenia for categorical signicance testing by


Introduction
Schizophrenia is a serious mental illness characterized by symptoms of psychosis (e.g., hallucinations, delusions, and thought disorder), negative symptoms (e.g., social and emotional withdrawal), and cognitive dysfunction (e.g., di culties with attention, memory, concentration). Schizophrenia can be a disabling condition, with 50% of those individuals who receive a diagnosis having long-term psychiatric problems and 20% having chronic symptoms and disability [1]. In the United States, the estimated prevalence of diagnosed schizophrenia is between 0.4% and 0.5% [2][3][4], affecting an estimated 1 million adults.
Due to the nature of the condition, individuals with schizophrenia are a particularly vulnerable and often underserved population [5]. Individuals with schizophrenia have higher rates of unemployment, are more likely insured via Medicaid, have lower socioeconomic status, lower levels of education, and reside in high risk, poverty environments [6].
These conditions and situations can be important determinants of health and as important, appear to increase both the risk of COVID-19 infection and the risk of severe morbidity and mortality following infection. Recently published research indicates that individuals with any serious mental illness, including schizophrenia, have 1.5 times the risk of COVID-19 infection [7], due at least in part to cognitive impairment, lower awareness of risk, and challenges with infection control [2]. Once infected, individuals with schizophrenia have substantially higher rates of hospitalization and mortality [7][8]. In a consecutive case series of 7,348 individuals with a positive COVID-19 laboratory test, Nemani, et al. [9] reported that history of a schizophrenia spectrum disorder diagnosis was associated with an increased odds of mortality (OR: 2.67; 95%CI: 1.48 -4.80), an association not reported for either mood disorders or anxiety.
The current research study builds on this preliminary evidence by evaluating the factors associated with risk of COVID-19 infection among a large, representative population of individuals with schizophrenia in the United States.

Methods
This study was a retrospective cohort study of 493,796 individuals who entered the COVID-19 pandemic with a diagnosis of schizophrenia. The study compared the population of individuals who did and did not receive a COVID-19 diagnosis by September 30, 2020. The study protocol was approved by the Advarra Institutional Review Board.

Data Sources and Management
Individuals were identi ed and selected from medical and pharmacy claims Real-World Data licensed from Decision Resources Group. In addition, these claims data were used to measure healthcare resource utilization. Social determinants of health were measured at the county level from the Area Health Resources File available from the Department of Health and Human Services (www.data.hrsa.gov/topics/health-workforce/ahrf). Each individual was assigned to a county based on a combination of their 3-digit ZIP code, linked to the 5-digit ZIP code associated with their provider, assigned hierarchically beginning with the primary care provider, then the mental health care provider or clinic.

Identi cation and Selection of Study Participants
The study period was January 1, 2019 through September 30, 2020. Eligible individuals met each of the following criteria: diagnosis of schizophrenia or schizoaffective disorder (ICD-10-CM: F20.x, F25.x) de ned by two or more outpatient claims (or one inpatient claim) between January 1, 2019 and June 30, 2020, 18 years of age or older as of January 1, 2019, a resident of the 50 United States or District of Columbia, and ongoing use of healthcare services as measured by at least one medical or pharmacy claim between each of January 1 and June 30, 2019; January 1 and June 30, 2020; and July 1 to September 30, 2020.

Statistical Analysis
Covariates included demographic characteristics (age and gender), insurance type (Medicare, Medicaid, commercial, or VA/other), medical comorbidity measured by the Charlson Comorbidity Index [10] psychiatric comorbidities, health care resource utilization prior to the pandemic, and social determinants of health.
Psychiatric comorbidities included anxiety (F41.x), adjustment disorders (F43.2x), bipolar depression (F30.x, F31.x), obsessive-compulsive disorder (F42.x), PTSD (F43.1x), sleep-wake disorders (G47.x), alcohol and substance use disorders (F10.x -F19.x, F32.x, F33.x, F34.1), and depressive disorders (F32.x, F33.x). Health care resource utilization was measured in the three-month period prior to March 1, 2020. Outpatient care was measured by an all-cause psychiatry o ce visit, psychotherapy session, or community mental health clinic visit. Acute care utilization was measured by partial-day hospitalization or schizophrenia-speci c emergency department or inpatient hospitalization. Medication treatment was measured as at least one claim for an oral or long-acting injectable (LAI) antipsychotic. Social determinants of health, measured at the assigned county of residence included population density (residents per sq. mile as of 2010), median household income, education (% graduated high school), race (% African American/black, % white, % Asian), and ethnicity (% Hispanic).
Frequencies and percentages were calculated for categorical variables: signi cance testing by chi-square test. Means and standard deviations for continuous variables: signi cance testing by two-sided T-test for mean.
Adjusted odds ratios (OR) and 95% con dence intervals (CI) were derived from a multivariate logistic regression modeling of COVID-19 infection (dependent variable). The threshold for signi cance was set at 0.05. All analyses were performed with SAS software, v9.4, SAS Institute Inc., Cary, NC, USA.
In the three months prior to March 1, 2020, an estimated 24.8% and 6.5% of the population had a claim for an oral antipsychotic medication and an LAI antipsychotic, respectively. A total of 8.9% had a community mental health center visit, 8.3% had a psychiatry o ce visit, and 6.1% had psychotherapy session. Among the countylevel social determinants of health, the average population density was 3,664 residents per square mile (SD: 9,914), the average median household income was $60,400 (SD: $15,800), with an average high school graduation rate of 87.0% (SD: 5.3%). The average county had 16.0% African American, 59.1% white, 16.4% Hispanic, and 5.7% Asian residents. (Table 2) By September 30, 2020, a total of 35,249 (7.1%) individuals had received a COVID-19 diagnosis. Individuals diagnosed with COVID-19 were more likely to be female (50.7% vs. 44.3%), older (59.9 vs. 49.4 years of age), more likely insured by Medicaid (80.7% vs. 67.2%), and less likely insured by a commercial plan (5.0% vs. 14.1%) or Medicare (14.2% vs. 18.5%). Further, those with a COVID-19 diagnosis resided in counties with slightly higher income ($61.24K vs. $60.37K), lower rates of education (86.5% vs. 87.1%), and lower population density (3,550 vs. 3,673). Race and ethnicity also varied by county among those with and without a COVID-19 diagnosis, with positive cases residing in counties with a higher percent African American (16.6% vs. 16.0%), Hispanic (18.1% vs. 16.3%), and Asian (6.2% vs. 5.7%) residents. All comparisons were statistically signi cant.
Pre-pandemic health care utilization also varied between those with and without a COVID-19 diagnosis.

Regression
With the exception of sex, the associations above remained after adjustment for all other factors. (Table 3; Figure   1) After adjustment, the factors most strongly positively-associated with COVID-19 infection were age (80+ years  [11][12][13]. Our results focus on identifying risk factors for infection and are consistent with prior research indicating that demographic characteristics [11,[14][15], social determinants [11,16,14,17], psychiatric comorbidities [7,11], and treatment [18][19] were associated with risk of COVID-19 infection. Among our study population, social determinants of health were signi cant contributors to individual risk of infection, with higher rates realized by individuals insured by Medicaid or who lived in counties with higher proportions of African American or Hispanic residents and higher rates among individuals residing in higher income areas. We identi ed three studies in general populations that reported similar results. In a retrospective analysis of COVID-19 infection among 34,503 individuals who sought care at a single regional health system, Rozenfeld [14] reported adjusted odds ratios of 1.51 for African Americans and 2.07 among individuals of Latino ethnicity. Lan[16] reported COVID-19 incidence rate ratios of 2.78 and 2.41 among African American and Hispanic healthcare workers, compared to non-Hispanic white colleagues and in a state-level analysis, Padalabalanarayanan [17] estimated 4.6% more COVID-19 cases for each percent increase in a state's African American population. In one of the only other studies to report on individuals with schizophrenia, Wang [11] reported an adjusted odds of infection of 2.3 among African Americans (compared to Caucasians) with schizophrenia. The consistently elevated risk of infection among African Americans and individuals in lowincome household is particularly concerning because these populations are at increased risk for poorer outcomes following COVID-19 infection [11,[20][21].
Further, after accounting for demographic and social determinants, comorbid depressive disorder, bipolar disorder, adjustment disorder, anxiety, and sleep-wake disorders were each independently associated with increased risk of COVID-19 infection. This is consistent with the ndings of Taquet [13], who reported that the hazard ratio for a diagnosis of COVID-19 infection was greatest for anxiety disorders, insomnia, and dementia and with those of Wang [11], who reported elevated adjusted odds of COVID-19 infection among individuals with ADHD (adjusted OR: 7.3), bipolar disorder (OR: 7.7), depression (OR: 10.4), and schizophrenia (OR: 9.9).
In contrast, a diagnosis of either alcohol use disorder or post-traumatic stress disorder (PTSD) was associated with lower odds of COVID-19 infection. These results are consistent with those of Yazdi et al. [22], who reported that 43% of alcohol use disorder patients lived alone during the pandemic, which likely reduced social interactions, leading to a decreased infection risk. Further, in a Veterans Health Administration study, Haderlein et al. [23] reported that veterans with clinically diagnosed PTSD were more likely to receive a COVID-19 test than those without a PTSD diagnosis; these patients were also less likely to test positive for COVID-19.
Consistent with the increased risk among older adults, individuals with higher comorbidity burden were also at increased risk of infection. This may be a consequence of poorer health, compromised immune function, greater healthcare needs, and the inability to maintain physical distancing [21]. Bitan et al.[8] reported that physical health comorbidities were associated with higher COVID-19 mortality rates, as schizophrenia patients were more likely to be obese, smoke, and be diagnosed with diabetes and COPD.
Among the most striking results, individuals with schizophrenia who were actively receiving mental health care in the three months prior to March 2020 had dramatically lower risk of COVID-19 infection. Individuals with a recent, pre-pandemic psychiatry visit, psychotherapy session or community mental health center visit were substantially less likely to receive a COVID-19 diagnosis. These results suggest that active management of schizophrenia reduces the behaviors that lead to exposure and improves patient awareness of the risks associated with COVID-19 [24].
Though the inverse association between COVID-19 infection and antipsychotic medication use documented in our study may also represent the value of active therapeutic management, other research conducted to date has yielded inconsistent and con icting results about the association between antipsychotic treatment and COVID-19 infection risk. Recently published studies have documented both an increased risk of infection [19] and a decreased risk of infection among patients on antipsychotics [18]. The in ammatory mechanisms involved in psychiatric illness and COVID-19 infection in combination with the anti-in ammatory and antiviral effects of antipsychotics appear to be a key driver of the ongoing research [25]. As an example, Crespo-Facorro[26] reported that aripiprazole and COVID-19 modulate the expression of genes that modulate immune and in ammatory response at a rate substantially higher than expected.

Strengths and Limitations
The study included several strengths and limitations. The primary strength is that the study population of 493,796 individuals, represents a substantial proportion of all individuals with diagnosed schizophrenia in the United States, increasing the generalizability of results. Second, the population included individuals across a spectrum of ages, insurance types, and geographic regions. Third, the study included both healthcare utilization and social determinants of health data. The most signi cant limitation is the open dataset, which does not include insurance eligibility information and thus may result in incomplete ascertainment of healthcare utilization. The second substantive limitation is that social determinants are measured at the county, rather than individual, level and thus may not represent an individual's speci c situation. Finally, the dataset includes neither symptom severity nor mortality information.

Conclusions
In the general U.S. population, individuals diagnosed with schizophrenia were at signi cantly higher risk of COVID-19 infection and our study suggests that risk was elevated further among the underserved, among African Americans and the Hispanic population, and among those with psychiatric comorbidity. The protective effect of psychiatric care and of antipsychotic medication use underscored the importance of continual care throughout the pandemic. The study protocol including use of the Clarivate claims dataset was reviewed by the Advarra Institutional Review Board (IRB) and was determined to be exempt from broad IRB approval, as this research study did not qualify as human subjects research. All data were de-identi ed and HIPAA compliant, and all methods were performed in accordance with the relevant guidelines and regulations.

Consent for Publication:
Not applicable.
Availability of Data and Materials: The administrative medical and pharmacy claims data that support the ndings of this study are available from Clarivate (https://clarivate.com) but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available.      Figure 1 Please See image above for gure legend.