Preexisting Neuropsychiatric Conditions and Associated Risk of Severe COVID-19 Infection and Other Acute Respiratory Infections

This cohort study investigates the potential association between a previous diagnosis of a neuropsychiatric condition and severe outcome from COVID-19 infection and other severe acute respiratory infections.

T he COVID-19 pandemic has caused at least 5 million deaths and continues to exert significant pressures on health care systems globally. 1 Despite successful vaccination strategies in many countries, it remains important to be aware of conditions that may predispose to worse outcomes from SARS-CoV-2 infection so that at-risk populations can be identified for the purposes of public health strategy.
Recent evidence indicates that the risk of developing an incident neuropsychiatric condition after severe COVID-19 infection and a severe acute respiratory infection (SARI) is similar. 2 Moreover, 3 systematic reviews with meta-analysis [3][4][5] and a subsequent major cohort study have 6 reported that people with preexisting neuropsychiatric conditions are at higher risk of COVID-19 mortality than people without such a diagnosis. These reports align with prepandemic data suggesting that severe mental illness, 7 schizophrenia, 8 and depression 9 are associated with increased risks of developing SARIs. It remains unclear, however, whether the increased risk of developing severe disease associated with these conditions differs between COVID-19 infection and other SARIs.
Before the COVID-19 pandemic, results of studies [10][11][12] suggested that some drug classes typically prescribed for these neuropsychiatric conditions have also been associated with increased risk of more severe respiratory infection. Evidence for some drug classes, however (eg, antidepressants), is conflicting. 13,14 There is a need for greater understanding of the associations between medications typically prescribed for psychiatric conditions and severe respiratory infection.
As COVID-19 becomes endemic, it is critical to understand which medical conditions and treatments may predispose to more severe disease. Furthermore, understanding whether neuropsychiatric conditions contribute to a generalized increase in risk for acute respiratory illness or whether this risk is disease specific is important for ongoing management of individuals and health systems. This study aimed, therefore, to evaluate the associations of preexisting neuropsychiatric conditions and treatments with COVID-19 outcomes compared to those with SARI.

Methods
This longitudinal cohort study was conducted in accordance with our prespecified protocol 15 ; deviations from the protocol are described in the protocol deviation section of the methods. The study was approved by the QResearch Scientific Committee, which has ethical approval capabilities from the East Midlands-Derby research ethics committee, including a project-specific patient waiver for patient consent forms due to data protections (including deidentification) implemented as part of the agreement. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. 16 We used the QResearch database of English primary care records, version 45 (EMIS Health), which has individual-level linkage to Hospital Episode Statistics (HES), Public Health England's second-generation surveillance system database regarding SARS-CoV-2 testing, the Intensive Care National Au-dit and Research Centre (ICNARC) database, and the Office for National Statistics' national mortality register. All records in the database meeting inclusion criteria were used. Patients self-reported their race or ethnicity as one of Asian, Black, White, or other if they did not identify with one of the previous 3 categories. Completion of this variable is not mandatory in primary care clinics; therefore, in some cases, it is missing.
We extracted 2 temporally distinct longitudinal cohorts: a prepandemic cohort comprising all adults (18 years and older) entering from January 24, 2015, until January 23, 2020 (the day before first recorded COVID-19 case in the UK), and a contemporary cohort comprising all adults entering from January 24, 2020, until the date of data extraction (May 31, 2021). The index date in the prepandemic cohort was the latest of the following: (1) January 24, 2015; (2) December 31 in the year that patients turned 18 years; or (3) 1 year after registration with a participating practice if registration was after January 24, 2014. The index date for the contemporary cohort was January 24, 2020 (first recorded SARS-CoV-2 infection in the UK), with participants younger than 18 years or not registered with a participating practice for at least 1 year on that date excluded. Patient follow-up time was from index date until the first record of any outcome (COVID-19-or SARI-related hospital or ICU admission, or death) or censoring: the earliest of date of deregistration from an EMIS practice or study end (January 23, 2020, for the prepandemic cohort and May 31, 2021, for the contemporary cohort).

Exposures
Detailed definitions of exposures are reported in full in the study protocol. 15 Briefly, we used SNOMED/Read Codes in the primary care records, and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes in the HES records to identify individuals with selected neuropsychiatric conditions. We used primary care records to identify prescriptions of relevant medications based on the British National Formulary. 17 We grouped similar neuropsychiatric conditions and their respective pharmacological treatments under the following categories: anxiety, mood, and psychotic disorders and hypnotic/anxiolytic, antidepressant, and antipsychotic medications. The primary care code lists used to create these groups are available on the QResearch website. 18 In addition, we investigated diagnosis only (ie, irrespective of treatment) of the following conditions: depression, dementia, schizophrenia, and bipolar disorder. Finally, we investigated any antidepressant use irrespective of concurrent diagnosis. Patients were considered exposed if they had a diagnosis of a neuropsychiatric condition at any time before the index date, and/or at least 2 prescriptions for neuropsychiatric medications in the 6 months before baseline. Patients with a first-ever diagnosis of a neuropsychiatric condition or 2 or more prescriptions for related medications within 6 months during the follow-up time were excluded from analysis for the relevant condition to avoid immortal time bias. 19 Outcomes Using these cohorts, we estimated associations between prior neuropsychiatric conditions and relevant medication use and subsequent severe COVID-19 or SARI-related illness (in the contemporary and prepandemic cohorts, respectively). A severe outcome was defined as hospitalization, intensive care unit (ICU) admission, or death related to COVID-19 infection or SARI. Hospital admissions of any duration were identified using linked HES data and were considered COVID-19 related if the person had a positive SARS-CoV-2 test result within 14 days before or during admission or if they had ICD-10 code U07.01 in their HES record (indicating confirmed COVID-19 infection). For ICU admissions, we used ICNARC's data set of COVID-19related admissions. Death from any cause within 28 days of a positive COVID-19 test or where COVID-19 was listed as a cause of death on the death certificate was considered COVID-19 related. Hospital admissions were considered SARI related if any ICD-10 code from J09-J22 occurred in the HES record. Deaths where the same ICD-10 codes were listed as a cause of death on the death certificate were considered SARI related. SARIrelated ICU admission was determined based on corresponding RE, or respiratory, diagnosis codes used internally by ICNARC. 15 For both conditions, a composite outcome of severe COVID-19 infection or SARI was created of the first recorded of hospitalization, ICU admission, or death.

Statistical Analyses
Flexible parametric survival models with clustering by general primary care practice were used to calculate hazard ratios (HRs) and 99% CIs, to assess whether prior neuropsychiatric diagnosis or treatment was associated with the risk of severe COVID-19 infection or SARI. All final analysis models were adjusted for demographic and clinical factors identified from clinical/epidemiologic understanding, as summarized in the directed acyclic graphs in the protocol. 15 Missing data were observed for race and ethnicity, smoking status, alcohol status, body mass index, and Townsend deprivation score quintile; multiple imputation with chained equations was used to replace these under the missing at random assumption. 20 Imputation models included all covariates and end points. Five imputations were generated, with model coefficients and SEs pooled in accordance with Rubin rules. 21 P values were 2-sided, and results were considered statistically significant if the 99% CI did not include 1. All analyses were conducted in Stata, version 17 (StataCorp).

Protocol Deviation
In some instances, we deviated from the original study protocol. Additional exposures were analyzed beyond those listed in the protocol based on clinical relevance; these included the following: bipolar disorder (ICD-10 codes F30-31 and corresponding Read Codes), depression (ICD-10 F06.3, F32-F33, F34.1, F41.2, F44.8, F92.0, and corresponding Read Codes), schizophrenia (ICD-10 F06.2, F20, F23.0-F23.2, and corresponding Read Codes), and antidepressant use (tricyclic and related antidepressant drugs, monoamine-oxidase inhibitors, selective serotonin reuptake inhibitors [SSRIs], other antidepressant drugs). Bipolar disorder is included in the broader category of mood disorders, schizophrenia is included in psychotic disorders, and antidepressant use is considered the treatment for mood disorders. Nevertheless, bipolar disorder, schizophrenia, and antidepressant use were highlighted for their particular clinical importance, and we have been careful not to overinterpret the results beyond those of the broader categories.
A frailty term for general practice was proposed for the proportional hazards models, but the use of Royston-Parmar models (Stata package stpm2) precluded this. Instead, we calculated SEs adjusted for clustering at the general practice level.

Sensitivity Analysis
A sensitivity analysis was added to investigate the potential confounding effect of COVID-19 vaccination. In this analysis, the end day in the contemporary cohort was changed to December 7, 2020, the day before the first COVID-19 vaccination was administered in the UK. Pilot analysis showed similar findings when hospitalization, ICU admission, and death were investigated separately; therefore, only the combined variable was used in the main analysis. An analysis of COVID-19-and SARI-specific mortality is included in eTables 1, 2, and 3 in the Supplement.

Results
There were 11 134 789 individuals in the prepandemic cohort (  Abbreviations: BMI, body mass index; ICU, intensive care unit. a Calculated as weight in kilograms divided by height in meters squared. b Indicates total numbers recorded for each event during the relevant study period; therefore, a person may appear in more than 1 row if they, for example, are hospitalized and subsequently die. The row first recorded is the earliest record of any event; therefore, people will only appear once. c Diagnoses defined as any record before baseline; treatment defined as 2 or more prescriptions in the 6 months before baseline; percentages of cases are of the total cohort; percentages of the number excluded are of total number diagnosed. d Other race and ethnicity includes people who did not identify with any of the previous 3 options. 28 days of a positive test result, 5183 had COVID-19 listed on the death certificate, and 15 534 had both. The numbers of patients with a neuropsychiatric condition who developed SARI or COVID-19 illness are shown in Table 3 for each analysis. Regression estimates from multiply imputed flexible parametric survival models are shown in Table 3. Log − log survival plots were used to check the proportional hazards assumption visually, and all models satisfied the assumption. In addition, the Figure displays estimates from the maximally adjusted multivariable models for both SARI and COVID-19 infection. All adjusted models showed that a severe outcome from COVID-19 and SARI was more likely in people with a preexisting diagnosis of a neuropsychiatric condition (COVID-19 smallest effect: anxiety diagnosis only, adjusted HR, 1.16; 99% CI, 1.12-1.20 and greatest effect: dementia diagnosis, HR, 2.85; 99% CI, 2.71-3.00; SARI smallest effect: anxiety diagnosis only, HR, 1.16; 99% CI, 1.13-1.18 and greatest effect: psychotic disorder diagnosis and treatment, HR, 2.56; 99% CI, 2.40-2.72). Effect estimates for all neuropsychiatric conditions and treatments were broadly similar for both outcomes, although the effect estimate for the association between severe COVID-19 and dementia appeared greater than for that of SARI, including an association between severe COVID-19 and dementia (COVID-19 infection: HR, 2.85; 99% CI, 2.71-3.00 vs SARI: HR, 2.13; 99% CI, 2.07-2.19) (Figure). Results from sensitivity analyses with COVID-19-and SARI-specific mortality as the outcome and restricting follow-up time in the contemporary cohort to the prevaccination period were similar to the main findings of the study (eTables 1, 2, and 3 in the Supplement).

Neuropsychiatric Conditions and COVID-19
Results of systematic reviews with meta-analysis of population-based cohort studies suggest that neuropsychiatric conditions (including anxiety, mood, psychotic, bipolar, personality, eating, and major depressive disorders; alcohol and substance abuse and misuse; and schizophrenia) are associated with an increased risk of COVID-19 mortality. [3][4][5] Despite using slightly different definitions of severe illness, the direction of the effect estimates was consistent and the magnitude broadly similar with those of the current study, which used a composite outcome of hospitalization, ICU admission, and death. Together, the findings suggest that people with a diagnosis of a neuropsychiatric condition have an associated increased risk of severe outcome from COVID-19 infection. To the prior work, we add that the associated increased risks with COVID-19 are broadly similar to those seen in other SARIs.

Neuropsychiatric Conditions and SARI
Cohort studies conducted before the COVID-19 pandemic have also reported an association between preexisting neuropsychiatric conditions and SARIs. 7,9 A Danish population study 9 conducted over 11 years found that people diagnosed with depression were at an associated increased risk of presenting with various respiratory infections. In addition, an English study 7 of people from the Oxford region from 1999 to 2011 found that preexisting bipolar disorder, depression, and phobic anxiety conditions were each associated with a 2-fold increase in risk of pneumococcal lung infections. These findings are largely in line with those of the current study. Thus, 3 large observational cohort studies have found evidence suggesting that people with a preexisting neuropsychiatric condition are at an associated increased risk of severe outcome from SARIs.

Psychotropic Medication
We found that people taking hypnotic, anxiolytic, antidepressant, or antipsychotic medication had an associated increased risk of more severe disease from both COVID-19 infection and SARI irrespective of whether they had a corresponding neuropsychiatric diagnosis, but findings from previous research are mixed. The current findings are consistent with evidence that SSRIs are associated with a higher risk of hospital mortality in patients admitted to ICU 13 and with a systematic review and meta-analysis 10 that found antipsychotic medication to be associated with increased risk of pneumonia. The authors of the systematic review, however, commented that there is a lack of randomized clinical trials in the area and that some observational studies failed to control for key confounders. In contrast, findings from a recent randomized clinical trial found that fluvoxamine, an SSRI, reduced the risk of hospitalization when given to symptomatic adult patients with an acute presentation consistent with COVID-19. 23 It is interesting to compare the timing of drug administration. The current study and Ghassemi et al 13 investigated SSRI use before tertiary care admission, whereas Reis et al 23 investigated administration after the onset of symptoms consistent with COVID-19. Ultimately, the reason for the discrepancy is unknown, but the current study adds that the associated increased risks were not COVID-19 specific; rather, they were broadly similar to the risks in other severe respiratory conditions.

Supplementary Analyses
Supplementary analyses showed that these associations were similar before the availability of vaccines for COVID-19 and when only the most severe outcome, mortality, was considered. This suggests that despite varying pressure on hospitals throughout the pandemic and the evolution of treatments and prophylactic measures, the association between neuropsychiatric conditions and severe COVID-19 infection and SARIs remained similar.
Although this study provided clear support for an association between neuropsychiatric conditions and more severe outcomes from respiratory infections, the observational design means it is not sufficient to demonstrate causality. We corrected for relevant demographic and clinical factors. However, it may be that neuropsychiatric illness occurs as part of a general picture of increased health disparity to the general population, 24 which could increase the risk of opportunistic infection and lead to increased severity of disease from such an infection. Alternatively, some people with neuropsychiatric conditions may not have access to suitable clinical facilities or may delay clinical presentation until later in the disease stage, by which time the severity of their respiratory illness has increased. 24 In either case, it seems clear that there is a generally elevated risk of severe acute respiratory illness associated with neuropsychiatric conditions and that this risk is similar with SARS-CoV-2 and other respiratory pathogens. This information is critical in determining the impact of neuropsychiatric illness on disease outcomes not only during the current pandemic but also into the future as COVID-19 becomes endemic.

Strengths and Limitations
Strengths of our study include the use of large cohorts that are representative of the English general population and the use of multiple validated electronic health care data sets linked at the individual level to provide accurate ascertainment of exposures, outcomes, and other relevant confounders. These data sets minimize selection and recall bias because they use prospectively collected data. Furthermore, we used comparable respiratory infections to determine whether the reported association between neuropsychiatric conditions and severe COVID-19 was due to the SARS-CoV-2 pathogen specifically or a more general association with various pathogens that cause respiratory illness.
This study also has limitations. The use of routinely collected health care data means that outcomes and exposures are not formally adjudicated, as the database is dependent on coding by individual practitioners. There may also be recording bias due to disruption in patient attendance at general practices during the study period, particularly during the COVID-19 pandemic. We did, however, collect records from multiple linked databases including general primary care practice, hospital, and registry data, and this should serve to minimize potential bias from underreporting. Additionally, exclusion of people diagnosed during the study period may have introduced selection bias, although pilot analyses suggested that this was not the case, and we used a valid technique. 19 Time-related heterogeneity can be masked by using HRs; however, the proportional hazards assumption held for all analyses, and sensitivity analysis varying follow-up time showed similar results to the main analyses. Finally, as with all observational studies, there is a risk of residual confounding, which we sought to minimize through adjustment for a range of confounders.

Conclusions
The current cohort study builds on previous work in this field by investigating not only a range of neuropsychiatric conditions and psychotropic medications but by also comparing their associations with both severe COVID-19 infection and other SARIs.
The adjusted results comparing severe outcomes from COVID-19 disease and other SARI largely seemed similar, suggesting that the associations were not disease specific. Although dementia was associated with a higher increased risk of severe outcome from COVID-19 than for SARI, well-documented impacts of the pandemic on specific care settings mean that this result should be interpreted with caution.