INTRODUCTION

The coronavirus disease 2019 (COVID-19) pandemic has had deleterious effects globally and has disproportionately impacted certain vulnerable populations, such as those with mental health (MH) conditions. One study found that patients who self-identified as having depression, anxiety, bipolar disorder, post-traumatic stress disorder, or schizophrenia had a variety of concerns about the pandemic related to their MH diagnosis. These concerns included the disruption of MH treatment services, running out of medication, social distancing, and worsening of their MH condition.1 Studies have also demonstrated increased risk related to poor health outcomes in those with MH conditions.2 Seon et al.3 found that the risk for COVID-19 infections and COVID-19-related deaths was high among those with mental illnesses. Additionally, results of a 2021 meta-analysis also indicated that patients with MH conditions were at higher risk for mortality and hospitalization.4

In addition to MH conditions, patients with neurological disorders, specifically dementia, have also been shown to experience increased risk related to poor health outcomes. One meta-analysis reviewed 149 studies and found that patients with mental and neurological disorders experienced greater risk of infection with COVID-19, and increased severity of illness and mortality than the control groups.5 A study performed on electronic records in the United Kingdom, which assessed comorbidities and the increased risk of mortality from COVID-19, found an increased risk of hospitalization and mortality associated with both pre-existing depression and dementia. The study also found that pre-existing dementia in adults over 65 was a significant risk factor for severe COVID-19 infections.6 Similarly, an English cohort study found that having a severe mental illness or dementia placed patients at higher risk of dying from COVID-19,7 even when controlling for chronic medical conditions, including kidney disease, cardiovascular disease, and chronic obstructive pulmonary disease.8

While previous research has repeatedly identified worse outcomes of COVID-19 patients with MH conditions and dementia separately, to our knowledge, studies have yet to investigate the impact of comorbid MH conditions and dementia on COVID-19 outcomes. The aim of the current study was to investigate the health outcomes of patients who were hospitalized due to COVID-19 and had comorbid psychiatric conditions in addition to dementia, in hopes of further assessing the vulnerability of these individuals during a worldwide pandemic. We hypothesized that COVID-19 patients with MH conditions and dementia (MHD) would have worse health outcomes (higher intensive care unit [ICU] admission and mortality rates) than those with MH conditions alone (MH) or without MH conditions (MH0).

METHODS

Data from a COVID-19 registry database of a total of 1,089 COVID-19 positive patients were used in the study.9,10 Patients included in the registry were between 18 to 100 years old and tested positive for COVID-19 via nasopharyngeal swab with a reverse transcription-polymerase chain reaction (RT-PCR) test upon presentation to Providence Hospital in Michigan between March 8th, 2020, and May 16th, 2020. A data warehouse query of ICD-10 codes at discharge was conducted to identify COVID-19 registry patients who had MH diagnoses. The psychiatric diagnoses included in the current study were schizophrenia, schizoaffective disorder, depressive disorders, anxiety/phobia disorders, drug use disorders (i.e., opiates, cocaine, marijuana, sedatives, and other stimulants), alcohol use disorders, and tobacco use disorders. Patients with dementia were identified using the Charlson Comorbidity Index (CCI),11 which was also part of the COVID-19 registry.

COVID-19 patients without a MH condition or dementia were assigned to the MH0 (control) group (n=877). Patients with ≥ 1 MH diagnosis alone (without dementia) were included in the MH group (n=90). The MHD group was composed of patients with a dementia diagnosis alone and those with dementia and with ≥ 1 MH diagnosis (n=122). Any patients with a delirium diagnosis were excluded to avoid any confounding effect. Institutional Review Board approval was obtained prior to use of registry and data warehouse data.

The COVID-19 registry variables utilized in the present study included patient demographic data, admission and discharge dates, discharge status (alive or expired), history of comorbid physical conditions, and whether patients were treated in the ICU. To account for comorbidities and risk of mortality beyond their MH conditions and dementia, the CCI, quick Sequential Organ Failure Assessment (qSOFA),12 and the Glasgow Coma Scale (GCS)13 for each patient were also collected. The CCI was modified (mCCI) by the study team to remove dementia from the criteria as this would have biased the CCI score toward the dementia-containing groups. Other variables used as predictors of mortality were age (in years), body mass index (BMI), and laboratory values including total white blood cell count (WBC), hemoglobin, platelets, blood urea nitrogen (BUN), and oxygen saturation. ICU admission and mortality were the outcome variables and were used as measures of severe disease.

Statistical analyses were conducted with SAS Software version 9.4 (SAS Institute Inc, Cary, NC). Chi-square tests were used to determine differences between the MHD, MH, and MH0 in terms of demographics, BMI, mCCI, and qSOFA. The outcomes, mortality and ICU admission, were analyzed separately by logistic regression to assess the effects of the aforementioned potential confounders, which we categorized over and under the following thresholds: BMI (30kg/m2), hemoglobin (12.5 g/dL), WBC 9.5 109 /L, BUN (36.1 mg/dL), platelets (268 109 /L), qSOFA (≥ 1 vs 0), mCCI (4), GCS (15), age (65 years old), sex (male versus female), race (African American, Caucasian, or ‘Other’), and oxygen saturation (<95%). ‘Other’ races include Hispanic, Asian, and Native American. A p-value of <.05 was considered statistically significant.

To reduce the impact of correlation among explanatory variables and missing data, the initial multivariable logistic regression analyses were performed separately with the following sets of variables: 1) sex, age, race, and BMI (n=1089); 2) mCCI, qSOFA, and GCS (n=1089); 3) WBC, hemoglobin, platelets, and BUN (n=86); 4) oxygen saturation only. Other correlated variables with oxygen saturation, e.g., SaO2/FiO2 ratio, PaO2/FiO2 ratio had excessive missing values. The final model used all potential correlates recognized in steps 1-4. No further reduction was made.

For the final model, we gauged goodness of fit by the deviance statistic, Hosmer-Lemeshow and Spiegelhalter tests. Regression diagnostics were conducted to ascertain if there were potential influential observations based on their large contribution to Pearson and/or deviance chi-square statistics. None were found that would severely impact the overall fit of the final model. Finally, we summarized the effects of all factors in the final model by odds ratios (ORs) and 95% confidence intervals (CIs).

RESULTS

Demographic data from the three groups demonstrated a statistically significant difference between the groups for all characteristics except for gender (Table 1). The overall mortality rate was 20.2% (220/1089) for the cohort but rates differed significantly among the 3 groups (MH0 group = 16.1%, MHD group = 54.1%, and MH group = 14.4%). Mortality among the three groups was statistically significant (p <.0001). In unadjusted analyses, compared with the MH0 group, the MHD group had significantly higher mortality (OR=6.15, 95% CI: 4.13, 9.17, p<.0001). However, there was no difference in mortality when comparing the MH group with the MH0 group (OR=0.88, 95% CI: 0.48, 1.63, p=0.687). A third comparison of the MHD group with the MH group showed statistically significantly higher mortality (OR=6.98 (95% CI: 3.51, 13.88), p < .0001). Because differences between groups on other factors associated with mortality may have contributed to the increase in mortality, a multivariable logistic regression analysis was conducted to adjust for the effects of confounding.

Table 1.Demographic differences between the three groups.
Characteristic Subgroup Control N (%) Dementia and MH
N (%)
MH only
N (%)
p-⁠value
Age < 65 years 433 (49.4) 12 (9.8) 44 (48.9) <0.0001
≥ 65 years 444 (50.6) 110 (90.2) 46 (51.1)
Gender Female 437 (49.8) 67 (54.9) 50 (55.6) 0.37
Male 440 (50.2) 55 (45.1) 40 (44.4)
Race Caucasian 191 (21.8) 60 (49.2) 28 (31.1) <0.0001
African
American
656 (74.8) 62 (50.8) 61 (67.8)
Other 30 (3.4) 0 1 (1.1)
BMI (kg/m2) < 30 433 (49.4) 92 (75.4) 43 (47.8) <0.0001
≥ 30 444 (50.6) 130 (24.6) 47 (52.2)
mCCI < 4 448 (51.1) 10 (8.2) 41 (45.6) <0.0001
≥ 4 429 (48.9) 112 (91.8) 49 (54.4)
qSOFA 0 490 (55.9) 18 (14.8) 41 (45.6) <0.0001
≥ 1 387 (44.1) 104 (85.3) 49 (54.4)

Note: MH = mental health conditions; BMI = body mass index; mCCI = modified Charlson Comorbidity Index; qSOFA = quick Sequential Organ Failure Assessment; p-values from Chi-square tests.

Table 2 summarizes the effects on mortality from the multivariable logistic regression analysis. In comparison with the MH0 group, mortality was significantly higher in the MHD group, but not in the MH group. The overall group effect was significant. Other significant effects were sex, age, race (African American versus Caucasian), mCCI, GCS, and O2 saturation. BMI and qSOFA were not significant. The ORs show that the effects are in the expected direction. Overall, the regression model had good discriminative power (c-statistic=0.85) and gave excellent goodness-of-fit as indicated by non-significant Pearson deviance statistics, Hosmer-Lemeshow and Spiegelhalter tests.

Table 2.Odds ratio estimates and 95% confidence intervals for effects on mortality.
Effect Estimate 95% Confidence Limits p-⁠value p-⁠value*
MHD vs MH0 2.130 1.319 3.439 0.002 .002
MH vs MH0 0.645 0.322 1.292 0.216
Sex: Male vs Female 1.966 1.368 2.826 <.0001
Age: ≥65 vs <65 years 2.246 1.393 3.622 <.0001
Race: AA vs Caucasian 0.639 0.430 0.949 0.027 .08
Race: Other vs Caucasian 0.549 0.109 2.775 0.468
BMI: ≥30 vs <30 kg/m2 1.055 0.726 1.533 0.778
mCCI: ≥4 vs <4 2.587 1.587 4.216 <.0001
qSOFA: ≥1 vs 0 1.797 0.986 3.276 0.056
Glasgow Coma Scale: <15 vs 15 3.739 2.147 6.512 <.0001
O2 Saturation: <0.95 vs ≥0.95 1.848 1.284 2.660 0.001

Note: MHD = mental health and dementia group; MH0 = control group; MH = mental health only group; AA = African American; BMI = body mass index; mCCI = modified Charlson Comorbidity Index; qSOFA = quick Sequential Organ Failure Assessment; O2 = oxygen; *p-value for overall effect.

Excluding race, BMI, and qSOFA from the model had little impact on the ORs of the other variables, except for GCS. The revised OR=3.74, 95% CI: 2.15, 6.51 for comparing GCS<15 to 15 was significant. This finding was expected because of the substantial correlation between the qSOFA and GCS measures. Our primary comparison of MHD versus MH0 was substantively the same: OR=2.46, 95% CI: 1.56, 3.93, p<.0001. Restricting race to African American and Caucasian (i.e., dropping all 31 patients of ‘Other’ race) had practically no effect on the estimated ORs.

Overall, there were a total of 204 ICU admissions, which is 18.7% of the total sample but significantly different among the 3 groups (MH0 group = 18.1%, MHD group = 13.9%, and MH group = 31.1%; p=0.007). In unadjusted analyses, compared with the MH0 group, the MH group had significantly higher likelihood of ICU admission (OR=2.04, 95% CI: 1.26, 3.29, p=0.004). However, there was no statistically significant difference in ICU admission when comparing the MHD group with either the MH0 group or MH group. Since there were differences between groups on other factors associated with ICU admission that may have contributed to the increase in this severity outcome measure, a multivariable logistic regression analysis was conducted to adjust any confounding effects.

Table 3 summarizes the effects on ICU admission from the multivariable logistic regression analyses. The overall group effect was significant. In comparison with the MH0 group, ICU admission was significantly higher in the MH group and statistically significantly lower in the MHD group. Other significant effects were race, BMI, qSOFA, GCS, and O2 saturation. Age, sex, and mCCI were not significant. Similar to the model for mortality, the regression model for ICU admission had good discriminative power (c-statistic=0.75) and gave excellent goodness-of-fit as indicated by non-significant Pearson deviance statistics, Hosmer-Lemeshow and Spiegelhalter tests.

Table 3.Odds ratio estimates and 95% confidence intervals for effects on Intensive Care Unit admission.
Effect Estimate 95% Confidence Intervals p-⁠value p-⁠value*
MHD vs MH0 0.465 0.257 0.844 0.0118 0.0007
MH vs MH0 1.951 1.157 3.289 0.0122
Sex: Male vs Female 1.160 0.832 1.616 0.3824
Age: ≥65 vs <65 years 0.908 0.598 1.378 0.6501
African American vs Caucasian 1.517 1.001 2.299 0.0497 0.0192
Other vs Caucasian 3.478 1.344 9.002 0.0102
BMI: ≥30 vs <30 kg/m2 1.805 1.273 2.561 0.0009
mCCI: ≥4 vs <4 1.290 0.847 1.964 0.2351
qSOFA Score: ≥ 1 vs 0 3.206 2.013 5.106 <.0001
Glasgow Coma Scale: <15 vs 15 1.918 1.210 3.039 0.0056
O2 Saturation: <0.95 vs ≥0.95 1.962 1.405 2.739 <.0001

Note: MHD = mental health and dementia group; MH0 = control group; MH = mental health only group; BMI = body mass index; mCCI = modified Charlson Comorbidity Index; qSOFA = quick Sequential Organ Failure Assessment; O2 = oxygen; *p-value for overall effect

Excluding race and BMI had no statistically significant effect on the model. However, the remaining variables, including age, sex, mCCI, qSOFA, had some impact on the ORs of the other variables, but in this analysis the most prominent was the GCS<15 to 15, which showed a revised OR=3.74, 95% CI: 2.15, 6.51, p< 0.0001. As above with the mortality analysis, the restriction of race to African American and Caucasian (i.e., dropping all 31 patients of ‘Other’ race) similarly had basically no effect on the estimated ORs.

DISCUSSION

The current study aimed to investigate the health outcomes of mortality and ICU admission of patients who were hospitalized due to COVID-19 and had comorbid psychiatric conditions in addition to dementia to further understand the vulnerability of these individuals during a worldwide disease pandemic. With regard to mortality, results indicated that patients with both dementia and MH diagnoses were at higher risk for mortality when compared with patients without MH conditions, which was consistent with our hypothesis. However, the lack of a significant difference in mortality of the MH only group when compared with the MH0 control group was somewhat unexpected as previous research has shown that COVID-19 patients with MH diagnoses were at a higher mortality risk.3,4

Regarding ICU admission, our hypothesis that patients with MH conditions and dementia would have more ICU admissions than patients without MH conditions or dementia was not confirmed. In fact, our results indicated that the likelihood of being admitted to the ICU was lower in these patients. These findings may have been due to the influence of other comorbidities, which may not have necessarily urgently required ICU level care but that due to comorbidity burden more frequently led to mortality. Another consideration is that patients with dementia and MH conditions may have had greater difficulty with communication of their physiological symptoms depending on the severity of the dementia, which may have led to greater mortality in conjunction with this disease burden. It is also possible that patients with dementia in addition to MH conditions may have been more likely to have advance directives that indicated a preference for no ICU admission and/or do not resuscitate/do not intubate orders depending on the stage of their dementia.

Findings of the present study also indicated that patients with COVID-19 and MH diagnoses only (without dementia) had higher rates of ICU admission when compared with control patients. This finding makes sense conceptually and is consistent with previous research indicating increased ICU admission has been found in those with COVID-19 and substance use disorders as our study included substance use disorders in the MH group.14 However, other studies have shown COVID-19 patients with MH conditions are not at higher risk of ICU admission4 and ICU admission may be decreased in those with COVID-19 and schizophrenia.15 Taken together, these findings continue to call attention to potentially discrepant data with regard to ICU admission, which may be problematic when determining resource allocation during a possible future disease pandemic. Future studies should also focus on clarifying reasons for discrepant findings with regard to ICU admission in patients with MH conditions.

One strength of the study was the use of data from an extensive COVID-19 registry database comprised of patient data from two different hospital campuses in southeast Michigan, an area significantly affected by COVID-19.10 The registry also included multiple pieces of laboratory data (e.g., WBC counts, platelets, oxygen saturation, etc.), several formal indices (e.g. GCS, CCI and qSOFA), and demographic data. Additionally, the study time frame was intentionally constrained as to avoid widespread public vaccination, which would have likely impacted the outcomes.

The results of the study must be interpreted within the context of its limitations. An initial aim of the current study was to examine whether the risk of severe outcomes differed between the individual psychiatric conditions. However, these differences could not be calculated due to small sample sizes of the individual psychiatric conditions and thus lack of power. Another limitation may be the effect of the adjustment of the CCI to the mCCI. Although the modification was made in order to limit the bias from dementia, when the CCI was modified it lost its established predictive value. Since this change was made, it may have led to bias in the opposite direction, leading to a comparison that minimizes or underestimates the true burden of comorbidity without the inclusion of dementia.

Finally, there were limitations within the medical record system, which made it difficult to discern patients who had pre-existing MH conditions in comparison with those who developed their condition during their hospital stay. Timing of symptoms/diagnosis of MH conditions may affect the results since newly diagnosed MH patients may not have carried the same risk for worse COVID-19 outcomes as patients with pre-existing MH conditions. Future research studies could aim to cross match inpatient and outpatient data to clearly delineate pre-existing conditions from new onset conditions.

CONCLUSIONS

The current study examined the outcomes for the MH population early in the COVID-19 pandemic. The present study found several indications that patients with dementia and one or more MH conditions were particularly vulnerable to poorer outcomes during the first wave of a disease pandemic. More studies are needed in the future to assess the dementia and MH populations together as these populations are at significant risk for mortality from COVID-19. Our study findings have implications for this group being considered a high-risk population in the event of future disease pandemics and may inform allocation of resources (e.g., healthcare resources, vaccinations).


FUNDING

The authors have no funding to report for this study.

DISCLOSURES

The authors have no disclosures or conflicts of interest to report for this study.

CORRESPONDING AUTHOR

Laci L Zawilinski, PhD, LP
Associate Chair for Graduate Medical Education
Michigan State University College of Human Medicine
Director of Behavioral Health for Family Medicine/
Clinical Psychologist
Henry Ford Health Providence Hospital
Department of Medical Education
16001 West Nine Mile Road
Southfield, MI 48075
Work phone: 248-849-5525
Work fax: 248-849-8120
Email: lzawili1@hfhs.org

Author contributions

Conceptualization: Daniel E Andrews IV (Supporting), Laci L Zawilinski (Supporting), Joseph Gardiner (Supporting). Data curation: Daniel E Andrews IV (Supporting), Nancy M Jackson (Lead), Lahib Douda (Supporting), Heraa Hasnat (Supporting), Jeffrey C Flynn (Supporting), Joseph Gardiner (Supporting). Validation: Daniel E Andrews IV (Supporting), Jeffrey C Flynn (Lead). Writing – original draft: Daniel E Andrews IV (Lead), Dora Mell (Supporting). Visualization: Laci L Zawilinski (Supporting). Supervision: Laci L Zawilinski (Lead), Abdulghani Sankari (Lead). Writing – review & editing: Laci L Zawilinski (Lead), Jeffrey C Flynn (Supporting), Abdulghani Sankari (Supporting). Methodology: Dora Mell (Supporting), Nancy M Jackson (Supporting), Lahib Douda (Supporting), Heraa Hasnat (Supporting). Investigation: Lahib Douda (Supporting), Heraa Hasnat (Supporting), Abdulghani Sankari (Supporting). Formal Analysis: Joseph Gardiner (Lead).