Clinical Features and Outcomes of Patients with Dementia Compared to an Aging Cohort Hospitalized During the Initial New York City COVID-19 Wave

Study design and data source

This study was approved by the Institutional Re-view Board (IRB) of Columbia University under protocol AAAS9834. We expand on a previously described, retrospective, manually abstracted data-base of all consecutive patients who presented to the New York-Presbyterian/Columbia University Irving Medical Center (NYP/CUIMC) ED or were hospitalized at NYP/CUIMC and had a positive result on reverse transcriptase polymerase chain reaction assay for severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), indicating active COVID-19 illness [17]. Data was manually abstracted from the electronic medical record (EMR) into a Research Electronic Data Capture (REDCap) instrument [17, 18]. In compliance with the IRB-approved protocol, patient consent was not applicable in this study.

The original REDCap database contained information on demographics, body mass index (BMI), comorbidities, and common presenting signs and symptoms as descriptive variables [17]. For outcome variables, the database contained information on status for death, discharge, intubation, level of care, length of stay, and in-hospital complications [17]. Length of stay was defined as the date and time from initial ED presentation to the date and time of either discharge or death.

Data collection and definitions of variables

Additional abstraction was performed in this study to supplement the previously created REDCap database with descriptive information regarding dementia status, additional signs and symptoms, and code status. Dementia was identified through chart review if Alzheimer’s dementia, vascular dementia, Lewy-Body dementia, Parkinson’s dementia, or any other dementias of unspecified type were indicated in the EMR provider notes, such as admission notes, progress notes, or discharge summaries. Dementia was not stratified based on severity. Additional signs and symptoms in this study included delirium, chest pain, and abdominal pain. Delirium was identified if the patient was described as altered or confused in the EMR.

Do-Not-Resuscitate/Do-Not-Intubate (DNR/DNI) was indicated by either a) DNR and DNI order or b) DNR order only at death or date of discharge. If no code status was indicated at discharge or death, patients were assumed to be full code. DNR/DNI orders placed before inpatient admission or in prior hospitalizations were also recorded. Patients who were DNR/DNI at discharge or death and were never full code during their hospital course were included in a separate category as “consistent DNR/DNI.”

Study sample

The processed REDCap dataset only included visits with the highest level of care for patients with multiple visits [17]. Levels of care in this study included admission to the intensive care unit (ICU), inpatient admission (not ICU), or ED visit only. For patients with multiple visits of the same level of care, the most recent visit was selected [17]. Additional checks for data quality and random subsample review were performed for accuracy and consistency [17].

This study included patients who were 65 years of age or older at the time of ED presentation and followed them until completion of their hospital course. We excluded patients with incomplete REDCap records or missing length of stay fields. In total, 531 unique patients aged 65 and above with a positive result on PCR assay for SARS-CoV-2 who first presented to the ED or were hospitalized at NYP/CUIMC between March 6 and April 7, 2020 were included (Fig. 1).

Fig. 1

STROBE Flowchart of Study Sample Selection. Selection of the study sample is presented in a flowchart. The steps from “Unprocessed REDCap database (1,737 records)” to “Unique patients presenting to the ED or hospitalized at NYP/CUIMC with COVID-19 (n = 1,168)” were performed with an R script used in the prior study that created the REDCap database [17]. Additional exclusions were made in this study for patients with a presenting age less than 65, with records not marked “Complete” in REDCap, and with missing values for length of stay.

Statistical analysis

We first characterized the prevalence of presenting signs and symptoms, excluding those with < 5%prevalence. Then, to examine the association between dementia status and in-hospital mortality, we created an unadjusted Kaplan-Meier curve and four Cox proportional hazards models: an unadjusted analysis, an analysis adjusted for age and sex, and two multivariate models adjusted for comorbidities, with/without code status as an additional covariate. The outcome was in-hospital death. Discharged patients were censored from the analysis at the time of discharge for the primary Cox analyses but were included in a sensitivity analysis of the final model with code status and in the Kaplan-Meier curve until the time of maximum length of stay (which was approximately 100 days) to account for potential bias.

Comorbidities adjusted for were based on those established by the CDC as risk factors for severe COVID-19 illness: active cancer, chronic kidney dis-ease/end-stage renal disease (CKD/ESRD), chronic obstructive pulmonary disease, immunosuppressed state (defined as having had solid/liquid transplant, HIV, or being defined as “immunosuppressed” in the REDCap database), obesity (defined as BMI greater or equal to 30), coronary artery disease, heart failure, and diabetes (both types I and II) [19]. The total number of patients included in our two multivariate Cox models with/without code status was 482 patients, as obesity was adjusted for in these models, and 49 patients did not have data for BMI. No imputation was performed in this study. Cause-specific hazard ratios (HRs) from the Cox multivariate regression analyses were visualized with forest plots and reported with 95%confidence intervals and p-values for each covariate.

We used R version 4.0.2 for our statistical analysis [20–26]. Categorical variables were reported as proportions (%) while continuous variables were reported as medians [interquartile range (IQR)]. For categorical variables, the χ² test with Yates’ continuity correction was used. Fisher’s exact test was used for categorical variables that had a total prevalence of less than 5%. Mann-Whitney U test was used for continuous variables. Statistically significant was considered p < 0.05 in this study.

Baseline characteristics

We identified 531 patients at least 65 years of age at presentation who were hospitalized with COVID-19. There were 116 patients with previously diagnosed dementia, and 415 patients with no known history of dementia. Patients with previously diagnosed dementia were older (median 84.00, IQR [78.00–89.00] versus 73.00, IQR [69.00–80.00], p < 0.001), more often female (56.9%versus 45.1%, p = 0.03), and had a higher prevalence of DNR/DNI orders at discharge or death when compared to patients with no known history of dementia (60.3%versus 35.2%, p < 0.001; Table 1). There were no significant differences in race/ethnicity, smoking status, or comorbidities, except that more patients with previously diagnosed dementia had a prior cerebrovascular accident (28.4%versus 10.8%, p < 0.001) and none had a previous transplant (versus 4.1%for patients with no known history of dementia, p = 0.03).

Signs and symptoms

Presenting signs and symptoms for COVID-19 differed among patients with and without dementia (Fig. 2). Patients with previously diagnosed dementia more often presented with delirium (36.2%versus 11.6%, p < 0.001), and post-hoc multiple logistic regression found this remained significant after adjusting for age and sex (Supplementary Figure 1). These patients also less often manifested or communicated dyspnea (45.7%versus 65.8%, p < 0.001), diarrhea (8.6%versus 21.9%, p = 0.002), myalgias (8.6%versus 21.0%, p = 0.004), chills (4.3%versus 19.3%, p < 0.001), nausea/vomiting (6.9%versus 16.9%, p = 0.01), and headache (0.9%versus 7.2%, p = 0.02).

Fig. 2

Unadjusted Bar Graph for Signs and Symptoms. Signs and symptoms with greater than 5%total prevalence are presented in a bar graph from highest to lowest prevalence in the total cohort. Patients with previously diagnosed dementia were more likely to present with delirium (altered mental status) and less likely to present with dyspnea, diarrhea, myalgias, chills, nausea/vomiting, and headache. Asterisks indicate statistical significance: ^***p < 0.001; ^**0.001 < p < 0.01; ^*0.01 < p < 0.05.

Outcomes and in-hospital complications

Patients with previously diagnosed dementia had higher in-hospital mortality in this cohort than patients with no known history of dementia (50.0%versus 35.4%, p = 0.006; Table 2). They also had a shorter length of stay in days (median 4.60, IQR [2.21–7.55] versus 6.96, IQR [2.98–13.70], p < 0.001) and more often died earlier in their hospital course (Fig. 3A). A post-hoc subgroup analysis of all patients aged 78 or above (the lowest IQR boundary found for age in the dementia cohort) found that these differences in mortality and earlier hospital death did not remain significant once accounting for age as a confounder (Fig. 3B). There was no increase of in-hospital complications for patients with previously diagnosed dementia; instead, patients with no known history of dementia had a higher incidence of septic shock (23.6%versus 12.1%, p = 0.01) and new onset arrhythmia (12.3%versus 5.2%, p = 0.04; Table 2).

Fig. 3A

Kaplan-Meier Curve for In-Hospital Mortality Relative to Dementia History. Overall survival for patients with dementia and those with no known history of dementia are presented in an unadjusted Kaplan-Meier curve. Shading indicates 95%confidence intervals. Discharged alive patients were considered event-free and only censored at the maximum length of stay value in this study, which was approximately 100 days. Patients with dementia died earlier in their hospital course and had significantly lower overall survival than patients with no known history of dementia.

Fig. 3B

Kaplan-Meier Curve for In-Hospital Mortality Relative to Dementia History, Subgroup of All Patients Aged 78 or Above. Post-hoc subgroup analysis of overall survival for patients with dementia and those with no known history of dementia aged 78 or above are presented in a Kaplan-Meier curve. Shading indicates 95%confidence intervals. Discharged alive patients were considered event-free and only censored at the maximum length of stay value in the subgroup of patients aged 78 or above, which was approximately 89 days. Earlier in-hospital death for patients with dementia was no longer significant in this analysis, and mortality at the end of the study period for both cohorts was also no longer statistically significant (53.3%versus 46.8%, p = 0.41).

Cox multivariate regression analysis

In our multivariate Cox model without code status (Table 3, Model 3), the HR for dementia was 1.27 (95%CI: 0.87–1.85). Covariates that showed statistically significant increased risk of in-hospital mortality in this model were age and CKD/ESRD, while female sex conferred lower risk. After adding consistent DNR/DNI status to the model (Table 3, Model 4), dementia showed a similar HR to the model without code status (HR = 1.29, 95%CI: 0.88–1.88). Covariates which remained statistically significant in the multivariate Cox model with code status were age and female sex, but not CKD/ESRD. Diabetes became significantly associated with in-creased in-hospital mortality, and the HR for consistent DNR/DNI status was 2.16 (95%CI: 1.52–3.08). The significant covariates in the final multivariate Cox model with code status remained significant in the sensitivity analysis. All covariates are shown for the multivariate Cox models with and without code status in forest plots (Supplementary Figures 2and 3A and 3B).

Limitations

There are several important limitations in this study. First, dementia diagnoses are well recognized as being historically underreported in medical records [33], and the dementia indicator in this dataset could instead be a subset mostly comprised of those with more apparent, advanced stages of dementia. However, the prevalence of dementia in this cohort is similar to well-established estimates in the same community served by this hospital [34, 35]. The separation of symptoms in multiple domains relative to dementia lends further credence to an accurate diagnosis. Although the record review extracted meaningful clinical data about acute neuropsychiatric symptoms, a formal determination of toxic metabolic delirium due to COVID-19 cannot be definitively determined. Moreover, despite efforts to specifically identify new symptoms of delirium, it is possible that for some patients, the symptoms identified in data extraction are simply reflective of underlying dementia.

Given that the most common sign in dementia is memory loss, this may contribute to poor recognition and underreporting of COVID-19 symptoms; with the exception of fever and delirium, all other COVID-19 symptoms were less often evident in patients with dementia. Additionally, obtaining an accurate history from a patient with dementia could have been more difficult. This was especially pertinent in the context of a busy ED during a pandemic, where family members and caregivers may not have accompanied patients or were primarily engaged by phone or virtual means, and thus introduced challenges with the history-taking process. Nonetheless, these findings reflect the way in which care was delivered and recorded in the imperfect reality of the early days of what became the epicenter of the US COVID-19 epidemic, which increasingly correlates with care being delivered around the world in a worsening pandemic.

While dementia was not a statistically independent risk factor for in-hospital death, our study was limited in sample size and may have been under-powered to detect some outcomes. Additionally, there could be other residual confounders in our analyses. Race and ethnicity are well recognized as complex surrogates for social determinants of health; they could not be included in our models because of missing data for many in the cohort. Nonetheless, significant associations were identified in other established comorbidities for severe COVID-19 illness, namely CKD/ESRD and diabetes [19].

While code status was associated with in-hospital mortality in this cohort, illness severity at the time of a DNR/DNI order could not be accounted for in our model and could be an important residual confounder [36]. Following a diagnosis of a potentially severe illness like COVID-19, advanced directives are more likely to be discussed, and this becomes even more the case should the illness become severe or appear futile. With the premise that discussions of advanced directives are clinically indicated and ethically justified, it is likely that severe illness is driving increased in-hospital mortality rather than code status [36]. These code status discussions were especially challenging in early periods of COVID-19 care in NYC when hospital resources, particularly ventilators, were in short supply. Other centers may have faced similar challenges during the COVID-19 wave of winter 2021 as well as other contexts. Aside from controlling for code status in regression models, we considered alternative approaches such as inversion propensity weighting. However, such an approach presents pragmatic challenges including sample size which was too small to sufficiently enable appropriate balance across all covariates. Moreover, any additional modeling would not be able to further manage the challenging contextual determinants of code status that we have highlighted.

Finally, potential errors could have occurred during the manual abstraction of the EMR into the REDCap database, although quality control checks were performed to minimize these errors. All findings in this study were from a single urban medical center, which limited the sample size and generalizability. It is also recognized that collective understanding of how to effectively treat COVID-19 has evolved since this cohort was encountered early in the US pandemic.