Age and frailty are independently associated with increased COVID-19 mortality and increased care needs in survivors: results of an international multi-centre study

1Age and frailty were independently associated with increased risk of mortality in hospitalised patients with COVID-19.

2Delirium was not predictive of mortality but was predictive of critical care admission with COVID-19.

3Age, frailty and delirium were associated with increased odds of transitions of care needs at discharge in survivors.

4Coronavirus disease 2019 (COVID-19) is a multi-system disease caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Early data suggested older and/or co-morbid adults were at increased risk of adverse outcomes. The clinical frailty scale (CFS) featured in critical care escalation guidelines in UK and Canada. Early evidence of impact of frailty produced mixed findings; studies suggested frailty was, and was not associated with increased mortality with COVID-19. Our collaborative previously demonstrated in other conditions that delirium is common and associated with adverse outcomes, especially in frail older adults. Two distinct cohorts identified that delirium is more common in frail older adults with COVID-19. Thus, a need for research on delirium and frailty with COVID-19 was recognised. Additionally, prolonged recovery has been reported in survivors, although the impact of COVID-19 upon transitions of care needs with age and frailty has not been previously studied.

5To evaluate the association of age, frailty and delirium with adverse outcomes including mortality and secondary outcomes to include critical care admission, incident delirium and transitions of care needs in survivors in hospitalised patients with COVID-19.

6We included unscheduled hospital admissions of adults aged ≥ 18 years old with COVID-19 infection in this observational study. Emergency department discharges and nosocomial COVID-19 were excluded. Prospective data upload upon clinical suspicion was encouraged; clinicians identified patients during medical clerking, or ward transfer. Retrospective identification was dependent on local COVID-19 coding processes, involving medical records, informatics or microbiology.

7Investigation was led through the Geriatric Medicine Research Collaborative (GeMRC). The protocol was openly available on GeMRC, British Geriatrics Society and University of Birmingham REDCap webpages, and disseminated via emails and social media. Sites were required to obtain local, regional and national approvals, and declare these were in place when registering. Data sharing agreements were arranged where required. Sites were provided REDCap data upload logins; secure encrypted web-based data management software. Sites uploaded anonymised patient-level data onto REDCap. Independent data managers ensured quality control.

8Suspected COVID-19 infection was diagnosed clinically considering symptoms, radiology and laboratory tests. Laboratory confirmation was conducted according to local policies and World Health Organization guidance; identification of SARS-CoV-2 from reverse transciptase polymerase chain reaction from oropharyngeal or high nasal swabs (RdRp gene assay), or antibodies against SARS-CoV-2 in serum samples ≥ 14 days after symptom onset. Patients were included if there was strong clinical suspicion but no laboratory confirmation.

9Data were extracted from routinely collected clinical information; variables are outlined in the online supplement (Supplementary Table S1). Screening with the 4 ‘A’s Test (4AT) on admission was recommended; ≥4 was suggestive of prevalent delirium. Incident delirium was defined as documented emergent delirium during admission. Frailty was derived based on function 2 weeks before admission using the 9-point CFS, by prospective clinical assessment or retrospectively from medical records.

10Primary end point was death during index admission. Secondary end points were critical care admission, incident delirium and increased care requirements at discharge (as defined below).

11Data analysis was performed using STATA SE version 16 (StataCorp LLC, Texas, USA) by an independent statistician (KO). Descriptive variables were expressed as median and interquartile range (IQR), and counts; chi-squared and Mann–Whitney U tests were applied for statistical significance of mortality differences.

12We used Cox proportional survival analysis to assess impact of delirium and frailty upon inpatient mortality. Univariable and multivariable analyses were conducted as follows:

13Model 1: variables previously associated with COVID-19 adverse outcomes; age, sex, C-reactive protein (CRP), Ferritin, body mass index (BMI), Alanine aminotransferase, lymphocyte: neutrophil ratio, Glomerular Filtration Rate [Modification of Diet in Renal Disease, MDRD, formula], co-morbidities coded individually and illness severity by admission national early warning score (NEWS).

14Model 2: variables above with CFS; CFS 1–3 was the comparison group, and CFS 9 and missing CFS were separate discrete groups. Additional analyses were conducted with age and frailty (excluding CFS 9) as continuous variables.

15Model 3: variables in Models 1 and 2 and delirium (prevalent/incident).

16We performed Wald and Likelihood Ratio tests for model fit for age and frailty individually and together as predictors in all models. In addition, we assessed for multiplicative interactions between age and frailty upon the primary outcome of mortality censored at the point of discharge. Frailty was grouped with the exclusion of CFS 9 in these models (CFS 1–3, CFS 4–6, CFS 7–8, CFS missing).

17We performed sensitivity analyses on models for the primary outcome:

18Excluding patients aged < 65 years old.

19Excluding patients admitted to hospitals outside the UK.

20We used binary logistic regression to assess impact of variables on critical care admission and incident delirium (excluding prevalent delirium), and ordinal logistic regression to assess impact upon care requirements at discharge. Incident delirium was considered as any new diagnosis of delirium by a healthcare professional at any time during admission, where this was not present at admission. Increased care was defined as transitions across three care levels: living at home without formal care, living at home with formal care or living in a 24-h long-term care facility.

21The study includes data from 5,711 individuals with COVID-19 admitted to 55 hospitals in 12 countries. Supplementary Figure S1 shows reasons for data exclusion. Supplementary Table S2 demonstrates participating site locations. Median age was 74 and 55.2% were male. Table 1 shows full baseline patient characteristics.

22Risk of death increased independently with age and frailty in univariable and multivariable analyses (Table 2), including with age and frailty as continuous variables (Supplementary Table S3, online supplement). Risk of death tripled >80 years old (hazard ratio [HR] 3.57, 95% confidence interval [CI] 2.54–5.02), compared with 18–50, and in very severely frail individuals (CFS 8 versus CFS 1–3) (HR 3.03, 95% CI 2.29–4.00). Models 1 and 2 are available online (Supplementary Table S4). Age and frailty together and not individually as predictor variables improved model of fit (LR χ2 = 91.3, P < 0.001; Wald χ2 = 207.9, P < 0.001). Additionally, mortality risk increased with age and frailty together in multiplicative interactions (Supplementary Table S5 and Supplementary Figure S2). Delirium was predictive of mortality in univariable but not multivariable analysis. Risk of death increased with higher CRP or ferritin, more severe renal disease and cancer. Mortality did not differ across BMI cut-offs; risk of death was increased with missing BMI. Figure 1 demonstrates Kaplan–Meier curves for risk of death for frailty and delirium. Results were not affected by sensitivity analyses for ≥65 years old, or UK data only (Supplementary Tables S6 and S7).

23Critical care admissions were more likely with greater illness severity, CRP, or ferritin or BMI < 18.5 or ≥30 (Table 3), and less likely with age, frailty and dementia. Admissions were six times less likely in >80 years old compared with 18–50 (odds ratio [OR] 0.13, 95% CI 0.08–0.21). Delirium was independently associated with critical care admission (OR 2.67, 95% CI 2.06–3.46). Models 1 and 2 are available online (Supplementary Table S8).

24Delirium incidence was 9.6%. Incident delirium odds increased with age but not frailty (Table 4). Risk in >80-year olds was double that of 18–50 (OR 2.21, 95% CI 1.37–3.59). Incident delirium odds were independently associated with male sex, illness severity and cardiovascular disease. Dementia was not associated with incident delirium. Model 1 is available online (Supplementary Table S9).

25Increased care risk increased with age, frailty, delirium, dementia and mental health problems (Table 5). Likelihood of increased care >80 years old was triple that for 18–50 (OR 3.07, 95% CI 2.25–4.20). Increased care levels were seven times more likely with severe frailty (CFS 7) than without frailty (CFS 1–3) (OR 7.00, 95%CI 5.27–9.32). Models 1 and 2 are available online (Supplementary Table S10).

26Age and frailty were independently associated with COVID-19 mortality. This is consistent with risk exhibited for nearly all other illnesses, and does not represent relative risk for COVID-19 compared with other illnesses; risk of dying increases with age and frailty within ‘normal’ risk. However, increases in absolute mortality risk will be most pronounced in these groups, even if relative risk is equivalent to young or robust individuals. It is important to consider the results of likelihood testing, which demonstrated that age and frailty as individual predictors improved the model of fit. Risk continued to increase with increasing age and with increasing frailty. Therefore, the greatest risk will have been exhibited by the oldest and most frail patients. Underlying mechanisms for increased mortality with age and frailty with COVID-19 may include endothelial dysfunction leading to vasoconstriction and organ dysfunction, heightened inflammation and pro-coagulant state, dysregulated angiotensin-converting enzyme 2 activity promoting viral uptake and immunesenescence. Immunesenescence is associated with immune system changes that are age-related, frailty-related, or inactivity-related.

27Delirium was not independently predictive of mortality, but was associated with critical care admission; delirium itself is an illness severity marker. This may relate to exclusion of prevalent delirium cases, or represent different delirium pathophysiology with COVID-19 compared with other conditions. We demonstrated novel results that frailty, age, delirium, dementia and mental health diagnoses were independently associated with transitions of care in survivors i.e. adverse functional outcomes. Quality of life is individual and subjective, but increased dependency will have been hugely significant for many individuals. Transitions of care are also likely to have been associated with state-funded health and social care system costs, during a time of international economic recession, which has wider health impacts.

28Previous studies assessing COVID-19 mortality with frailty showed mixed results. These have been predominantly small single site studies. Our results are consistent with another study including 1,410 UK-hospital and 154 Italian-hospital patients; sub-categorised CFS and age were independently associated with mortality. A second UK single site study involving 677 patients demonstrated increased mortality in CFS ≥ 6. A small Italian study demonstrated that a Frailty Index was also predictive of mortality, even after removing co-morbidities from the index. However, a small UK study showed that age but not continuous CFS was predictive of mortality in univariable analysis; CFS was not included in multivariable analysis. Similar results were shown in another UK study. Differences may relate to under-powering in smaller studies, or how CFS was recorded or extracted from clinical records. In our study, few patients had missing CFS (11% versus 32% in the latter study). However, these studies also assessed mortality following discharge, whereas we report mortality during index admission. Whilst this is important to distinguish, we do not consider this explanatory for differences; it is unlikely robust patients were more likely to have been discharged to die outside of hospital. As our data were censored at hospital discharge, this has also been accounted for.

29Consistent with results elsewhere, male sex, inflammation and cancer were associated with mortality. However, BMI was not independently predictive of death, which is contrary to previous research, although high and low BMI were associated with critical care admission. Studies previously adjusted for age, sex and co-morbidities, but this is the first study to adjust for frailty. Both being underweight and obese have been associated with frailty.

30Previous research demonstrated that delirium is a common COVID-19 presentation. It is surprising that delirium did not predict mortality in our study. Delirium has been consistently associated with mortality in hospitalised patients with other illnesses. A single-site Italian study demonstrated that delirium, diagnosed against reference criteria by geriatricians, was associated with 4-fold increased COVID-19 mortality risk. Under-recognition is unlikely to fully explain differences as prevalence was high in our study. Delirium was associated with critical care admission, which is consistent with a Brazilian study.

31Frailty has been associated with prevalent delirium in COVID-19 and other conditions. Incident delirium was not associated with frailty in this study. Additionally, dementia, a classical delirium risk factor, was not associated with risk. Higher risk was demonstrated with cardiovascular disease and illness severity. These differences in patient groups affected may explain differences in COVID-19 mortality with delirium compared with other conditions. Severely ill robust patients may have been as likely to develop delirium as frail patients, but less likely to die. We are not aware of other studies reporting transitions of care in patients with COVID-19. Studies outside of COVID-19 have shown that frailty and cognitive spectrum disorders are associated with increased risk of new discharge to a care home.

32A major strength of our study is that it was large and multi-centre. This is the largest study to date evaluating how frailty and delirium relate to outcomes in hospitalised patients with COVID-19. Data collectors were not involved in analysis; statistical analysis was conducted by an independent statistician. We included large numbers of variables in multivariable analyses, which had been previously associated with adverse outcomes with COVID-19. We did not collect ethnicity data. Previous research identified that Black African or Caribbean and Bangladeshi individuals are at increased risk of adverse outcomes from COVID-19, although this also relates to socioeconomic status. This personal information requires stricter safeguards, and is not internationally standardised.

33Data collectors were provided delirium and frailty diagnosis guidance. Prospective data collection was encouraged and diagnoses were made by clinicians. Results provide predictive value of real-world delirium and frailty diagnoses. Given the high overall delirium prevalence, we consider it unlikely that under-diagnosis of delirium significantly impacted upon our overall results. Retrospectively identified data may be vulnerable to documentation errors, and we cannot rule out possibility of data entry errors via REDCap. To counteract this, data managers performed quality control checks on uploaded data, and contacted sites where data were missing or outlying values recorded.

34Overall, missing data rates were low; discrete missing categories were included to account for those that were. The highest missing data rates related to BMI (height and weight). This could explain why obesity was not predictive of mortality; missing BMI was associated with mortality. It may have been most likely to be missing in most unwell or possibly most obese patients. Higher odds of mortality were demonstrated with missing CFS, which is consistent with previous studies. Multiple imputation has shown consistent effect of frailty on mortality in other populations.

35Inclusion of hospitalised patients only is a limitation. The majority of people with COVID-19 had mild symptoms, particularly those who were young and robust; inclusion of community cases may have amplified association of frailty with mortality. Conversely, frail individuals may have died unexpectedly, or advance care planning decisions may have been made to avoid admission. Internationally, significant numbers of people who died from COVID-19 died in 24-h long-term care facilities. We also recognise that our sample may not be internationally representative. As dissemination was via GeMRC, more older adults may have been identified if clinicians were working on geriatric medicine wards. This in itself should not have affected main results; data collection was not biased towards outcomes.

36Healthcare policy should recognise heightened vulnerability in older adults, particularly those living with frailty. Caution should be exhibited to ensure older adults with frailty are shielded from high risk COVID-19 exposure, such as ensuring isolation procedures during elective surgery admissions. Older and/or frail patients admitted with COVID-19 should undergo holistic assessment, ensuring treatment is proportionate and in accordance with their wishes. Research identifying underlying mechanisms of adverse outcomes with age and frailty may enable novel intervention development. It is vital older adults with frailty are adequately represented in all COVID-19 research. Vaccines and COVID-19 treatments may have different responsiveness with age or frailty.

37Considering high odds of increased care in patients with frailty, urgent funding is needed to enhance community and hospital rehabilitation services. At present, longer-term consequences of COVID-19 are unknown. Chronic symptoms of fatigue and systemic upset have been reported even in community-dwelling previously robust adults. Further research should focus on understanding mechanisms and adverse recovery predictors, particularly in patients who developed acute sarcopenia or induced frailty.

38In this international multi-centre study, age, frailty and morbidity were independently associated with adverse outcomes with COVID-19. Patients who were older or more severely frail were more likely to die, less likely to be admitted to critical care, and more likely to require higher care levels on discharge in survivors. Increased awareness of importance of measuring frailty alongside age and co-morbidities in hospitalised patients will assist clinicians making holistic decisions involving treatment of reversible pathology, prevention of unwanted or burdensome treatment and early rehabilitation.