key: cord-0789907-e29rtrpi
authors: Hägg, Sara; Jylhävä, Juulia; Wang, Yunzhang; Xu, Hong; Metzner, Carina; Annetorp, Martin; Garcia-Ptacek, Sara; Khedri, Masih; Boström, Anne-Marie; Kadir, Ahmadul; Johansson, Anna; Kivipelto, Miia; Eriksdotter, Maria; Cederholm, Tommy; Religa, Dorota
title: Age, frailty and comorbidity as prognostic factors for short-term outcomes in patients with COVID-19 in geriatric care
date: 2020-08-14
journal: Journal of the American Medical Directors Association
DOI: 10.1016/j.jamda.2020.08.014
sha: 9b521716b29959da21ad709613f188846f3a115d
doc_id: 789907
cord_uid: e29rtrpi

Abstract Objectives To analyze whether frailty and comorbidities are associated with in-hospital mortality and discharge to home in older adults hospitalized for coronavirus disease 2019 (COVID-19). Design Single-center observational study. Setting and Participants: Patients admitted to geriatric care in a large hospital in Sweden, between March 1st – June 11th 2020; 250 were treated for COVID-19 and 717 for other diagnoses. Methods COVID-19 diagnosis was clinically confirmed by positive RT-PCR test or, if negative, by other methods. Patient data were extracted from electronic medical records, which included Clinical Frailty Scale (CFS), and were further used for assessments of the Hospital Frailty Risk Score (HFRS) and the Charlson Comorbidity Index (CCI). In-hospital mortality and home discharge were followed up for up to 25 and 28 days, respectively. Multivariate Cox regression models adjusted for age and sex were used. Results Among the COVID-19 patients, in-hospital mortality rate was 24% and home discharge rate was 44%. Higher age was associated with in-hospital mortality (hazard ratio [HR]=1.05 per each year, 95% confidence interval [CI]=1.01-1.08) and lower probability of home discharge (HR=0.97, 95% CI=0.95-0.99). CFS (>5) and CCI, but not HFRS, were predictive of in-hospital mortality (HR=1.93, 95% CI=1.02-3.65 and HR=1.27, 95% CI=1.02-1.58, respectively). Patients with CFS>5 had a lower probability of being discharged home (HR=0.38, 95% CI=0.25-0.58). CCI and HFRS were not associated with home discharge. In general, effects were more pronounced in men. Acute kidney injury was associated with in-hospital mortality and hypertension with discharge to home. Other comorbidities (diabetes, cardiovascular disease, lung diseases, chronic kidney disease and dementia) were not associated with either outcome. Conclusions and Implications Of all geriatric COVID-19 patients, three out of four survived during the study period. Our results indicate that in addition to age, the level of frailty is a useful predictor of short-term COVID-19 outcomes in geriatric patients.

The SARS-CoV-2 virus pandemic, which causes coronavirus disease 2019 , has particularly 32 high morbidity among the older segment of the population. COVID-19 mortality rates show sharp 33 increase with increasing age; the vast majority of deaths in Sweden are reported among those aged 70 34 and older. There seems to be an increasing age-gradient and sex-difference among the ICU-treated 35 patients 1 , but the reasons for these variations in the clinical outcomes are not clear. It has been 36 speculated that certain somatic conditions like diabetes, hypertension and obesity are risk factors for 37 worse outcomes 2 . Due to the difficulties in identifying the underlying risk factors, restrictive 38 recommendations have been directed broadly to people older than 70 years of age irrespective of 39 health and activity status. 40

The older population is characterized by large heterogeneity in terms of health and vigor. Depending on 42 factors like life-style, socioeconomic status and genetic predisposition, health trajectories develop 43 differently between individuals. Irrespective of chronological age and concurrent disease, some age 44 faster and become vulnerable and susceptible to disease and disability earlier than others. In order to 45 recognize this condition, the concept of frailty has been introduced over the last decades 3 . WHO defines 46 frailty as "a progressive age-related decline of body functions resulting in vulnerability and reduced 47 resilience to physical and mental stressors with an increased risk of negative health outcomes" 4 . Frailty 48 can be assessed using various approaches, such as scales that take fitness and dependency into account. 49

The Clinical Frailty Scale (CFS) is one such tool that classifies subjects according to a 9-graded scale, from 50 very fit (=1) to terminally ill (=9) 5 . Another measure developed for identification of older hospitalized 51 patients with frailty is the Hospital Frailty Risk Score (HFRS), based on diagnostic codes in the 52

International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) 6 . 53

Although not yet confirmed in larger samples, there is preliminary evidence that the susceptibility to 54 J o u r n a l P r e -p r o o f COVID-19 and the increased mortality risk in older people is linked to frailty 7 . This is partly supported by 55 the high death tolls in residents of sheltered housing (e.g., nursing homes), where most residents are 56 frailer than community-dwelling older individuals. Other factors, such as group dining and staff 57 challenges may as well have contributed to the high death rates. 58

The Stockholm area has around 2 million inhabitants and was hit by COVID-19 in early spring 2020 with 60 more than 2000 casualties within the first two months of the outbreak. The aim of this article is to use 61 data from a large hospital in Stockholm with 967 patients treated at a geriatric ward unit during the first 62 three months of the COVID-19 outbreak, whereof 250 were diagnosed with COVID-19. The primary 63 hypothesis is that frailty according to CFS and the recently developed HFRS 6 are stronger risk factors for 64 negative outcomes than chronological age and comorbidities. 65 66

Patients admitted to the geriatric care unit at a large hospital in Stockholm, Sweden, from March 1 st to 69

June 11 th 2020 were included in the study. In total, 967 individuals with 250 COVID-19 patients, 232 70 (92.3%) confirmed (ICD-10 code U07.1) and 18 (7.2%) suspected (ICD-10 code U07.2), were studied for 71 short-term outcomes: i.e., in-hospital mortality or being discharged to home. Most patients came 72 directly from home and transitioned through the emergency section before being admitted to geriatric 73 care. Follow-up was done through electronic medical records throughout the study period, resulting in 74 different follow-up times for different patients. At a maximum, patients were followed up for up to 25 75 days for in-hospital mortality and up to 28 days for discharge to home. Patient demographics, diagnoses 76 and death data were collected through electronic health records in the Take Care® system, where all 77 deaths occurring in COVID-19 patients were assumed to be due to J o u r n a l P r e -p r o o f

A clinical COVID-19 diagnosis was determined using reverse transcriptase polymerase chain reaction 80 (RT-PCR) using extracts from nasopharyngeal swabs. Clinical diagnosis in patients with a negative RT-PCR 81 was made in collaboration with infection disease specialists if the patient had typical symptoms and 82 typical findings on a CT scan with no other explanation of the symptoms (i.e., bacterial infection). As 83 stated above, 18/250 (7.2%) of the COVID-19 patients were tested negative in the RT-PCR. In a preprint 84 systematic review, the false negative rate has been estimated between 2% and 29% 8 . The most likely 85 reasons for false negatives include too low amount of the virus for the test to detect it (e.g. the swab 86 not inserted deep enough in the nose), non-optimal sample type, a person is tested too early or too late 87 in the infection, technical issues in the assay or if the sample sits too long before being tested, allowing 88 the viral RNA to degrade 9, 10 . The false positive rate has also been shown to vary with time from 89 exposure and symptom onset 10 , making decision-making based on other assessments, such as the CT 90 scan pivotal. 91 92 Frailty 93

Frailty was defined using the CFS and HFRS. The CFS score was assigned by a geriatrician trained in 94 scoring with the CFS scale within 1-2 hours of admission. A chart review and face-to-face assessments 95 with patients and families were used to determine the CFS level as it was some time before admission. 96

The CFS scale ranges from 1 (very fit) to 9 (terminally ill) 5 . The CFS scores are commonly divided into the 97 following three categories: non-frail (CFS 1-4), mild-to-moderately frail (CFS 5-6), and severely frail (CFS 98 7-9). However, in the present study, we wished to test different cut-offs to determine the most 99 meaningful categorization for COVID-19 -related outcomes, and used dichotomization at CFS >5. 100 HFRS was assessed according to the model by Gilbert et al 6 using all available ICD-10 diagnostic codes 101 that were documented for the sample. The HFRS is a composite score based on 109 ICD-10 diagnostic 102 codes that have been associated with the risk of frailty 6 . Each of the 109 ICD-10 diagnostic codes are 103 assigned with a weight, ranging from 0.1 to 7.1, based on how strongly they associate with frailty, and 104 the weights are summed across the ICD-10 codes to yield the HRFS. In our analyses, we used the HFRS as 105 continuous measure. 106

Comorbidity 108

The different comorbidities were defined using ICD-10 codes for diabetes ( Differences across the study variables between COVID-19 patients and other patients, as well as 116 between men and women, were assessed using T-test, Mann-Whitney test and chi-squared test 117 statistics as appropriate. To analyze whether the CFS, HFRS, CCI and comorbidities (listed in Table 1 ) 118 were predictive of in-hospital mortality we performed Cox proportional hazard model analyses. For 119 being discharged to home, proportional subdistribution hazards model analysis 12 , where death is treated 120 as a competing risk, was used. We first tested the CFS, HFRS, CCI and the comorbidities individually in 121 univariate Cox models adjusted for age and sex. Those predictors that were significant in the univariate 122 models were further entered in multivariate models, again adjusting for age and sex. All the reported 123 variables were tested for the proportional hazard assumption. Harrell's C was used to assess the overall 124 predictive accuracy (discrimination) of each Cox proportional hazard model for mortality. R version 3.6.3 125 and the packages "survival" and "cmprsk" were used in the analysis. Significance level was set to α<0.05. 126

The study was approved by the Swedish Ethical Review Authority in Stockholm on April 14 th 2020 with 129

Dnr 2020-01497. 130

Sample characteristics 132

Characteristics of the study sample are presented in Table 1 . Proportions of in-hospital mortality in the 133 COVID-19 patients was 24% (59/250) and for being discharged to home 44% (110/250). Mortality rate 134 was higher among the COVID-19 patients, and they were younger, more frail, less likely to be discharged 135 to home and had higher prevalence of diabetes and hypertension, compared to the non-COVID-19 136 patients. Sex-stratified characteristics of the COVID-19 patients are presented in Supplementary Table 1 Survival and discharge to home regression analyses 146

Higher age was predictive of in-hospital mortality (Table 2 ) and decreased the probability of being 147 discharged to home (Table 3) . Being frail, corresponding to a CFS>5, and higher co-morbidity index were 148 both associated with in-hospital mortality (Table 2) . Being frail (CFS>5) was also associated with a 149 decreased probability of being discharged to home (Table 3) , whereas CCI was not (Table 3) . Higher 150 J o u r n a l P r e -p r o o f HFRS was not associated with in-hospital mortality (Table 2) but it was associated with a decreased 151 probability of being discharged to home (Table 3) . With the exception of AKI, which was associated with 152 in-hospital morality (Table 2) , and hypertension, which was associated with being discharged to home 153 (Table 3) , none of the comorbidities -diabetes, cardiovascular disease, chronic obstructive pulmonary 154 disease, asthma, chronic kidney disease, and dementia -were associated with either of the outcomes. 155

Stratifying patients based on usage of antihypertensive drugs (ATC code C09) demonstrated more 156 pronounced effects on mortality and being discharged to home (data not shown). Sex was not 157 associated with in-hospital mortality or discharge to home (Tables 2 and 3 ). Sex-stratified analyses for 158 in-hospital mortality are presented in Supplementary Table 2. CFS>5 and CCI conferred a relatively 159 greater risk in men, whereas AKI was a significant risk factor in women only. The Harrell's C indices for 160 the models indicated that age and sex ( The results of this study in hospitalized geriatric COVID-19 patients show that frailty assessed as CFS>5 166 and CCI were predictive of in-hospital mortality, whereas HFRS was not. Both CFS>5 and higher HFRS 167 were associated with decreased probability of being discharged back to home, but similar to the former 168 association, CFS presented the strongest effect. Apart from AKI, which was predictive of in-hospital 169 mortality, and hypertension, which was predictive of discharged to home, none of the individual 170 comorbidities were associated with either outcome. Higher age was also associated with in-hospital 171 mortality and decreased probability of being discharged to home; however, the model including only 172 age and sex (Model 1) had a poor predictive accuracy. Adding CFS>5, CCI and AKI to the model for in-173 hospital mortality increased its predictive accuracy to a fair level. A sex-stratified analysis further 174 showed that the CFS>5 and CCI were associated with greater risk of in-hospital mortality in men, 175

whereas AKI was a significantly greater risk factor in women. However, due to the relatively low power 176 in the sex-stratified analysis, larger studies are needed to confirm whether some of the risk factors for 177 COVID-19 -related outcomes are sex-specific. 178

To the best of our knowledge, our study is the first one to analyze the joint associations of frailty and as a continuous measure, we tested different cut-offs for the CFS to identify the clinically most 184 meaningful cut-off for increased risk. We observed that the risk for in-hospital mortality showed most 185 prominent increase from scores more than 5. Our results on CCI are in accordance with those of Bezzio 186 et al who analyzed COVID-19 outcomes in an Italian sample of inflammatory bowel disease (IBD) 187 patients with median age of 48. They found that in addition to older age and active IBD, CCI score>1 was 188 associated with negative outcomes, such as pneumonia, hospitalization, respiratory therapy and 189 death 13 . CCI was also assessed in a sample of patients hospitalized with COVID-19 in the New York City 190 area 14 . Although the prognostic value of CCI on the COVID-19 outcomes was not modeled, the authors 191 reported a median CCI score of 4 in the patients, which is seemingly higher than in our sample, and 192 reflects a significant comorbidity burden in hospitalized COVID-19 patients. It should be noted, however, 193 that our data included comorbidities only if they were of relevance to the current admission. Hence, it is 194 possible that our study has underreported other diagnoses, leading to less likelihood of detecting effects 195 from comorbidities on COVID-19 outcomes. Nevertheless, we found AKI to be associated with in-196 hospital mortality, with stronger effects in women. However, the acute state of AKI is likely an indicator 197 of the COVID-19 severity rather than a marker of an underlying (kidney) disease. 198

In a recent analysis on population vulnerability to COVID-19, Sweden was identified among the high-risk 200 countries. This was concluded based on the high proportions of older adults, and high rates of years 201 lived with disability due to medical conditions considered risk factors for severe COVID-19 15 . Frailty has 202 also been highlighted as one of the conditions posing an increased risk in aged individuals with 17 . It is therefore pertinent to set the focus on identifying those factors that predict COVID-19 -204 related outcomes in older populations. Especially for a new disease like COVID-19, studies improving 205 prognosis are urgently needed because there are no evidence-based clinical guidelines to follow 18 . 206

Although the overall COVID-19 death rate in our sample was high compared to the patients hospitalized 207 for other diagnoses (24% vs. 4%), and higher age was an independent risk factor for mortality, not all old 208 individuals had the same risk. That is, three quarters of the geriatric COVID-19 patients survived and 44% 209 were able to return home directly after hospitalization. Our results thus suggest that both CFS and CCI 210 can be used to complement risk assessments and identify geriatric patients who are in need of more 211 focused care. However, as noted in a recent commentary on frailty in the face of COVID-19 17 , frailty is 212 not synonymous to end-of-life. Hence, the knowledge now accumulating on how to treat hospitalized 213 COVID-19 patients should be used to improve survival of the most vulnerable individuals. 214 215 Our study has several strengths. Compared to other studies published on this topic, we use a large 216 sample of hospitalized geriatric patients including both those diagnosed with and without COVID-19. We 217 also include two different assessments of frailty, both the established CFS, and the more recently 218 developed HFRS, and assess comorbidity not only as individual disease diagnoses but also using the CCI. 219

In addition, we perform multivariate statistical modeling taking into account the competing risk of death 220 on discharged to home, and analyze the added predictive value of frailty and comorbidity using the 221

Harrell´s C statistic that facilitates interpretations. 222

Our study also comes with some limitations, and the results needs to be interpreted in light of these 224 limitations. Firstly, the population under study is drawn from a geriatric hospitalized sample of older 225 adults in need of hospital care. Hence, the results are not generalizable to the wider population of 226 Sweden and other older individuals. Secondly, because of the selected sample, some part of the analysis 227 may suffer from selection bias and effects such as collider bias -when a risk factor is interpreted as a 228 protective factor in multivariate models because of underlying correlations 19 -may be in place. For 229 example, we observe in our data that hypertension shows a protective effect on COVID-19 survival, 230 although this may be an artifact from correlations with the usage of antihypertensives or other factors. 231

Thirdly, we only report short-term outcomes as our data are collected from in-hospital records only. A 232 longer follow-up with record linkage is planned as a continuous study, and adding more variables to the 233 models may also improve the prognostic C-statistic value. Forth, as the CFS is assessed at admission, 234 recall bias may be an issue here. Finally, as alluded to above, diagnoses available in the electronic 235 medical records were only those related to the current admission. Hence, it is likely that we missed a 236 number of comorbidities in the COVID-19 patients, resulting in a lower CCI and less ability to test other 237 diseases as risk factors. 238 239

In conclusion, the results in this study show that 76% of COVID-19 patients survived, indicating that 241 providing hospital level care to frail older COVID-19 patients is not futile. While higher age is also 242 associated with in-hospital mortality and decreased probability of discharged back to home in geriatric 243 patients, including frailty and comorbidity assessments to the models improves their predictive 244 accuracy. Therefore, such assessments can help to better identify older COVID-19 patients who are at 245 risk of adverse outcomes and thus in need of a more multi-dimensional care, both from an acute care as 246 well as from a rehabilitation care perspective. 

The Swedish Intensive Care Registry

Prevalence of comorbidities and its effects in patients infected 273 with SARS-CoV-2: a systematic review and meta-analysis

Frailty in elderly people

A global clinical measure of fitness and frailty in 279 elderly people

Development and validation of a Hospital Frailty Risk 281

Score focusing on older people in acute care settings using electronic hospital records: an 282 observational study

Frailty and mortality in hospitalized older 284 adults with COVID-19: retrospective observational study

FALSE-NEGATIVE RESULTS 287 OF INITIAL RT-PCR ASSAYS FOR COVID-19: A SYSTEMATIC REVIEW

Real-time RT-PCR in COVID-19 detection: issues affecting the results

Variation in False-Negative Rate of Reverse 292

Transcriptase Polymerase Chain Reaction-Based SARS-CoV-2 Tests by Time Since Exposure. Ann 293 Intern Med 2020

A new method of classifying prognostic comorbidity in 295 longitudinal studies: development and validation

A Proportional Hazards Model for the Subdistribution of a Competing Risk

Outcomes of COVID-19 in 79 patients with IBD in Italy: an 299 IG-IBD study

Presenting Characteristics, Comorbidities, and 301 Outcomes Among 5700 Patients Hospitalized With COVID-19 in the New York City Area

Population vulnerability to COVID-19 in Europe: a 304 burden of disease analysis

Decision-Making in COVID-19 and Frailty

Frailty in the Face of COVID-19

Making the Call

Collider bias undermines our understanding of COVID-19 310 disease risk and severity

The authors are supported by grants from the Swedish Research Council, the Stockholm County Council (ALF Project and SU-Region Stockholm Project), Stiftelse Stockholms Sjuhem, Knut and Alice Wallenberg Foundation, Karolinska Institutet, King Gustaf V:s and Queen Victorias Freemason Foundation, Osterman Foundation, the Strategic Research Program in Epidemiology at Karolinska Institutet. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.