Predictors of survival in elderly patients with COVID-19 admitted to the hospital: derivation and validation of the FLAMINCOV score.

OBJECTIVE: To identify predictors of 30-day survival in elderly patients with COVID-19.
METHODS: Retrospective cohort study including COVID-19 patients≥65 years old hospitalized in 6 European sites (January 2020-May 2021). Demographics, comorbidities, clinical characteristics and outcomes were collected. A predictive score (FLAMINCOV) was developed using logistic regression. Regression coefficients were used to calculate the score. External validationina cohort including elderly patients from a major COVID-19 center in Israel was performed. Discrimination was evaluated by the area under the receiver operating characteristic curve (AUC)in the derivation and validation cohorts. Survival risk groups based on the score were derived and applied to the validation cohort.
RESULTS: Among 3010 patients included in the derivation cohort, 30-day survival was 74.5% (2242/3010). Intensive care unit (ICU) admission rate was 7.6% (228/3010).The model predicting survival included independent functional status (OR 4.87, 95%CI 3.93-6.03), SpO2/FiO2 ratio>235 (OR 3.75, 95%CI 3.04-4.63), C-reactive protein<14 mg/dl (OR 2.41, 95%CI 1.91-3.04), creatinine<1.3 (OR 2.02, 95%CI 1.62-2.52) mg/dl and absence of fever (OR 1.34, 95%CI 1.09-1.66). The score was validated in 1174 patients. The FLAMINCOV score ranges from 0 to 15 and showed good discrimination in the derivation (AUC 0.79, 95%CI 0.77-0.81, p<0.001) and validation cohort (AUC 0.79, 95%CI 0.76-0.81, p<0.001). Thirty-day survival ranged from 39.4% (203/515) to 95.3% (634/665)across four risk groups according to scorequartiles in the derivation cohort. Similar proportions were observed in the validation set..
CONCLUSIONS: The FLAMINCOV score identifying elderly with higher or lower chances of survival may allow better triage and management, including ICU admission/exclusion.

INTRODUCTION

Since the start of the COVID-19 pandemic, elderly were identified as one of the most vulnerable patient groups [1, 2]. Mortality rates change across age categories, ranging from 9.5% in patients 60-69 years old up to 29.6% in those aged >80 years [3]. The highest mortality rates are reported in elderly patients admitted to intensive care unit (ICU) [4, 5]. Thus, intensivists were initially discouraged to admit elderly patients to ICU and age has been often considered the only determining factor in ICU triage decision. This approach raised ethical concerns, since the poor outcome of elderly reported in some studies have been related to the delayed ICU admission of these patients [6]. The clinical frailty scale (CFS) seems to better predict the outcome of elderly patients instead of age itself [7]. However, the use of CFS for critical care decision has been debated since mildly frail older adults may still have enough intrinsic capacity to withstand the stressors of hospitalization and achieve clinical cure [8]. Thus, the identification of elderly patients with COVID-19 who have higher chance of survival might be useful to better decide treatments and allocation of these patients, while reducing the risk of therapeutic obstinacy in those with reduced probability to recover.

The aim of our study was to identify predictors of 30-day survival in a large cohort of elderly patients with COVID-19 and stratify patients according to their probability to survive.

METHODS

Patient cohort and study design

This is a retrospective study including hospitalized patients≥65 years old with COVID-19 in 6 sites (University hospital of Pisa, Italy; Rabin Medical Center, Beilinson Hospital, Israel; Istituto Auxologico Italiano, Milan, Italy; Hospital of Modena, Italy; San Raffaele Scientific Institute, Milan, Italy; Geneva University Hospitals, Switzerland) from January 2020 to May 2021. Inclusion criteria were: 1) age ≥65 years old; 2) laboratory-confirmed COVID-19, diagnosed by a positive SARS-CoV-2 real-time polymerase chain reaction test on a nasopharyngeal swab. Model development and reporting followed the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement and published recommendations for prediction models [9, 10]. Patients who met the inclusion criteria constituted the population of interest and were included in the derivation set to develop a score for survival prediction in elderly patients. The derived score was validated in a validation set of elderly patients hospitalized inthe Rambam Health Care Campus, Haifa, Israel, from April 2020 to January 2022. The same inclusion criteria used for the derivation population were used to select patients in the validation set.

The protocol was approved by the medical ethics committee of Area Vasta Nord Ovest (ID 19283) and the institutional ethics review boards of participating hospitals. Written informed consent was obtained from participants according to local rules.

Data collection and Potential Predictive Variables

Epidemiological and demographic information, medical history, comorbidities, information on clinical symptoms on admission, treatments, and interventions received during the hospital course, including need for oxygen or invasive mechanical ventilation support, were collected from medical records using a prespecified case report form.

The functional capacity was evaluated according to patient's ability to perform activities of daily living by using the Norton scale on admission (Supplementary Table 1) [11]. Clinical signs and symptoms included fever (body temperature >38 °C), dyspnea, and confusion/altered mental status on admission. SpO2 values on admission were collected and SpO2/FiO2 was also calculated [12]. A SpO2/FiO2 ratio was categorized as > or ≤235, since its correlation with a PiO2/FiO2 ratio > or ≤ 200 [13]. Laboratory findings on admission included white blood cell count, lymphocyte, platelet counts, C-reactive protein, procalcitonin, D-dimer and ferritin levels. Data about treatments (steroids, immunosuppressive drugs) and interventions (low-flow oxygen, high-flow oxygen therapy, non-invasive and invasive mechanical ventilation) were collected.

The clinical information used to calculate prognostic score was taken on the day of admission to hospital.

Outcome

The primary outcome measure was 30-day survival.

Statistical analysis

Continuous variables are presented as medians and interquartile ranges (IQRs). Categorical variables are presented as frequencies and proportion. The comparison between patients who survived and those who did not was performed using the Mann Whitney U test, Pearson’s Chi-squared test, or Fisher’s exact test, as appropriate. Continuous variables were dichotomized according to Classification and Regression Tree analysis, apart from SpO2/FiO2 ratio (categorized as > or ≤ 235) and the PiO2/FiO2 ratio (categorized as > or ≤ 200).

To explore factors associated with survival, univariable and multivariable logistic regression models were used. A multivariable analysis was performed to identify factors independently associated with 30-day survival using a forward regression model. Variables with statistical significance (p<0.05) on univariate analysis were included in the multivariable model. Details about included variables and score selection are reported in Supplementary Materials.Odds ratio (OR) and 95% confidence interval (CI) were calculated.

All patients from the 6 participating centers were included in the derivation cohort. The predictive score (FLAMINCOV) was developed using the regression coefficients as in Sullivan’s scoring system by dividing each regression coefficient by the smallest and rounding to the nearest unit. Imputation for missing variables was considered if missing values were less than 20%. We assessed discrimination by using the area under the receiver operating characteristic curve (AUC). A value of 0.5 indicates no predictive ability, 0.7 to 0.8 is considered good, 0.8 to 0.9 is considered excellent, and more than 0.9 is considered outstanding [14]. The Lemeshow– Hosmer goodness-of-fit test was used to evaluate calibration. The cohort was split to quartiles based on the regression probabilities and survival rates were calculated for the 4 risk groups both in the derivation and validation cohorts.

The variables required for calculating the FLAMINCOV score were collected for external validation. The AUC, Lemeshow– Hosmer goodness-of-fit test and 30-day survival by risk groups, defined by the same score thresholds as in the derivation cohort,were calculated also for the validation cohort. All statistical analyses were performed using IBM SPSS Statistics, version 27.0 (IBM Corp., Armonk, NY, USA), and were considered signifcant at p<0.05 (two-tailed).

RESULTS

Study population

A total of 3010 elderly patients were included in the derivation cohort. Table 1 shows demographic and clinical characteristics for the derivation cohort. The median age was 77 years (interquartile range [IQR] 70-84); 42.2% of patients were women. The 30-day survival rate was 74.5% (2242/3010 patients). A progressive increase in 30-day mortality rates was observed during the course of the study (Figure 1 ). Overall, 228/3010 (7.6%) patients were hospitalized in intensive care unit (ICU). Time from Emergency Department (ED) admission to ICU transfer was 2 (IQRs 0-6) days. Data about coinfections on hospital admission were available in 1886/3010 (62.6%) patients. Among them, coinfections were detected in 7.4% of patients (140/1886) and represented by respiratory bacterial infections (n=101/1886, 5.4%), urinary tract infections (n=19/1886, 1%) and other types (1 Clostridium difficile infection, 1 Enterococcus faecalis bacteremia).

TABLE 1

Comparison of elderly patients with COVID-19 who died and those who did not within 30-day from hospital admission.

	TotalN=3010	30-day survivorsN=2242	Non-survivorsN=768	p value
Age, median (IQRs)	77 (70-84)	75 (69-82)	82 (75-87)	<0.001
Female sex, n (%)	1270 (42.2%)	982 (43.8%)	288 (37.5%)	0.002
Functional status, n/N (%)Independent functional status	1454/2879 (50.5%)	1275/2112 (60.4%)	179/767 (23.3%)	<0.001
BMI, median (IQRs)	27 (24-30.5)	27.9 (24.8-31.2)	25.9 (23.3-28.9)	<0.001
Comorbidities, n (%)Diabetes mellitusCardiovascular diseaseHypertensionCerebrovascular diseaseChronic pulmonary diseaseChronic kidney diseaseChronic liver diseaseSolid cancer	861 (28.6%)1090 (36.2%)1256 (41.7%)311 (10.3%)307 (10.2%)319 (10.6%)38 (1.3%)415 (13.8%)	615 (27.4%)762 (34%)946 (42.2%)218 (9.7%)220 (9.8%)189 (8.4%)29 (1.3%)295 (13.2%)	246 (32%)328 (42.7%)310 (40.4%)93 (12.1%)87 (11.3%)130 (16.9%)9 (1.2%)120 (15.6%)	0.015<0.0010.3750.0610.231<0.0010.7940.087
Immunosuppressive treatment before admission, n (%)	355 (11.8%)	260 (11.6%)	95 (12.4%)	0.567
Clinical presentation on admission, n/N(%)Absence of feverNormal mental statusAbsence of dyspneaPaO2/FiO2 >200SpO2/FiO2 >235	2091/2988 (70%)2562 (85.1%)1925 (64%)1561/2791 (55.9%)1495 (49.7%)	1603/2228 (71.9%)2032 (90.6%)1561 (69.6%)1350/2023 (66.7%)1303 (58.1%)	488/760 (64.2%)530 (69%)364 (47.4%)211/768 (27.5%)192 (25%)	<0.001<0.001<0.001<0.001<0.001
Physical examination on admission, n/N (%)Absence of hypotensionNo tachycardia (HR<100)No tachypnoea (RR<20)SOFA score, N=2746	2815/2959 (95.1%)2370/2896 (81.8%)2132/2747 (77.6%)3 (3-4)	2126/2191 (97%)1787/2182 (81.9%)1731/2173 (79.7%)3 (3-4)	689/768 (89.7%)583/714 (81.7%)401/574 (69.9%)3 (3-4)	<0.0010.883<0.0010.716
Laboratory exams at EDCreatinine, mg/dl, median (IQR)Creatinine <1.3 mg/dlLymphocytes >800/mcLPlatelet count>150x103/mcLFerritin <1325 ng/mLD-dimer<1650mg/LC-reactive protein<14 mg/dl	1 (0.8-1.3)2385 (79.2%)2138 (71%)2295/2918 (78.6%)1183/1431 (82.7%)1307/1949 (67.1%)2086/2613 (79.8%)	1 (0.9-1.2)1886 (84.1%)1651 (73.6%)1744/2154 (81%)906/1068 (84.8%)1037/1421 (73%)1617/1919 (84.3%)	1 (0.8-1.7)49 (65%)487 (63.4%)551/764 (72.1%)277/363 (76.3%)270/528 (51.1%)469/694 (67.6%)	<0.001<0.001<0.001<0.001<0.001<0.001<0.001
COVID-19 treatmentLow molecular weight heparinRemdesivirSteroidsConvalescent plasmaImmunomodulatory drugsa	2496 (82.9%)489 (16.2%)1700 (56.5%)58 (1.9%)365 (12.1%)	1861 (83%)380 (16.9%)1195 (53.3%)44 (2%)266 (11.9%)	635 (82.7%)109 (14.2%)505 (65.8%)14 (1.8%)99 (12.9%)	0.8370.074<0.0010.8080.452

BMI body mass index, ED Emergency Department, IQRs interquartile ranges.

p value calculated using Mann-Whitney U test for continuous variables and Chi square test for categorical variables; p values <0.05highlighted in bolda either tocilizumab or baricitinib.

Figure 1

(legend). Thirty-day mortality rates across different time period.

Compared to patients who survived within 30 days from admission, non-survivors were significantly older and more frequently males (Table 1). Dependent functional status, diabetes mellitus, cardiovascular disease and chronic kidney failure were more common in non-survivors. Survivors were more likely to present with no fever, normal mental status, no dyspnea and a SpO2/FiO2>235 on hospital admission. Furthermore, survivors had less frequenlty lymphopenia, thrombocytopenia, high ferritin and C-reactive protein values on admission (Table 1).

Determinants of survival and derivation of the FLAMINCOV score

The FLAMINCOV score was derivated in 2586/3010 (85.9%) patients with complete data. The study flow chart is reported in Figure 2 . Comparison between patients with missing data and those included showed that missing data occurred more commonly among younger patients and those with less comorbidities; 30-day survival was higher among patients with missing data (Supplementary Table 2). The 30-day survival rate was 73.5% (1901/2586).

Figure 2

(legend). Study flow chart (both derivation and validation cohort).

On multivariable analysis (Table 2 ), independent functional status (OR 4.87, 95% CI 3.93-6.03, p<0.001), SpO2/FiO2 ratio>235 (OR 3.75, 95% CI 3.04-4.63, p<0.001), C-reactive protein <14 mg/dl (OR 2.41, 95% CI 1.91-3.04, p<0.001), creatinine <1.3 mg/dl (OR 2.02, 95% CI 1.62-2.52, p<0.001) and absence of fever (OR 1.34, 95% CI 1.09-1.66, p=0.006) were factors independently associated with 30-day survival. These findings were confirmed also considering time periods and center as variables in the multivariable model (Supplementary Table 3).

TABLE 2

Multivariate logistic regression analysis of factors independently associated with 30-day survival and score points.

	β coefficient	OR (95% CI)	p value	Points‐based risk score
Independent functional status	1.584	4.87 (3.93-6.03)	<0.001	+5
SF>235	1.322	3.75 (3.04-4.63)	<0.001	+4
CRP <14mg/dl	0.881	2.41 (1.91-3.04)	<0.001	+3
Creatinine <1.3 mg/dl	0.705	2.02 (1.62-2.52)	<0.001	+2
Absence of fever	0.296	1.34 (1.09-1.66)	0.006	+1

Multivariable analysis performed using a forward regression model. Variables entered but not retained: age, female sex, diabetes mellitus, normal mental status.

CRP: C-reactive protein; CI confidence interval; SF SaO2/FiO2; OR odds ratio.

Table 2 shows the FLAMINCOV score and the designation of points. The score ranged from 0 to 15. The AUC of our model was 0.79 (95% CI 0.77–0.81, p<0.001) (Figure 3 , panel A). The goodness-of-fit HosmerLemeshow χ2 was 3.6 (p = 0.822), indicating a good calibration.

Figure 3

(legend). ROC curve of the FLAMINCOV score in the derivation cohort (panel A) and in the validation cohort (panel B).

The FLAMINCOV score was classified into 4 risk groups according to percentiles of the score: 1) risk group 1 (score ≤5, observed 30-day survival 39.4%), risk group 2 (score 6-9, observed survival 65.8%), risk group 3 (score 10-11, observed survival 85.9%) and risk group 4 (score 12-14, observed 30-day survival 95.3%). Survival rates across the different risk groups are reported in Figure 4 .

Figure 4

(legend). Thirty-day rates across different risk groups of the FLAMINCOV score (the risk groups were calculated according to the percentiles of the score) in the derivation and in the validation cohort.

1: low probability of survival; 2 low-intermediate probability of survival; 3: low-high probability of survival; 4: high-probability of survival.

External validation

The external population from the Rambam Health Care Campus, Haifa (Israel) included 1342 elderly patients. The FLAMINCOV score was validated in 1174/1342 (87.5%) patients with complete data (Figure 2). Comparison between patients with missing data and those included showed that missing data occurred among patients with less comorbidities, but lower 30-day survival and short time to death (Supplementary Table 4). The 30-day survival rate was 68.1% (799/1174 patients). The AUC of the model was 0.77(95% CI 0.75–0.8, p<0.001) (Figure 3, panel B). The goodness-of-fit HosmerLemeshow χ2 was 7.9 (p= 0.340), indicating reasonable calibration. When applying the FLAMINCOV score risk group definitions in the validation cohort, 30-day survival rates were: 38.4% in risk group 1, 60.2% in risk group 2,77.8% in risk group 3 and 94.5% in risk group 4 (Figure 4).

DISCUSSION

In this multicenter observational cohort study we propose the FLAMINCOV score to predict 30-day survival of elderly patients with COVID-19 and guide clinicians to their optimal management and allocation. The score comprises of variables easily obtainable at the ED, prior to patient triage.

During the first wave when hospital faced with significant challenges, elderly patients were usually excluded from ICU care because advanced age appeared strongly associated with poorer outcomes [15]. In a survey from 21 countries, one third of the responders declared that elderly patients were not candidate to the ICU care in their hospitals [16]. The allocation of patients based only on age generated some concerns. There is evidence that age on its own can be misleading in outcome prediction, and scientific societies advocated the use of CFS in clinical decisions for elderly patients with COVID-19 [8, 17]. However, the use of frailty alone as instrument to decide the patient allocation is questionable, since in many studies categorization of CFS was arbitrary. Compared to the Norton scale used in our study, CFS is more specific to evaluate frailty, but it has been not specific for patients with COVID-19 in this special population.

In our study the absence of dependency, and not age itself, is the most important factor associated with survival. Several studies highlighted the importance to prioritize the functional capacity as a principal endpoint in elderly care admitted to the hospital for different diseases [18, 19]. Among different available scales the Norton scale is a simple assessment tool traditionally used for the assessment of the risk of pressure ulcers, but may be useful to predict also other complications and in-hospital mortality [20]. Recent single-center observational studies including 186 and 375 elderly patients with COVID-19, respectively, showed that functional status predicts death in hospitalized patients with COVID-19 [21, 22]. The poor prognosis of elderly patients with COVID-19 and functional dependency may be related to different factors, including comorbidity burden, poor nutritional status, and impaired cell-mediated immunity, cytokine production and phagocytosis[23].

We found that increase in CRP serum levels were associated with higher risk of mortality. Several factors may explain this correlation: 1) CRP is directly related to the production of IL-6, that might reflect the activation of immune response and cytokine storm [24], 2) the inflammatory state illustrated by CRP elevation may reflect the activation of coagulation cascade and pro-thrombotic state [25, 26], 3) CRP may be the marker of a pre-existing chronic activation of innate immune system.[27].

The FLAMINCOV score has some strengths. It allows to predict 30-day survival according to different classes of risk: low, intermediate-low, intermediate-high and high probability to survive. Patients were well distributed across risk classes both in the derivation and in the validation cohort. Thus, it may be useful from a clinical point of view, because it well reflects the variety of patients admitted to the hospital. This stratification may support clinicians because allows the identification of elderly with high or low chances to survive. Patients with less severe disease are usually at low risk for a fatal outcome and not candidates for ICU. Conversely the FLAMINCOV score may be more useful in patients with severe COVID-19. In this category, a low FLAMINCOV score may help physicians to exclude patients from intensive care, while patients included in the intermediate or high FLAMINCOV classes (e.g. a patient with independent functional status and no significant increase in inflammatory markers) should not be excluded only on an age-based evaluation. Surprisingly, we found an increase in mortality across 3 different periods. Although this is not an objective of the study, it should be underlined that this finding is in line with previous ones [6]. This may be due to several reasons, including changes in treatment of COVID-19 with widespread use of steroids, that remains debated in elderly [2], and reduced access to intensive care with the increase in the absolute number of COVID-19 patients and reduced hospital capacity.

Our study has several limitations. First, the retrospective nature of this study may have affected data collection. There are some missing data (for example, procalcitonin values were unavailable for a large percentage of patients and we cannot establish if this is a prognostic biomarker tooand data about “Do not resuscitate order” were not available for all centers). However, we excluded cases with missing data and provided their description. Second, the evaluation of functional status at the ED was not a multiparametric assessment and we use the Norton scale, that is not a specific scale to evaluate frailty [28]. Anyway, this reflects the real world experience and we tried to provide a specific tool for elderly with COVID-19 admitted to the ED. Third, the vaccination campaigns together with the spread of new variants of concern changed the severity of COVID-19 in elderly patients. In this context, the FLAMINCOV score may have a reduced applicability. However, the access to vaccination is not equal all over the world, and as Omicron spreads globally, the majority of people in low-income countries remain unvaccinated and unprotected against COVID-19. Moreover, elderly patients may have a reduced response to vaccine and continue to represent subjects at high risk of progression [29]. Finally, the score didn’t consider therapies and some heterogeneity in the COVID-19 treatment might have been among centers. Some treatments, such as steroids, may affect parameters of the score including CRP. Although we collected all laboratory findings on admission, we did not take into account the impact of treatments started at home. A further study should be planned to evaluate the impact of treatments of the outcome of this special patients population.

In conclusion, we developed the FLAMINCOV score to identify elderly patients with COVID-19 with different probabilities to survive. The FLAMINCOV may be useful to better triage elderly with severe COVID-19 allowing the identification of patients with low and those with intermediate or high probabilities to survive that may beneficiate from intensive care.

Conflict of interest

MF received grants and/or has been on the speakers' bureau from Angelini, Menarini, Pfizer, TermoFisher, GSK, MSD, Gilead, Shionogi. All conflict of interests are outside the submitted work and did not influence its results. All other authors have no conflicts of interest to declare.

Funding

no funding.

Contribution

MF, DY, CM, VP, MT, MP designed the study; GT created the case report form and developed the database; GT, MI, RM, BV, GH, PV, RN, DY, PGB, GA, GL collected data; GTanalysed and interpreted data; GT, MF and MP performed the statistical analysis of data; GT and MF wrote the manuscript; MT, AC, CM, DY, MP and MF revised and contributed to the critical revision of the final manuscript.