Mortality reduction in older COVID-19-patients hospitalized in Spain during the second pandemic wave from the SEMI-COVID-19 Registry

In 2020, the COVID-19 pandemic followed a two-wave pattern in most countries. Hospital admission for COVID-19 in one wave or another could have affected mortality, especially among the older persons. The objective of this study was to evaluate whether the admission of older patients during the different waves, before SARS-CoV-2 vaccination was available, was associated with a different mortality. We compared the mortality rates of patients hospitalized during 2020 before (first wave) and after (second wave) July 7, 2020, included in the SEMI-COVID-19 Registry, a large, multicenter, retrospective cohort of patients admitted to 126 Spanish hospitals for COVID-19. A multivariate logistic regression analysis was performed to control for changes in either the patient or disease profile. As of December 26, 2022, 22,494 patients had been included (17,784 from the first wave and 4710 from the second one). Overall mortality was 20.4% in the first wave and 17.2% in the second wave (risk difference (RD) − 3.2%; 95% confidence interval (95% CI) − 4.4 to − 2.0). Only patients aged 70 and older (10,973 patients: 8571 in the first wave and 2386 in the second wave) had a significant reduction in mortality (RD − 7.6%; 95% CI − 9.7 to − 5.5) (unadjusted relative risk reduction: 21.6%). After adjusting for age, comorbidities, variables related to the severity of the disease, and treatment received, admission during the second wave remained a protective factor. In Spain, patients aged 70 years and older admitted during the second wave of the COVID-19 pandemic had a significantly lower risk of mortality, except in severely dependent persons in need of corticosteroid treatment. This effect is independent of patient characteristics, disease severity, or treatment received. This suggests a protective effect of a better standard of care, greater clinical expertise, or a lesser degree of healthcare system overload.


Clinical differences in patients ≥ 70 years hospitalized for COVID-19 by survival status
A univariate analysis of mortality was performed.Almost all variables were statistically associated with mortality, denoting the large sample size.Table 2 shows data on demographics, clinical manifestations, laboratory findings, and treatments received.Although some of the associations were strong, the effects were often small.
The variables and interactions included in the maximal logistic regression model estimated are shown in Table 3.
The odds ratios and risk ratios of mortality of being admitted in the second wave versus the first one for the four different combinations of interaction covariates values are shown in Table 4.The protective effect of being admitted during the second wave of COVID-19 is highest for patients without severe dependence who are not treated with corticosteroids, and it diminishes when either of these circumstances is present, becoming neutral when both converge.

Discussion
This study confirms a difference in mortality in patients hospitalized for COVID-19 in Spain between the first and the second wave of the pandemic.The lower mortality rate observed in the second wave is due to the lower mortality in patients ≥ 70 years; no differences in mortality were observed among younger subjects.To our knowledge, this is a novel finding that has not been previously described.
This reduced mortality rate found in older patients hospitalized during the second wave of the COVID-19 pandemic compared to the first wave could be due to some unmeasured or unknown confounders which may be broadly grouped into three categories: differences in the patients, differences in the disease, or differences in treatment and overall management.
The second-wave patients in this study were slightly older, had a higher degree of dependence (as measured by Barthel index), and a greater comorbidity burden.Given that these factors are associated with a worse COVID-19 prognosis, the clinical and epidemiologic differences between patients hospitalized in the first and the second wave 20 do not explain the lower mortality observed in the second wave.
Although patients admitted during the second wave were slightly less severe at admission, after adjustments in the multivariate analysis, admission during the second wave remained an independent protective factor.In the second wave, the use of treatments that have been shown to reduce COVID-19 mortality increased, namely corticosteroids 21 , tocilizumab 22,23 , and remdesivir 24 .However, the lower mortality in the second wave was unchanged after adjusting for the use of these therapies.
Our results suggest that there may be some factor (or, more probably, combination of factors), associated with hospitalization that influences mortality and changed between the waves.Some potential candidates are changes in the overall management of patients, improvements in clinical expertise, and a lesser degree of hospital overload (Fig. 2).
Interestingly, the analysis of the interactions between severe dependency and systemic corticosteroid treatment with the hospital admission wave has allowed us to discover that the protective effect of the second wave on mortality is highest when both factors are absent, somewhat reduced when either of them is present, and neutral when both are present.It is likely, then, that the factors, largely unknown, that explain an average protective effect of the second wave, may not succeed in improving the prognosis of the most fragile and severe patients, such as those requiring corticosteroid treatment.www.nature.com/scientificreports/ We have learned that there are a lot of "intangibles" that influence the prognosis of COVID-19 hospitalization.Quick identification of respiratory failure, thromboprophylaxis and early mobilization, an adequate state of hydration, proper management of stress hyperglycemia, nutritional support, physical rehabilitation, and more have become the new standard of care and are potential uncontrolled factors that could explain the better prognosis in the second wave.Most of these factors will have a greater impact on the older people, as they are frailer and thus prone to physical deconditioning, dehydration, or confusional states.
Another possibility is that greater clinical expertise led to improved prognosis.Indeed, in the USA, a progressive decline in COVID-19-related mortality was described after the passage of just a few months 9 .
Healthcare system overload could be an important driver of mortality in the COVID-19 pandemic [25][26][27] .The first wave in Spain was explosive and overwhelmed hospitals in some areas.For instance, in Madrid, COVID-19 occupancy reached nearly 300% of the nominal ICU capacity and nearly 105% of the general ward capacity 28 , paralyzing non-emergency surgical procedures.On the contrary, the second wave has been less dramatic, leading to a smaller impact on hospital occupancy and healthcare activity.It may well be that the lower mortality in the second wave is mainly a reflection of less healthcare system overload.As our study does not include data on the true workload borne by the hospitals, this notion remains a hypothesis.
In extreme cases, healthcare system overload leads to shortages, which can also have a greater effect on the older persons due to implementation of triage criteria.If it were confirmed that healthcare system overload causes greater mortality in the older persons, it would be a moral imperative for us as a society to quickly adopt robust preventative measures as soon as another wave is upon us and there is risk of healthcare system overload.
Our registry cannot answer this crucial question, as we lack data on hospital or ICU patient loads at the time of the patients' admissions.
We recognize several limitations in our study.The large number of researchers involved and/or variability in the availability of data from each hospital could have led to information bias.Selection bias could have been introduced given the voluntary participation of each center.A potential source of uncontrolled confounding factors is the severity of underlying conditions and overall frailty of patients.The lower mortality in the second wave could also be explained by a "harvesting effect" that may be present if the most severely ill patients had already died in the first wave, though patients admitted during the second wave were older and had more dementia and comorbidities.However, we did not analyze frailty, a prognostic factor that is more potent than dependence or age in older patients 20 .
In terms of limitations regarding treatment-related variables, the effect of remdesivir and tocilizumab on mortality are strongly time-dependent with a narrow window of opportunity and both tocilizumab and corticosteroids are indicated for a worsening respiratory or inflammatory condition.Our registry includes data on the timing of the drug initiation but does not include clinical and laboratory findings at that moment, so it is not Malignancy: solid tumor, leukemia, lymphoma; f CKD: chronic kidney disease (patients on dialysis or with serum creatinine > 3 mg/dL); g Arterial stiffness: pulse pressure ≥ 60 mmHg; h Tachypnea: > 20 breaths per minute; i Fever: temperature > 37.8 °C; j Tachycardia: heart rate > 100 bpm; k Inflammatory pattern: see description in "Methods".Categoric variables are expressed as percentages and compared using likelihoodratio chi-square test for statistical significance.OR = odds ratio; 95% CI OR: 95% confidence interval for odds ratio.The odds ratios have been calculated with respect to the first category.Quantitative variables are expressed as mean (standard deviation) or median [interquartile range] and were compared for statistical significance using Student's t-test (with equal or unequal variances) or Mann-Whitney U test, as appropriate.www.nature.com/scientificreports/possible to evaluate the exact effect of the drugs.The deleterious effect of corticosteroids or tocilizumab in our multivariate analysis should be interpreted as a marker of the patient's worsening condition.
The strengths of this study include its multicenter, nationwide design as well as the large number of patients included, which provides strong statistical power.The consecutive inclusion of patients in each center limits selection bias.
In conclusion, mortality in the older patients hospitalized in Spain with COVID-19 has been significantly lower in the second wave even after adjusting for baseline clinical condition, disease severity upon admission, and pharmacological treatment with proven benefits in treating COVID-19, except in severely dependent persons in need of corticosteroid treatment.Our results suggest that this reduction of mortality could be related to a better standard of care, improvements in clinical expertise, less healthcare system overload, or a combination of these three factors, though other unknown confounding factors cannot be ruled out.

Study design
This is a retrospective cohort study comparing the first and second waves of the COVID-19 epidemic in Spain.The first wave was defined as the period between January 1 and July 7, 2020.The second wave was defined as the period between July 8, 2020 and December 26, 2020, before SARS-CoV-2 vaccination was available.
The final weeks of the first wave and the initial weeks of the second one thus defined periods with a low incidence of COVID-19 and few hospital admissions.However, this cut-off point reflects the transition from the greater healthcare system overload which occurred in the initial months to the lesser healthcare system overload of the later months.

Registry design
The SEMI-COVID-19 Registry is an ongoing, nationwide, retrospective cohort launched in March 2020 that comprises most consecutive patients hospitalized in Spain who are discharged with confirmed COVID-19 disease.It has become one of the largest repositories of COVID-19 patient data and includes more than 20,000 patients to date.Its characteristics have been thoroughly described elsewhere 6 .
Inclusion criteria for the registry were age ≥ 18 years and first hospital discharge with a confirmed diagnosis of COVID-19.Exclusion criteria were subsequent admissions of the same patient and denial or withdrawal of informed consent.
Consecutive patients who required hospital admission and who had SARS-CoV-2 infection confirmed by a positive result on real-time polymerase chain reaction (RT-PCR) testing of a nasopharyngeal, bronchoalveolar lavage, or sputum sample and who provided verbal consent were included in the registry.With the advent of the second wave, the inclusion criteria were expanded with two modifications: antigen testing was accepted as a method for confirming diagnosis and reinfections (> 3 months from the initial infection) of the same patient were accepted for inclusion.From March 23, 2020, to December 26, 2021, a total of 22,494 patients from 126 hospitals throughout the country were included in the registry.Patients were treated at their attending physician's discretion according to local protocols and clinical judgment.Patients included in open-label clinical trials could be included in the registry provided that all information about treatment was available.Due to its observational nature, the registry caused no inconvenience to the patients included.

Data collection
Clinical investigators all over the country collected data from medical records using a standardized online data capture system (DCS).The DCS includes both a database manager and the set of procedures for the verification of data.Patient identifiable data are dissociated and pseudonymized using an alphanumeric sequence and each researcher keeps a protected registry (patient log) for the purpose of data verification and quality control.The database platform is hosted in a secure server and both the database and each client-server transfer are encrypted.The pseudonymization system allows for safeguarding patient privacy while also complying with ethical considerations and data protection regulations.
Data on more than 300 variables are collected retrospectively after patient discharge and grouped under various headings: inclusion criteria, epidemiological data, RT-PCR and serology data, prior comorbidities and medication history, findings (symptoms and physical examination) at admission, laboratory (blood gases, metabolic panel, complete blood count, coagulation) and diagnostic imaging tests at admission, additional data at seven days after admission or at admission to the intensive care unit, pharmacological treatment (antiviral drugs, immunomodulators, antibiotics) and ventilatory support during the hospitalization, complications during the hospitalization, and progress after discharge and/or 30 days from diagnosis.The variables in the registry have previously been described 6 .
A number of secondary variables were calculated from the primary variables in the registry.Some qualitative variables were classified into binary categories whereas some quantitative variables were categorized as normal or abnormal; age was categorized into decades.Arterial stiffness was defined as a pulse pressure greater than or equal to 60 mmHg 29 .Blood glucose levels were categorized into three groups according to standard glycemic targets in hospitalized patients: < 140 mg/dl, 140-180 mg/dl, and > 180 mg/dl 30 .The risk categories based on the pattern of inflammation used in this study were a modified version of risk categories recently reported in another work from the SEMI-COVID-19 Registry 31 .The low-risk category was defined as lactate dehydrogenase (LDH), C-reactive protein (CRP), and D-dimer (DD) values in the first tercile and lymphocyte count in the third tercile.The high-risk category was defined as any LDH, CRP, or DD values in the third tercile or lymphocyte count in the first tercile.The moderate-risk category was defined as patients who did not meet the criteria of the low-or high-risk categories.

Statistical analysis
In a descriptive analysis, we compared epidemiological data, demographics, signs and symptoms on admission, comorbidities, laboratory results, chest x-ray findings, treatment received, and clinical outcomes.Continuous variables were expressed as mean and standard deviation or median and interquartile range (IQR), according to distribution assessed by the Shapiro-Wilk test and standardized normal probability plots.Categorical variables were expressed as frequencies and percentages.Differences between groups were compared using Student's t-test or the Mann-Whitney U test for continuous variables and the likelihood-ratio chi-square test for categorical variables.
A univariate analysis was performed to explore possible risk factors for all-cause death during admission or the next 30 days from discharge and variables associated with the exposure (pandemic wave) using binomial logistic regressions.The variables were chosen from an array of clinical and laboratory findings, previous comorbidities, and treatments received according to local protocols.Due to the large sample size, almost all variables showed significant differences in the comparisons between exposure and outcome groups in the univariate analysis.
We created a logistic regression model to assess the effect of being admitted during the first or second wave on all-cause mortality risk.We selected a series of predictors associated with the exposure (pandemic wave) and the outcome (mortality) as potential confounding factors.The selection criteria also took theoretical arguments or findings from other studies into consideration in order to adjust for factors that could explain a potential difference in the risk of death between the two waves.
The admitting variables that were ultimately included as possible confounders of the wave effect were age (categorized into decades from 70 years), sex (reference: male), age-adjusted Charlson Comorbidity Index (reference: moderate comorbidity), degree of dependence (reference: none or mild dependence), hypertension, arterial stiffness (pulse pressure ≥ 60 mmHg), clouding of consciousness, tachypnea, oxygen saturation/FiO2 ratio (%), blood glucose level categories, risk category based on the pattern of inflammation, and bilateral pneumonia as well as tocilizumab, remdesivir, or corticosteroid therapy during hospitalization.
In addition, first-order interactions between the waves and all potential confounding factors were included in the initial model.Multicollinearity was detected for several terms of interaction, and they were removed from the model.A chunk test for the rest of interaction terms did show statistical significance (p < 0.001), so individual likelihood ratio tests were performed for every one of them.Three interactions with the variable "Wave" remained statistically significant: severe dependency, corticosteroid treatment, and ventilatory support, which has three different categories.In order to achieve an interpretable estimation and reduce the number of combinations for which to calculate the wave effect on mortality, we decided to omit the interaction of the wave with ventilatory support.So, the final logistic regression model included all the confusion terms and the interactions of Wave with Severe dependence and Systemic corticosteroid therapy.We did not conduct variable selection once the model was estimated, as this maximal model is the best fit for calculating the wave's effect on mortality.Adjusted odds ratios and risk ratios were estimated for each combination of the values of interaction terms.Adjusted risk ratios were calculated with delta-method standard errors for the wave covariate.All analyses were conducted using Stata version 18.0 (StataCorp.2023.Stata Statistical Software: Release 18. College Station, TX: StataCorp LLC).

Figure 1 .
Figure 1.Case Fatality Rate (CFR) according to age in patients hospitalized during the first and second waves of COVID-19, expressed as percentage (deceased patients/total patients).The diamonds indicate the point estimate of the risk difference, and the horizontal bars represent its 95% confidence interval.

Figure 2 .
Figure 2. Several characteristics of patients, differences in COVID-19 severity and management could confound the estimation of mortality between waves.After adjusting for these factors, differences in the outcome could be explained by unmeasured patient characteristics and COVID-19 severity factors, or uncontrolled COVID-19 management.

Table 1 .
Clinical differences in patients ≥ 70 years admitted in the first and second waves of COVID-19.a Severe dependence: Barthel Index < 60; b Obesity: BMI ≥ 30 kg/m2; c Cardiovascular disease: Ischemic heart disease, heart failure, transient ischemic attack, stroke, or peripheral artery disease; d Obstructive respiratory disease: chronic obstructive pulmonary disease, asthma, chronic bronchitis, or obstructive sleep apnea; e

Table 2 .
Clinical differences in patients ≥ 70 years hospitalized for COVID-19 by survival status.a Severe dependence: Barthel Index < 60; b Obesity: BMI ≥ 30 kg/m 2 ; c Cardiovascular disease: Ischemic heart disease, heart failure, transient ischemic attack, stroke, or peripheral artery disease; d Obstructive respiratory disease: chronic obstructive pulmonary disease, asthma, chronic bronchitis, or obstructive sleep apnea; e Malignancy: solid tumor, leukemia, lymphoma; f CKD: chronic kidney disease (patients on dialysis or with serum creatinine > 3 mg/dL); g Arterial stiffness: pulse pressure ≥ 60 mmHg; h Tachypnea: > 20 breaths per minute; Fever: temperature > 37.8 ºC; j Tachycardia: heart rate > 100 bpm; k Inflammatory pattern: see description in "Methods".Categoric variables are expressed as percentages and compared using likelihood-ratio chi-square test for statistical significance.OR = odds ratio; 95% CI OR: 95% confidence interval for odds ratio.The odds ratios have been calculated with respect to the first category.Quantitative variables are expressed as mean (standard deviation) or median [interquartile range] and were compared for statistical significance using Student's t-test (with equal or unequal variances) or Mann-Whitney U test, as appropriate. i

Table 3 .
Logistic regression multivariate model showing the average effect of being admitted in the second wave after adjusting for confounding variables.a,cSeedescription in "Methods".bBarthelindex < 60. d Interaction between wave and severe dependence.eInteraction between wave ant systemic corticosteroid therapy.The odds ratio of Second wave represents the average effect of hospital admission during the second wave, adjusted for confounding terms and interaction.Table4displays the effect for each combination of interaction term values.

Table 4 .
Effect of second wave on mortality in patients ≥ 70 years hospitalized for COVID-19.AOR: adjusted odds ratio; ARR: adjusted risk ratio calculated with delta-method standard errors for the wave covariate.