Is the long-term mortality similar in COVID-19 and community-acquired pneumonia?

Introduction There are no data on the association of type of pneumonia and long-term mortality by the type of pneumonia (COVID-19 or community-acquired pneumonia [CAP]) on long-term mortality after an adjustment for potential confounding variables. We aimed to assess the type of pneumonia and risk factors for long-term mortality in patients who were hospitalized in conventional ward and later discharged. Methods Retrospective analysis of two prospective and multicentre cohorts of hospitalized patients with COVID-19 and CAP. The main outcome under study was 1-year mortality in hospitalized patients in conventional ward and later discharged. We adjusted a Bayesian logistic regression model to assess associations between the type of pneumonia and 1-year mortality controlling for confounders. Results The study included a total of 1,693 and 2,374 discharged patients in the COVID-19 and CAP cohorts, respectively. Of these, 1,525 (90.1%) and 2,249 (95%) patients underwent analysis. Until 1-year follow-up, 69 (4.5%) and 148 (6.6%) patients from the COVID-19 and CAP cohorts, respectively, died (p = 0.008). However, the Bayesian model showed a low probability of effect (PE) of finding relevant differences in long-term mortality between CAP and COVID-19 (odds ratio 1.127, 95% credibility interval 0.862–1.591; PE = 0.774). Conclusion COVID-19 and CAP have similar long-term mortality after adjusting for potential confounders.


Introduction
COVID-19 has shown a capacity to cause devastating outcomes.Fortunately, the emergence of effective vaccines and treatments have significantly mitigated this effect (1).However, there remains a concern about long-term mortality and other health consequences.Previous studies have demonstrated an increase in long-term mortality in patients with community-acquired pneumonia (CAP) compared to the general population.In those with CAP, approximately 9% of survivors with an acute episode who have required hospitalization die within a year (2,3).Contrarily, such mortality in patients COVID-19 has been reported as very low (4).To our knowledge, though, there are no data comparing the association of type of pneumonia (COVID-19 or CAP) with long-term mortality after an adjustment for potential confounding variables.Therefore, our aim was to assess the contribution of potentially confounding risk factors to the long-term odds of death in COVID-19 versus CAP.

Methods
We designed a retrospective analysis of two large, prospective and multicentre cohorts of hospitalized patients with COVID-19 and CAP.Patients were follow-up for a year after discharge.The COVID-19 cohort comprised patients from the RECOVID study and included patients admitted during 2020.Briefly, RECOVID is a registry of 49 Spanish hospitals with hospitalized patients with confirmed SARS-CoV-2 infection by reverse transcription polymerase chain reaction and with new infiltrates on chest X-ray or CT scan.The study received approval by the local ethics committees (UIC-IBU-2020-03).Furthermore, the CAP cohort included hospitalized patients from 15 Spanish hospitals between January 2012 and August 2015 (NEUMONAC study).For this cohort, patients met inclusion criteria if they received a diagnosis for pneumonia based on a new radiologic infiltrate and the presence of at least two compatible clinical symptoms.Exclusion criteria for both cohorts included admission within the previous 15 days and immunosuppression.We also excluded all those admitted to the intensive care unit (ICU; 13.5% of the total cohort) due to the high variability of ICU admission criteria during the pandemic.The study received approval by the ethics committee of each hospital, and patients signed an informed consent (2013/0204).
The main outcome under study was 1-year mortality in patients who were hospitalized in conventional ward and later discharged.We compared the association of type of pneumonia with long-term mortality after adjusting for variables selected by clinical decisions and per previous literature.Those variables were age, sex, nursing-home residency, smoking history, Charlson Comorbidity Index (CCI), bilateral radiologic involvement, CURB65 and vascular complications during hospitalization (3).Cardiovascular complications included acute coronary syndrome, heart failure, arrhythmia, cerebrovascular accident and venous thromboembolic disease (5).We summarized data as n (%) or median (interquartile range).Furthermore, controlling for all previously mentioned covariates, we adjusted a Bayesian logistic regression model to assess associations between the type of pneumonia type and 1-year mortality.The main reason for using Bayesian logistic regression was to accommodate the multiple imputation procedure in a natural and straightforward way.Using Bayesian inference provides more robust estimates of the model parameters and also captures the uncertainty of the different parameters in a more reliable way (6).A multiple imputation procedure was implemented with missing data to improve the robustness of our results.Briefly, we created 50 different imputed datasets using the Multivariate Imputation by Chained Equations (MICE) approach (7).Then, a Bayesian logistic regression model was fitted on each of the varying imputed datasets; all models were combined to get the pooled estimates.Flat prior distributions (non-informative) for each of the coefficients were used.We computed 95% credibility intervals for all the estimates and estimated the probability of effect (MPE) following the procedure of Makowski et al. (8).Credibility intervals have a probabilistic interpretation regarding the parameter's possible values (i.e., the area under the posterior distribution's curve between two points is the probability that the value of the parameters lies between these two points).Statistical analyzes were performed using R (version 4.2.2) and R packages clickR (version 0.9.27),MICE (version 3.15.0)and BRMS (version 2.18.0).

Discussion
Age, nursing-home residency, smoking history and comorbidity are recognized factors related to long-term mortality (3).We found that the odds of death for each 10-year interval in age or nursing-home residency.Secondly, as has been previously shown, the greater the CCI and CURB65 scores, the greater the mortality (9).Vascular events, especially in vulnerable patients, will leave sequelae that condition higher mortality.We showed that vascular complications are highly probable in conferring a negative effect on 1-year mortality.CAP severity is known to cause cardiovascular complications, which can worsen long-term prognosis (10,11).The COVID-19 cohort showed lower mortality, perhaps due to a younger age, than the CAP cohort.However, in the analysis, there were no relevant differences in mortality across both cohorts.This finding highlights long-term impact, not only in morbidity or long COVID.
Our study has several strengths and limitations.This is the first study to compare long-term mortality in individuals with COVID-19 or CAP Here we show that, similar to CAP, COVID-19 has considerable long-term mortality that has not been adequately addressed.Nevertheless, recruitment of these cohorts took place in different periods so we cannot rule out its possible effect on the results.Second, we used the CURB65 score, which has not been completely tested in COVID-19 (12,13).conclusion, COVID-19 and CAP have similar long-term mortality after we adjusted for potential confounders.Efforts to treat COVID-19 must also focus on the long-term consequences and not just on the acute phase.

TABLE 1
Characteristics for CAP and COVID-19 cohorts.Approximate probability distributions for estimates of different logistic regression model parameters for 1-year mortality.Odds ratio are shown in natural log scale.95 and 100% highest density intervals (HDI) for each estimate are colored in red and blue, respectively.