key: cord-0692781-mdbfb3vh
authors: Razjouyan, Javad; Helmer, Drew A; Lynch, Kristine E; Hanania, Nicola A; Klotman, Paul E; Sharafkhaneh, Amir; Amos, Christopher I
title: Smoking Status and Factors associated with COVID-19 In-hospital Mortality among U.S. Veterans
date: 2021-10-25
journal: Nicotine Tob Res
DOI: 10.1093/ntr/ntab223
sha: 7013c735d83f9e42240876e94778496e2d42f688
doc_id: 692781
cord_uid: mdbfb3vh

INTRODUCTION: The role of smoking in risk of death among patients with COVID-19 remains unclear. We examined the association between in-hospital mortality from COVID-19 and smoking status and other factors in the United States Veterans Health Administration (VHA). METHODS: This is an observational, retrospective cohort study using the VHA COVID-19 shared data resources for February 1 to September 11, 2020. Veterans admitted to the hospital who tested positive for SARS-CoV-2 and hospitalized by VHA were grouped into Never (as reference, NS), Former (FS), and Current smokers (CS). The main outcome was in-hospital mortality. Control factors were the most important variables (among all available) determined through a cascade of machine learning. We reported adjusted odds ratios (aOR) and 95% confidence intervals (95%CI) from logistic regression models, imputing missing smoking status in our primary analysis. RESULTS: Out of 8,667,996 VHA enrollees, 505,143 were tested for SARS-CoV-2 (NS=191,143; FS=240,336; CS=117,706; Unknown=45,533). The aOR of in-hospital mortality was 1.16 (95%CI 1.01, 1.32) for FS vs. NS and 0.97 (95%CI 0.78, 1.22; P > 0.05) for CS vs. NS with imputed smoking status. Among other factors, famotidine and non-steroidal anti-inflammatory drugs (NSAID) use before hospitalization were associated with lower risk while diabetes with complications, kidney disease, obesity, and advanced age were associated with higher risk of in-hospital mortality. CONCLUSIONS: In patients admitted to the hospital with SARS-CoV-2 infection, our data demonstrate that FS are at higher risk of in-hospital mortality than NS. However, this pattern was not seen among CS highlighting the need for more granular analysis with high quality smoking status data to further clarify our understanding of smoking risk and COVID-19-related mortality. Presence of comorbidities and advanced age were also associated with increased risk of in-hospital mortality. IMPLICATIONS: Veterans who were former smokers were at higher risk of in-hospital mortality compared to never smokers. Current smokers and never smokers were at similar risk of in-hospital mortality. The use of famotidine and non-steroidal anti-inflammatory drugs (NSAIDs) before hospitalization were associated with lower risk while uncontrolled diabetes mellitus, advanced age, kidney disease, and obesity were associated with higher risk of in-hospital mortality.

M a n u s c r i p t 4 Implications  Veterans who were former smokers were at higher risk of in-hospital mortality compared to never smokers. Current smokers and never smokers were at similar risk of in-hospital mortality.  The use of famotidine and non-steroidal anti-inflammatory drugs (NSAIDs) before hospitalization were associated with lower risk while uncontrolled diabetes mellitus, advanced age, kidney disease, and obesity were associated with higher risk of in-hospital mortality.

M a n u s c r i p t 5

The impact of cigarette smoking on the risk of death or serious outcomes among patients with COVID-19 remains unclear 1 . Current and former smoking status are both associated with more frequent and severe upper and lower respiratory illnesses, and associated complications 2,3 . One might expect that smoking would have a similar negative impact on health outcomes in COVID- 19 infected patients, but the evidence is unclear and contradictory. For example, smoking is associated with lower BMI 4 , while high BMI is a risk factor for more severe illness from COVID-19 infection and is an established major contributor to COVID-19 mortality 5 . Furthermore, the effects of smoking on gene expression in current smokers is complex and incompletely understood 6, 7 , the fact that may also impact the course of COVID-19 infection. For example, the effects of nicotine on upregulation of the angiotensin converting enzyme (ACE-2) receptors on COVID-19 may influence the course of illness 8, 9 .

A systematic review of the observed effects of smoking on COVID-19-related outcomes from March 2020 identified five studies from China with small sample sizes (N ranged from 41-1,099) 10 . The combined analysis showed no differences between current smokers and other groups while the constituent study with the largest sample size (N = 1,099) estimated higher relative risk (RR) of severe symptoms, ICU admission, and mortality among the current smoker group compared to others 10 . A subsequent meta-analysis of 12 different studies (11 from China and one from Netherlands) demonstrated that smoking was associated with a 1.54-fold higher risk of a severe outcome among smokers vs. non-smoking patients infected with COVID-19. A more recent systematic review which included 47 studies (including studies previously reviewed) showed that current smokers had an increased risk of severe COVID-19, such as respiratory failure necessitating mechanical ventilation compared to non-smokers 11 . In contrast, a letter to editor after reviewing 11 articles reported that prevalence of hospitalization among smokers is low 6 due to possible interaction between SARS-CoV-2 and the nicotinic cholinergic receptor 7, 12 . A recent study showed a significant association between former smoker status and hospitalization, although the statistical difference faded after adjusting the model with confounding factors 13 . Limitations of the above studies include lack of adjustment for confounding factors, imprecise and incomplete documentation of smoking status, and small sample sizes.

There is an urgency to understand better the associations between smoking status and adverse outcomes due to COVID-19 infection to guide public health messaging and inform individual decisions about COVID risk mitigation measures. While the extant literature supports an association between current smoking and increased risk of complications of COVID-19, several questions remain, such as the relationship between former smoking status and COVID-19 outcomes. To examine such an association, we utilized the U.S. Department of Veterans Affairs (VHA) COVID-19 Shared Data Resource. VHA is a national integrated healthcare provider that operates more than 170 medical centers and more than 1,000 outpatient clinics across the U.S. It provides medical care and related social services for more than nine million enrolled Veterans with robust racial and ethnic diversity generally representative of the military population 14 The Shared Data Resource provides a large, curated data source from VHA national electronic medical records (EMR) for answering important COVID-19 related questions..

The main objective of the study report was to test the association between smoking status and inhospital mortality. Our approach also allowed us to identify demographic, clinical, pre-existing condition, and pre-hospitalization medication factors associated with in-hospital mortality of hospitalized patients with COVID-19.

M a n u s c r i p t 6 2. Methods

We performed an observational, retrospective cohort study using the VHA's Corporate Data Warehouse (CDW), which is a relational database that aggregates patient data from all VHA facilities from 1999 to present 15 . We used the COVID-19 Shared Data Resource that contains information related to COVID-19 treated within VHA 16 . It encompasses a wide range of information of SARS-CoV-2 tested patients (e.g., timing and nature of test results, pharmacological and non-pharmacological interventions, patient outcomes, and pre-existing conditions and medication) 17 .

For the present study, we included all veterans who were active users of VHA services and excluded patients who did not have at least one outpatient visit or one inpatient stay during the period from 

Our key variable was smoking status (never, former, current, and unknown) gathered from VHA Electronic Medical Records (EMR) Health Factors (HF) dataset. HF table contains data that are derived from clinical reminders within the EMR and include information on a variety of clinically relevant risk behaviors, such as smoking status. Smoking status in the COVID-19 Shared Data Resource included data from the 2 years prior to index date. Health Factor data were mapped to distinct smoking categories (current, former, never) based on the most recent health factor data if the most recent indicated current or former smoker. If the most recent was never, prior data was used to confirm and resolve inconsistencies. For example, a most recent never smoker who had a prior former smoker health factor was defined as former smoker. A most recent never smoker with a past current smoker was defined as former. Patients with no smoking health factors were defined as unknown. Previous studies showed agreement between EMR records and self-reported smoking from a survey with reported kappa statistics ranging from 0.66-0.74 [18] [19] [20] Our primary outcome for the statistical modeling was in-hospital mortality. Mortality was assessed from the patient treatment file reflecting death prior to hospital discharge 21 .

Imputation of Smoking Status: To address the uncertainty related to unknown smoking status, we imputed current, former, or never smoking status from other variables in the dataset using Multiple Imputations by Chained Equations (MICE) 22, 23 . MICE is an empirically and theoretically-supported tool to address missing data 24 . To increase the generalizability of the imputation model, we included all available variables. We performed five iterations with five levels of imputation in each iteration.

Potential confounders of COVID-19 in-hospital mortality were abstracted from the VA COVID-19 Shared Data Resources 25 . These included demographics, pre-existing conditions, and prehospitalization medications. The Elixhauser comorbidity index and Charlson Comorbidity index were determined using International Classification of Diseases (ICD)-10 codes from patients' encounters A c c e p t e d M a n u s c r i p t 7 (inpatient and outpatient) during the 2 years prior to index date 26, 27 . The prescribed medications were also reported for the two years prior to index date. To select the most important predictors, we used machine learning feature (or variable) selection techniques ( Figure 2 ) with four steps. The dependent variable for the most important variable selection process was in-hospital mortality. Age and body mass index (BMI) were converted to categorical variables; age was mapped to seven variables (<30, ≥30 <40, ≥40 -<50, ≥50 <65, ≥65 <75, ≥75 <85, and ≥85 years) and BMI was mapped to five variables (<18.5, ≥18.5 <25, ≥25 <30, ≥30 <40, and ≥ 40 kg/m2).

We utilized a cascade of variable selection approach to identify the most salient features. These steps were performed twice: a) without imputing 'unknown smoking' and b) with imputing 'unknown smoking' status into three other groups, i.e., 'Never smoker', 'Former smoker', and 'Current smoker'.

Step 1, prevalence: any variables with prevalence less than 1%, were removed.

Step 2, univariate filter: we excluded variables that were not statistically associated with in-hospital mortality at p<0.05.

Step 3, least absolute shrinkage and selection operator (LASSO): we used LASSO with 10-fold cross validation to select the most important candidate variables. The LASSO is a regression analysis method that performs both variable selection (filter method) and regularization (wrapper method) to enhance prediction accuracy and interpretability of the statistical model 28 .

Step 4, sequential forward variable selection: To avoid inter-variable correlations, we used sequential forward variable selection. Before we applied the sequential method, we enriched the list of variables from step 3 by forcing BMI and age variables into the model.

To understand the tendency of the most important variables to raise or lower risk of in-hospital mortality, we applied a binary logistic regression. The positive coefficients from binary logistic regression indicate association with increased risk and negative coefficients indicate association with decreased risk of in-hospital mortality. Odds ratios and confidence intervals were obtained by exponentiating the beta coefficients and confidence intervals around these coefficients.

To characterize the cohort, continuous variables are presented as mean with standard deviation, and categorical variables as number and percentage. We used logistic regression models to test the association between smoking status (current, former, and never) and in-hospital mortality on imputed and unimputed datasets. Odds ratios (OR) with 95% confidence intervals (95% CI) were reported. Each model was adjusted using the most important variables from the variable selection method described above. All statistical analyses were performed using SPSS, version 26 (SPSS Inc, Chicago, Illinois), the machine learning algorithms were performed using MATLAB (MathWorks Inc., Natick, Massachusetts, United States), and MICE imputation was performed using R (MICE-package 3.12.0). Figure 1 ]. The mean (standard deviation (SD)) age of former smokers (70.6 (12.2) years) was higher than never smokers (65.6 (14.8) years) and current smokers (64.3 (12.8) years). The mean (SD) of the Charlson Comorbidity Index in former smokers (3.5 (2.8)) was higher than never smokers (2.8 (2.6)) and current smokers (3.4(2.9)). We observed 1,299 deaths with the lowest M a n u s c r i p t 8 proportion observed in current smokers (9.0% (n=97)) compared to never smokers (11.8% (n=362)) and former smokers (14.8% (n=665)) [ 

Out of 152 variables (demographics, pre-existing conditions, and pre-hospitalization medications), 14 variables remained after the variable selection process (Figure 2 

The imputed, unadjusted OR of in-hospital mortality was 1. 36 Table 3 ). Identical analyses of the unimputed data revealed essentially the same strength of associations and confidence intervals for the associations between in-hospital mortality and smoking status (Supplement 3 & Supplement 4). Additionally, the association between smoking status and in-hospital mortality remained consistent after adjusting with each group of most important variables (demographics, pre-hospitalization conditions, and pre-hospitalization medications) (Supplement 5).

In this analysis of a large, national, retrospective cohort of veterans cared for by the VHA health care system, we found that odds of in-hospital mortality among Former Smoker veterans infected with SARS-CoV-2 is higher than Never Smokers and Current Smokers.

The role of smoking on mortality due to COVID-19 infection has been unclear and controversial. A series of systematic reviews reported an association between smoking status (Former Smoker and Current Smoker) and adverse COVID-19 outcomes, but these reports were limited by small sample sizes, homogeneous populations, incomplete capture of smoking status, and other flaws 10, 11, 29 . Our results perhaps indicate a limited or more nuanced role for smoking in risk of adverse outcomes. In our analysis, the adjusted odds ratio (aOR) of in-hospital mortality among former compared to never smokers was small, albeit statistically significant, indicating an elevated risk of mortality for former smokers. Also, we found a lower, but not statistically significant aOR of mortality among current smokers compared to never smokers. These findings were consistent in unadjusted and unimputed analyses, as well. Our findings add to the literature by highlighting the different risk of in-hospital mortality from COVID-19 among Former, Current and Never Smokers using a large sample with wellcurated data and more complete information on smoking status.

As a possible explanation of our findings, we hypothesize that the ongoing exposure to oxidative stress and the resultant increased mucus secretion and neutrophil accumulation in the airways among active smokers may potentially protect them from severe COVID-19 outcomes. The Angiotensin Converting Enzyme-2 (ACE-2) receptor is the main receptor for entry of SARS-CoV-2 in the host cells 30 . We and others have previously demonstrated that smokers (Former and Current Smokers) have higher expression of the ACE-2 receptor 31 which is particularly upregulated in the goblet cells 31 . Active smoking results in an increase the number of goblet cells and decrease ciliated cells resulting in the replacement of of normal epithelium and mucous metaplasia 32 . It is, therefore, possible that the decreased in ciliary cells and increased mucus secretion by goblet cells in active smokers have a protective effect in current smokers from complications of COVID-19 infection 1 . Future studies should explore our hypothesis further.

To our knowledge, this is the first study to demonstrate that former smokers are at higher risk of inhospital mortality from COVID-19. If corroborated in future studies, public health messaging about the increased risk of smoking with regard to death and complications from COVID-19 may need to be more nuanced. The CDC currently highlights "current or former cigarette smoker" status as a single risk factor 33 . Further studies in organoid cultures might help refine the mechanisms by which A c c e p t e d M a n u s c r i p t 10 tobacco exposure influences COVID-19 infection and why former smokers are at higher risk for death from COVID-19 compared to never or current smokers.

Among individual factors that increase risk for in hospital mortality from COVID, we corroborate some previously reported results, but the sample sizes and quality of data available through the VHA provide clearer and more precise information about the effects than has been available. Of note, after adjusting for confounders such as BMI, we did not find that ethnicity was a significant predictor of in-hospital mortality (Supplement 2) consistent with previous reports 34 . Among the most important predictors of in-hospital mortality, older adults were at disproportionate risk of mortality, likely reflecting disease severity, comorbidity burden and frailty not captured in the structured EMR data. This finding is consistent with Centers for Disease Control and Prevention (CDC) reports 35, 36 . CDC reported that mortality increased with age and eight out of ten COVID-19 deaths reported were in adults aged ≥ 65 years [37] [38] [39] . We observed that patients with obesity (BMI ≥ 40 kg/m 2 ) are at higher risk of mortality from COVID-19 infection which is also consistent with previous publications 40, 41 . Our study also showed that diabetes with complications, typically reflecting more severe, poorly controlled, or long-standing diabetes, are at greater risk of mortality, which is consistent with previous studies 42, 43 . The immune dysfunction associated with long standing diabetes may contribute to the elevated risk in this group 43 . The fact that this diabetes variable, and not other diabetes variables included in the variable selection process, remained important likely reflects ICD documentation practices more than a clinically relevant distinction among these variables. The association between Emphysema and higher in-hospital mortality is also consistent with previous reports 44, 45 . Once again, ICD coding practices likely explain why emphysema made the final list of important variables while Chronic Obstructive Pulmonary Disease did not and why Lower Respiratory Infection was marginally protective. The observed association between in-hospital mortality and two kidney-related pre-existing conditions (i.e., Acute Kidney Failure and Chronic Kidney Disease) was also reported previously 46 . It is speculated that relative immune dysregulation and exaggerated inflammatory responses in kidney disease contributes to worse COVID outcomes 46 .

Although hypertension proposed by CDC as a risk factor for severe COIVD-19 illness 47 , in our study it was excluded from the most important list of variables in the machine learning (LASSO) step. A systematic review of factors associated with COVID-19 mortality; hypertension also expressed its importance as a risk factor as 46% of patient who died from COVID-19 had such a pre-condition. In our veteran population, 74% (Supplement table 2 ) who died from COVID-19 had hypertension, hence, the LASSO step excluded this comorbidity due to low variation among hospitalized COVID-19 patients with in-hospital mortality.

Of interest, we did find that pre-hospital use of famotidine and NSAIDs was significantly protective after allowing for confounders. A recent retrospective study found that continuation of famotidine use in patients hospitalized with COVID-19 was associated with reduced risk of clinical deterioration leading to intubation or death 48 . Similar benefit was observed in another study of patients who received famotidine (either oral or intravenous at any dose) within +/− 7 days of COVID-19 positive test results and/or hospital admission 49 . Thus, famotidine has previously been reported as protective of severe COVID-19 outcomes but detailed analysis of its effects allowing for confounders in a large, national sample have been lacking 48, 49 . An early, rapid systematic review reported in March 2020 on NSAIDs and viral respiratory infections showed no evidence of increased severe adverse events as a result of the use of NSAIDs 50 . An early concerning announcement in LANCET about potential harm from NSAID use in patients with COVID was quickly contradicted in a short report which demonstrated no evidence of an increased risk of death with the use of NSAIDs in COVID-19 51 . Our study is the first study to suggest a protective role of NSAID use before hospitalization in hospitalized COVID-19 patients. The protective effects of NSAIDs may reflect their effects on coagulation or the inflammatory response or both. Further investigation is needed and warranted.

M a n u s c r i p t 11

The strengths of the study include a large cohort of patients from a healthcare system of national scope with a distribution of race and ethnicity approximating the U.S. population. Furthermore, we used a well-curated dataset with near complete demographics, pre-hospitalization medication prescriptions, pre-existing conditions, and up-to-date information of direct clinical and operational relevance to accurately assess testing, results, and outcomes. We applied a standard cross-sectional analytic approach to examine the association between smoking status and in-hospital mortality. Our approach to identifying the most important variables from the dataset was state of the art and our imputation of unknown smoking status enhances the internal validity and clinical relevance of our findings.

Importantly, our analysis is among the first and largest to segregate former from current smokers, revealing a potentially important difference in risk of adverse COVID outcomes between these two groups. The mortality assessment was limited to hospitalized patients due to delays in reporting outof-hospital deaths. Our data did not capture all possible confounding factors, including the social determinants of health that may confound the differences seen among different smoking status groups. Nicotine replacement therapy, behavioral tobacco cessation, vaping status, and details of smoking history (e.g., duration, years since quitting, total pack years) were not captured.

We focused our current study on examining the effect of smoking status and in-hospital mortality although the relationship between smoking status and the probability testing positive for COVID-19 and need for hospitalization is also very interesting. At this time, VHA data do not capture medical information for veterans who were tested or cared for outside of the VHA system unless the care is ordered and paid for by VHA. The data also captures only limited information about social determinants of health. The risk of incomplete information on the non-VHA outcomes and confounding factors limit the use of the VA COVID-19 Shared Data Resource for some analyses.

Our analyses of this large, national cohort of VHA users showed that former smokers are at increased risk of in-hospital mortality due to COVID-19, but current smokers and never smokers have a similar risk. Older adults, obese patients, kidney disease, and patients with diabetes mellitus with complications are at higher risk of mortality due to COVID-19 while use of famotidine and NSAID medications before hospital admission was associated with lower in-hospital mortality. These findings contribute to our understanding of how smoking affects COVID-19 disease course.

M a n u s c r i p t 12 Figure 1 : CONSORT Diagram M a n u s c r i p t 17 Table 3 : M a n u s c r i p t 18 

Cigarette Smoking and COVID-19: A Complex Interaction

The effects of waterpipe tobacco smoking on health outcomes: a systematic review

Smoking-related interstitial lung diseases: a concise review

Smoking and weight loss among smokers with overweight and obesity in Look AHEAD

Obesity and Mortality Among Patients Diagnosed With COVID-19: Results From an Integrated Health Care Organization

COVID-19 and the nicotinic cholinergic system

A potential interaction between the SARS-CoV-2 spike protein and nicotinic acetylcholine receptors

COVID-19 and nicotine as a mediator of ACE-2

Response to the emerging novel coronavirus outbreak

COVID-19 and smoking: A systematic review of the evidence

The effect of smoking on COVID-19 severity: A systematic review and meta-analysis

Nicotinic cholinergic system and COVID-19: In silico identification of interactions between α7 nicotinic acetylcholine receptor and the cryptic epitopes of SARS-Co-V and SARS-CoV-2 Spike glycoproteins

Smoking and risk of COVID-19 hospitalization

Overview of VA research on Health Equity

A 20-Year Evaluation of LOINC in the United States' Largest Integrated Health System

ORD COVID-19 Research Update for VHA Partners

Coronavirus disease 2019 in veterans receiving care at veterans health administration facilities

Validation of Veterans Affairs Electronic Medical Record Smoking Data Among Iraq-and Afghanistan-Era Veterans

Validity of Veterans Health Administration structured data to determine accurate smoking status

Validating smoking data from the Veteran's Affairs Health Factors dataset, an electronic data source

Ascertaining Veterans' Vital Status: VA data sources for mortality ascertainment and cause of death

A Frailty Index for UK Biobank Participants

Multiple imputation using chained equations: Issues and guidance for practice

Multiple Imputation by Chained Equations in Praxis: Guidelines and Review

Introduction to the VA COVID-19 Shared Data Resource and its Use for Research. Health Services Research & Development Web site

Validation of a combined comorbidity index

SNOMED CT Disease Hierarchies and the Charlson Comorbidity Index (CCI): An analysis of OHDSI methods for determining CCI

Feature selection for classification: A review

Features of severe COVID-19: A systematic review and meta-analysis

A review on the cleavage priming of the spike protein on coronavirus by angiotensin-converting enzyme-2 and furin

Tobacco Smoking Increases the Lung Gene Expression of ACE2, the Receptor of SARS-CoV-2

At the Root: Defining and Halting Progression of Early Chronic Obstructive Pulmonary Disease

People with Certain Medical Conditions

Differences in COVID-19-Related Testing and Healthcare Utilization by Race and Ethnicity in the Veterans Health Administration

Assessing the age specificity of infection fatality rates for COVID-19: systematic review, meta-analysis, and public policy implications

Severe Outcomes Among Patients with Coronavirus Disease 2019 (COVID-19)-United States

Estimating Older Adult Mortality From COVID-19

COVID-19 and Older Adults: What We Know

Obesity and mortality of COVID-19. Metaanalysis

Obesity is associated with increased risk for mortality among hospitalized patients with COVID-19

Impact of glycemic control in diabetes mellitus on management of COVID-19 infection

Are people with uncontrolled diabetes mellitus at high risk of reinfections with COVID-19? Prim Care Diabetes

COVID-19 and COPD: a narrative review of the basic science and clinical outcomes

COVID-19 and COPD

Outcomes for Patients With COVID-19 and Acute Kidney Injury: A Systematic Review and Meta-Analysis

People with Certain Medical Conditions

Famotidine Use Is Associated With Improved Clinical Outcomes in Hospitalized COVID-19 Patients: A Propensity Score Matched Retrospective Cohort Study

The use of non-steroidal anti-inflammatory drugs (NSAIDs) in patients with COVID-19: scientific brief

Is the risk of ibuprofen or other non-steroidal antiinflammatory drugs increased in COVID-19

The analysis was supported by seed funding from Baylor College of Medicine, Houston, Texas, United States, the Center for Innovations in Quality, Effectiveness and Safety (CIN 13-413), Michael E. DeBakey VA Medical Center, Houston, TX, United states and a national institute of health (NIH), National Heart, Lung, and Blood Institute (BHLBI) K25 funding (#:1K25HL152006-01). We are grateful to the VA Informatics and Computing Infrastructure (VINCI) and VA COVID-19 Shared Data Resource.

The authors do not have any competing interest.

The data are owned by the Veterans Health Administration (VHA) and can be maintained and analyzed only behind the VHA firewall. At this time, the data cannot be made available to non-VHA approved persons. For more details, please contact the corresponding author.

A c c e p t e d M a n u s c r i p t 19