In India, all citizens are assigned a unique identification number using a biometric ID system at an individual level (Unique Identification Authority of India, 2009). For this study, we collected data from three multispecialty hospitals in India. The patients’ data was verified using the unique identity number. Then, the patients’ demographic details and other relevant clinical and laboratory details were obtained from the medical records department. A superior level of confidentiality and electronically secured data storage systems were utilized throughout the study.

This retrospective cohort study involved COVID-19 patients who were hospitalized and treated in three multispecialty hospitals in India from the time interval of 1 June 2020, to 30 April 2021.

The inclusion criteria for the study population were COVID-19 patients hospitalized, discharged, or died during the period mentioned above. The confirmation of COVID-19 was based on positive results obtained from the SARS-CoV-2 polymerase chain reaction (PCR) test of respiratory (nasopharynx, oropharynx) specimens and oxygenation saturation (SpO2) of ≤93% or PaO2/FiO2 < 300 mmHg. Imaging studies (Chest computed tomography) supporting PCR test results were included. These testing were all conducted in the hospitals mentioned above. Patients below the age of 18 and above 80 years, those without proper medical records documentation, patients admitted for only 1 day, and those suffering from tuberculosis, hepatitis B, HIV, and carcinomas requiring immunomodulatory therapies were excluded. The initial case-finding tally of 1834 patients was identified and reviewed based on the inclusion criteria. Of these patients, following this scrutiny, a total of 1,626 patients were identified for our study.

The patient data was identified using an electronic data search at the medical records department in three hospitals. No manual abstraction was performed due to COVID-19 restrictions. Patient demographics and clinical data were filled in using the most commonly occurring clinical signs and symptoms. The same selection format was followed to collect laboratory results. Treatment history and clinical outcome were followed until the end of the study period, patients’ death in-hospital, or until the patients’ discharge. The data was stored using alphanumerical code, and the patient’s name was de-identified.

Baseline information, including age, gender, body mass index (BMI), and co-morbidities, were recorded. Clinical co-morbidities included hypertension, diabetes mellitus, coronary artery disease, chronic lung disease, chronic kidney disease, stroke, and heart failure. We included features of clinical presentations during the patient’s hospitalization (i.e., presence of fever, dyspnea, cough, chest pain, fatigue, and O2 saturation). Several laboratory parameters at the time of hospital admission were also collected from the electronic medical records, including white blood count (WBC), D-dimer, C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), total cholesterol, low-density lipoprotein (LDL), high-density lipoprotein (HDL) and triglycerides. As statins act by lowering lipid levels, we collected lipid values from COVID-19 patients from inpatient and outpatient medical records and then averaged them for each patient throughout the study. Some patients failed to present the laboratory values collected during their clinical checkups; the data was validated by involving patients in whom these were available. The details for missing laboratory values are provided in Table 1. In this study population, 524 patients (32.2%) were on statin therapy, and the remaining 1,102 (67.7%) were not on statins. Among 768 patients, a 1:1 propensity-score matching (384 statin users and 384 non-statin users) was done. The comparison between statin and non-statin patients was further categorized. A detailed medical record search for individual patient demographics, clinical features, COVID-19 disease, duration of statin use, and underlying co-morbidities was elicited either from medical records or from the patients/family members and then documented. The data was reviewed by the treating physicians, pulmonologists, and radiologists to exclude any data inaccuracies.

None of the patients and the public were involved during the design, data analysis, and manuscript preparation.

The Cross-Hospital Ethics Review Board approved this study, and the waiver was granted for informed consent as patient data were de-identified.

Antecedent statin use in this cohort study was defined as “medical data record with statin being prescribed by the patient’s physician as a routine home medication due to their previous history of hypertension, coronary artery disease, and stroke.” The statin in these patients was appeased from the patient or the family members, dependents, and pharmacies at the time of hospitalization. The primary endpoint was defined as “mortality during a hospital stay,” and the secondary endpoint was defined as “invasive mechanical ventilation during a hospital stay.” Our study outcome also included the duration of hospital stay, number of days on a ventilator, and use of hemodialysis.

This study’s characteristic variables included the basic demographics, associated co-morbidities, antecedent statin use, common clinical and laboratory findings, and clinical outcomes. The statistical estimates were previewed as number (percentage) [n (%)] for nominal variables, medians, and inter-quantile ranges (IQR) for measuring variables. The differences between this study’s unmatched and matched cohorts were examined using the two-sided independent t-test and the Chi-square test.

We constructed Cox proportional-hazard regression models to estimate the association between statin, use, and primary endpoint. Before building the Cox model, the Schoenfeld residual analysis test detected the proportional hazard (PH) for all confounders. The PH means “the ratio of the hazards for any two variables is constant over time."

The univariable logistic regression was performed for each confounding factor. The variables that had a significant association with the mortality hazard were probed into the multivariate Cox regression. To adjust the bias resulting from the confounders, we implemented four strategies; multivariate Cox regression, weighted PH Cox, PH Cox after the propensity score matching, and the PH Cox adjusted with propensity score (Figure 1). In our study, individual matching was impossible due to the variations in the drug types used in COVID-19 therapy and other confounding variables, including the clinical and laboratory characteristics.

A confounder is a significant factor in causing variance and bias in an observational study. Due to the absence of randomization of patients in the treatment category, the cohort patients are not identical in their clinical, laboratory characteristics, and treatment strategy. To minimize the effects of the confounding factor, we implemented propensity score methods. The propensity score is a statistical analysis to estimate treatment’s pure effect and reduce or eliminate the bias caused by other confounding variables. All confounding variables’ effects were summarized into a scoring pattern, referred to as the “propensity score.” The individual propensities were estimated using the multivariate logistic regression model that included all the confounding covariates, mixed Cox model, Cox model with time-varying exposure, and marginal-structural model.

We formulated a multivariable logistic regression model to determine the confounding factors. The patients with statin index treatment were considered a dependent variable, and the baseline covariates were considered an independent variable. The propensity score matching was implemented using a nearest-neighbor strategy with the specification of caliper width equal to 0.1 of the standard deviation of the logit propensity score. All the baseline variables in the propensity-matched cohort were evaluated by implementing descriptive analysis. Laboratory results and patient clinical outcomes were stratified based on statin use in the cohort group of hospitalized COVID-19 patients.

We constructed a logistic regression on the propensity-matched cohort with the control group as a reference to address the primary and secondary endpoints. We analyzed and examined if the effect estimate remained consistent by utilizing logistic regression with multivariable adjustment on the total cohort. We modified the multivariable models for variables that have been previously studied concerning COVID-19 mortality, including the baseline covariates. Statistical package for social science (SPSS) software 2.0 was used for statistical analyses (Arbogast et al., 2013; Ranganathan et al., 2017).

Sensitivity analyses were performed by defining the recent statin users as “either antecedent statin or in-hospital statin use.” We analyzed the data by defining statin users as “in-hospital statin users.” By implementing these amended definitions, we estimated the relation of any recent statin users with the primary endpoint using multivariable logistic regression. We also utilized the subgroup analyses to evaluate the association of antecedent statin use with the primary endpoint in a subset of patients where statins are generally prescribed (e.g., history of hypertension, coronary artery disease, and stroke).

Data search revealed missing BMI and prior statin use in 19 and 15% of the registered patients. We utilized multiple imputations with predictive mean matching to adjust the missing data entry variables. We imputed 100 datasets, estimated the odds ratios on each imputed data variable, and averaged the one hundred estimated values to acquire the pooled estimates. Rubin’s rules were implemented to calculate the model estimates and the standard errors (Lu et al., 2008). The patients’ co-morbidities and lipid results were missing in 18 and 55% of the cohort, and these were presented only at baseline. The remaining variables were missing in less than 5% of the cohort study.

Out of 1,626 patients included in our study, 524 (32.2%) had a history of antecedent statin use before hospital admission. Based on the average age of the patients, statin users (median age of 63 years with IQR 55-79) were older in comparison to non-statin users (median age of 59 years with IQR 45-77) (p < 0.001). COVID-19 was marginally more in males (55.84%) and in statin users (52.4%) (p = 0.05) (Table 2).

Patients using statins had a significant higher prevalence of co-morbidities such as hypertension (75.0 vs. 58.0%), diabetes mellitus (62.5 vs. 40.3%), coronary artery disease (47.5 vs. 37.1%), and chronic lung disease (39.1 vs. 24.3%) in comparison to patients without statin intake (p < 0.001 for all). Furthermore, patients using statins were more likely to experience chronic kidney disease (37.2 vs. 15.9%), cerebrovascular accidents (11.0 vs. 2.9%) (p < 0.01 for both) (Table 2).

At the time of initial clinical presentation, patients on statins had a high incidence of dyspnea, cough, and fatigue (95.8, 93.3, and 92.7%, respectively) in comparison to non-statin patients (78.0, 81.7, and 76.4% respectively. There were no notable differences in the patients’ presentation with chest pain and low O2 saturation among statin and non-statin users (Table 2).

At the time of discharge the laboratory results revealed that the patients with statins showed a lower mean of WBC count (7.6 × 10³/µL vs. 8.1 × 10³/µL, p < 0.01), and C-reactive protein (100 mg/L vs. 120.7 mg/L, p < 0.001) compared to non-statin COVID-19 patients. The same positive results followed in lipid profiles for patients on statins (Table 3). Patients with antecedent statin use showed no significant differences for mechanical ventilation (20 vs. 24.2%, p 0.07), and hemodialysis (5.4 vs. 7%, p 0.41), and in length of hospital stay and days on a ventilator (Table 4).

Comparison and variation in clinical outcomes in the propensity-matched cohort between statin and non-statin use are presented in Table 4. The mortality in statin users was 17.1% compared to 31% in non-statin users.

The clinical outcome in the multivariable-adjusted overall cohort [odds ratio (OR) 0.55, 95% confidence interval (CI) 0.37-0.63] revealed that statin users had a significant effect on the primary endpoint (mortality during hospitalization). The odds ratio was defined as “a measure of association between the statin use/exposure and the clinical outcome.” The secondary endpoint was attained in 77 (20%) patients receiving statins compared to 93 (24.2%) non-statin users. Statin users also showed reduced odds in the multivariable-overall (OR 0.76, 95%CI 0.58-1.00), but this was statistically not significant (Table 5).

The hospitalized COVID-19 patients were followed over a 5-day interval for over 30 days (Figure 2). The incidence rate of mortality during the 30-day follow-up was 0.09 cases per 100-person-day in statin users (the mortality rate was 17.1%) in comparison to 0.11 cases per 100-person-day in non-statin users (the mortality rate was 31%) (Table 6). The matched statin users had more severe baseline clinical characteristics and higher proportions of pre-existing co-morbidities. We used the mixed-Cox model without involving the time-varying exposures in our propensity score matched (PSM) cohort; the statins significantly reduced the mortality incidence (aHR:0.25, 95% CI: 0.12-0.38, p = 0.001) (Table 6). The graph (Figure 2) shows the survival pattern among COVID-19 patients between statin and non-statin users. The “at-risk” COVID-19 cases were defined as the total number of patients minus deaths and the “survival” COVID-19 cases were defined as the total number of at-risk COVID-19 patients minus deaths.

In our study, physicians prescribed Rosuvastatin, the most preferred statin, to COVID-19 patients (Figure 3). It was observed that Rosuvastatin was most effective than the other statin group in reducing LDL, triglycerides, and total cholesterol and increasing HDL levels. As presented in Table 7, in COVID-19 patients with co-morbidities, Rosuvastatin was the most preferred statin to be prescribed by the physicians in COVID-19 patients with diabetes mellitus (51.8%). In comparison, atorvastatin was the most preferred statin in CAD and CKD patients (46.7 and 41.6%, respectively). The highest tolerable dose for Rosuvastatin was 40 mg, and for atorvastatin and simvastatin, it was 80mg, respectively. However, Rosuvastatin was prescribed at a 10 mg daily dose, whereas atorvastatin and simvastatin were prescribed at a 20 mg dose for 6 weeks.

Our study was conducted from the time interval of 1 June 2020, to 30 April 2021. Since Genesequencing was not available in these three multispecialty hospitals where the study was conducted, we could not incorporate the details of our study’s gene sequencing and variants results. Furthermore, the samples from COVID-19-positive patients had to be sent to regional gene sequencing laboratories, which would require ethical clearance from the National Regional Research Centre for the results obtained to be used in our study. Since intra-mural or extra-mural grants did not fund our study, we could not afford to bear the enormous financial burden for this aspect.

The detailed age-wise separation was not reflected in detail in our manuscript as the other age groups were not statistically significant to provide a positive impact through a graphical representation using plot analysis by the statistical analyzer software as used in our study (both joint-plot and Forest plot analysis were attempted for this purpose). Hence, we decided only to present the age range that positively impacted our study, as shown above.