Statin potency and the risk of hospitalization for community‐acquired pneumonia

Data sources

We used data from the UK's Clinical Practice Research Datalink (CPRD), a clinical database that contains the medical records of 673 general practitioner practices 11. It covers 11.3 million individuals and is a 6.9% representative sample of the entire UK population. The CPRD includes demographic information, lifestyle variables, diagnoses, clinical information, prescription information and laboratory data. CPRD data were linked to Hospital Episode Statistics data, which include primary and secondary in‐hospital diagnoses as well as in‐hospital procedure codes. The CPRD has been extensively validated 12. This study was approved by the Independent Scientific Advisory Committee of the CPRD (protocol number 15_236, which was made available to journal reviewers) and the Research Ethics Board of the Jewish General Hospital in Montreal, Canada.

Study population

We identified all new users of statins (other than cerivastatin, which was withdrawn from the market in 2001) aged ≥40 years between April 1998 and October 2011. A new user was defined as a patient with no previous prescription for statins or other lipid lowering medications in the previous 365 days. The cohort entry date was defined by the date of the first statin prescription. We excluded patients who: (i) received prescriptions for multiple statins on the day of cohort entry to avoid exposure misclassification; (ii) were hospitalized with community‐acquired pneumonia during the year before cohort entry to ensure that the patient did not develop pneumonia prior to the statin prescription; (iii) were hospitalized on the date of cohort entry and those hospitalized for any cause with a length of stay >1 day within 30 days before cohort entry to ensure that all observed cases of pneumonia were community‐acquired and not hospital‐acquired; (iv) had a diagnosis of cancer (other than nonmelanoma skin cancer) in the year before cohort entry due to the harmful effects of cancer and its treatments on the lungs; and (v) received a prescription for antituberculosis drugs in the year before cohort entry due to their association with respiratory disease. Patients were followed until an HCAP (defined below), death, departure from the CPRD or Hospital Episode Statistics, prescription of cerivastatin, multiple statin prescriptions on the same day, the end of the study period (31^st October 2011), the end of follow‐up (730 days) or any hospitalization with a length of stay >1 day, whichever occurred first. Patients were censored upon hospitalization with a length of stay >1 day as any pneumonia that occurred soon after hospitalization would probably be nosocomial or hospital‐acquired rather than community‐acquired pneumonia, the outcome of interest. Since such patients were not considered at risk of the event of interest, their follow‐up was censored at the time of hospitalization.

Case–control selection

Cases were defined as all patients with incident HCAP. To differentiate between community‐acquired and nosocomial pneumonia, only those with an ICD‐10 code (Table S1) 13 indicating the presence of pneumonia recorded on the day of admission or the day after admission were considered cases. Furthermore, only hospitalizations with a length of stay ≥1 day were considered as events, except for patients who died on their admission date. The index date for cases was defined by the day of follow‐up on which hospital admission occurred. For each case, up to 10 controls were randomly selected using risk set sampling, with controls matched on sex, age (±1 year), cohort entry date (±90 days), and follow‐up duration; for one case, the age caliper was widened to 2 years to identify an eligible control. The mean number of controls per case was 9.1 (standard deviation: 2.6), and the median was 10 (interquartile range, 10–10; Table S2).

In secondary analyses, we restricted our case series to fatal HCAPs, defined as cases of HCAP who died from any cause while in‐hospital, those who died in‐hospital from any cause within 30 days of admission, and all HCAPs with death within 30 days of admission, regardless of location.

Exposure definition

Statin prescriptions were classified according to their LDL‐lowering ability 14, with rosuvastatin ≥10 mg, atorvastatin ≥20 mg and simvastatin ≥40 mg classified as higher potency statins. Rosuvastatin <10 mg, atorvastatin <20 mg, simvastatin <40 mg, fluvastatin and pravastatin were classified as lower potency statins. This dichotomy is based on the ability to produce a 45% reduction in LDL cholesterol and has been used previously 3, 4.

Cases and controls then were classified into four mutually‐exclusive exposure categories: (i) current use of higher potency statin; (ii) current use of lower potency statin; (iii) recent use of any statin (regardless of potency); and (iv) past use of any statin. Current use was defined by a prescription whose duration +50% of that duration overlapped with the index date. Recent use was defined by a prescription whose duration +50% of that duration ended in the 1–90 days before the index date. Past use was defined by a prescription whose duration +50% of that duration ended >90 days before the index date. The exposure categories of recent and past use were included in our models to account for all person‐time of observation during follow‐up and to allow for the proper estimation of treatment effects of current use of higher potency statin compared with current use of lower potency statin. However, estimates for recent and past use are not presented as we were not interested in drawing inferences regarding these categories. In secondary analyses, duration of current exposure was subclassified as <120 days, 120–365 days, and 366–730 days.

Statistical analysis

Patient characteristics were summarized as counts with proportions for categorical data and means with standard deviations for continuous data, except in the case of skewed distributions, where medians and interquartile ranges were used. The overall rate of HCAP and corresponding 95% CI were estimated using Poisson regression. In our primary analysis, we used conditional logistic regression to estimate ORs and corresponding 95% CIs for HCAP with higher potency statins vs. lower potency statins; these ORs represent unbiased estimates of the hazard ratios (HRs) due to our use of risk set sampling 15. To minimize confounding, we constructed high dimensional propensity scores 16. We first created a propensity model with higher potency statin at cohort entry as the dependent variable; this model included obesity, alcohol abuse, use of more than four drug classes in the prior year, more than four general practitioner visits in the past year, diabetes, hypertension, prior myocardial infarction, prior stroke, prior coronary revascularization and 500 empirically identified covariates using the validated high dimensional propensity score macro (available at http://www.drugepi.org/dope‐downloads/). From each prespecified data dimension, the algorithm assesses the prevalence of each covariate, the recurrence of each covariate, and finally prioritizes each covariate based on its bias‐inflation factor to empirically identify confounders of the exposure‐outcome association. Diagnoses and medications were measured in the year before cohort entry, and lifestyle variables were assessed in the 5 years before cohort entry using the value closest to the date of cohort entry. The c‐statistic for our propensity score model was 0.78 (Figure S1).

We adjusted our conditional logistic regression models (the outcome model) for smoking (ever, never, missing), history of asthma, chronic obstructive pulmonary disease (COPD) or nonhospitalized pneumonia in the year before cohort entry, use of immunosuppressive agents, inhaled bronchodilators, inhaled corticosteroids, systemic antibiotics, systemic corticosteroids, pneumococcal vaccine, influenza vaccine, and propensity score decile. We also included an indicator variable in our outcome model for an index date during summer (defined as April to September) due to the seasonal variations in respiratory outcomes such as HCAP.

In secondary analyses, we repeated our primary analyses with our case series restricted to fatal HCAP. In addition, we examined the impact of duration of current exposure on the rate of HCAP. We conducted three sensitivity analyses. First, we used grace periods of 25% and 75% of the prescription duration. Second, we repeated our case series restricted to those with a primary diagnosis of pneumonia. Third, we repeated our primary analysis using traditional adjustment to estimate the amount of residual confounding removed by our use of high dimensional propensity scores. Traditional adjustment involved the use of a conditional logistic regression model that included exposure and the predefined confounders as covariates but did not include the propensity score. All analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC, USA).

Study population

The study cohort consisted of 348 428 patients (Figure 1), 217 721 of whom entered the cohort on a lower potency statin and 130 707 of whom entered the cohort on a higher potency statin. The study cohort resulted in 2251 incident cases of HCAP during a total of 561 886 person‐years of observation (crude rate: 4.0 HCAP per 1000 persons per year, 95% CI: 3.8–4.2). A total of 547 (0.2%) patients were censored due to the prescription of cerivastatin, 770 (0.2%) patients were censored due to multiple statin prescriptions on the same day, and 57 952 patients (16.6%) were censored due to hospitalization with a length of stay >1 day.

There were several notable differences in the baseline characteristics of the 2251 cases of HCAP and their 22 178 matched controls (Table 1). Cases were more likely to have a history of ever smoking and a higher prevalence of asthma, COPD, pneumonia, and cardiovascular‐related comorbidities. In addition, cases had a higher prevalence of medication use, including use of respiratory drugs, immunosuppressive agents, systemic antibiotics and systemic corticosteroids. Importantly, cases and controls had a similar prevalence of use of pneumococcal vaccine.

There were also some differences between the 10 439 current users of lower potency statins and 7356 users of higher potency statin among controls (Table S3). Current users of a higher potency statin were more likely to have asthma, COPD, pneumonia and cardiovascular‐related comorbidities. In contrast, lower potency statin users had a higher prevalence of diabetes mellitus. There were no notable differences in sex, age, body mass index, smoking, and use of pneumococcal vaccine, immunosuppressive agents and system antibiotics.

Statin potency and HCAP

Compared with lower potency statins, higher potency statins were associated with an increased rate of HCAP (adjusted HR: 1.14; 95% CI: 1.03–1.27; Table 2). No clear duration–response relationship was present (Table 2).

In secondary analyses, higher potency statins were associated with a higher risk of fatal HCAP compared with lower potency statins (adjusted HR: 1.29; 95% CI: 1.04–1.59; Table 3). Similar results were obtained when using alternate fatal HCAP definitions.

Sensitivity analyses

We conducted three sensitivity analyses. First, similar results were obtained when varying the grace period to 25% and 75% of prescription duration (Table S4). Second, we repeated our primary analysis with our case series restricted to those with a primary diagnosis of pneumonia, which produced results that were consistent with those of our primary analysis, although not significant (Table S5). Third, we repeated our analysis using traditional adjustment, which also produced results that were consistent with those of our primary analysis (Table S6).