Statins and Mortality in COPD: A Methodological Review of Observational Studies

Abstract

Randomized controlled trials and observational studies have reported conflicting results on the potential beneficial effects of statins on mortality in patients with chronic obstructive pulmonary disease (COPD). We performed a systematic search of the literature to review all observational studies reporting relative risks of death with statin use in COPD, focusing on potential sources of bias. We identified 15 observational studies, out of 2835, of which 12 were affected by time-related and other biases and the remaining 3 by confounding bias. All 15 studies were also subject to confounding bias due to lack of adjustment for important COPD-related factors. The risk of death associated with statin use was reduced across all 15 studies (pooled relative risk (PRR) 0.66; 95% CI: 0.59-0.74). The reduction was observed in 7 studies with immortal time bias (PRR 0.62; 95%: 0.53-0.72), two with collider-stratification bias (PRR 0.60; 95% CI: 0.45-0.80), one with time-window bias (RR 0.61; 95% CI: 0.38-0.98), one with immeasurable time bias (RR 0.50; 95% CI: 0.40-0.62), and one with exposure misclassification (RR 0.86; 95% CI: 0.72-1.03). The three studies that avoided these biases were, however, affected by confounding bias resulting in a PRR of 0.77 (95% CI: 0.61-0.98). In conclusion, the observational studies investigating statin use and mortality in COPD are affected by major biases, many of which can result in spurious protective effects. Well-designed observational studies that carefully emulate randomized trials are needed to resolve this uncertainty regarding the potential beneficial benefits of statins on mortality in patients with COPD.

Keywords:

• Confounding
• database studies
• immortal time bias
• selection bias

Introduction

Chronic obstructive pulmonary disease (COPD) is a common, preventable, and treatable lung disease characterized by chronic airflow limitation and impaired lung function [1]. Currently available treatments, that aim to reduce symptoms and slow disease progression, include inhaled bronchodilators and inhaled corticosteroids, alone or combined. Alternative treatments, particularly those with anti-inflammatory properties that address the pathophysiologic components of COPD, are intensively sought [2].

Given that cardiovascular disease is a major comorbidity in COPD, there has been a growing interest in the role of cardiovascular medications such as statins, angiotensin receptor blockers (ARB), and beta-blockers on the risk of COPD-related outcomes [3]. Statins, which are 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors commonly used for their lipid-lowering effects, have also been found to have anti-inflammatory and immunomodulating properties that may be useful in the treatment of lung diseases [4, 5]. Several non-randomized observational studies have reported beneficial effects of statins in patients with COPD, including important reductions in the risk of exacerbations and mortality [6, 7]. Only two randomized controlled trials (RCTs) have been conducted to assess this effect, the largest being the Prospective Randomized Placebo-Controlled Trial of Simvastatin in the Prevention of COPD Exacerbations (STATCOPE) study, which found null effects of statins on COPD exacerbations and mortality relative to placebo over three years [8]. On the other hand, a one-year trial found that simvastatin significantly delayed time to first exacerbation by 50% and reduced exacerbation frequency by 23%, but no effect was observed on mortality [9].

The heterogeneity between the results of the observational studies and RCTs may be due in part to biases in the study design and analysis of the observational studies, some of which were shown to be affected by methodological shortcomings that are likely to exaggerate drug benefits [3, 10, 11]. On the other hand, the heterogeneity between the two RCTs may be due to the types of patients enrolled in the two trials [8, 9]. Moreover, while both trials were designed with sufficient power to detect a reduction in the rate of moderate or severe exacerbation with statin use, they were not powered for the effect on mortality, the primary outcome in most observational studies. It is thus unclear if the different findings with respect to the outcome of mortality between the observational studies and RCTs are due to biases in the former or low power in the latter.

This paper therefore aims to conduct a methodological review of all observational studies that investigated the effect of statin use on mortality in people with COPD, with a specific focus on time-related biases commonly found in pharmacoepidemiologic research.

Methods

We searched MEDLINE, EMBASE, and CINAHL from inception to November 2022 for non-randomized observational studies investigating the effect of statin use on the risk of mortality in people with COPD. The search strategy was peer-reviewed by a health sciences librarian [12] and keywords such as “COPD”, “Pulmonary Disease, Chronic Obstructive”, “emphysema”, “Hydroxymethylglutaryl-CoA Reductase Inhibitors”, and “statin” were used. Duplicate studies were removed, then the title, abstract, and keywords of studies were screened to identify potentially relevant studies. Both authors performed full text assessment of potentially relevant publications and examined the bibliography of existing reviews and all relevant studies identified by the above methods. We only included observational studies that compared the effect of statin use to nonuse on the risk of mortality in patients with COPD.

All selected studies were included regardless of study quality. We extracted data regarding the study size, study design, country of study participants, patient characteristics, exposure and comparator definitions, list of potential confounders, crude or adjusted relative risks (risk ratios, odd ratios, or hazard ratios) and their corresponding 95% confidence intervals (CI). Each study was assessed for potential sources of biases, particularly time-related biases such as immortal-time bias, immeasurable-time bias, and time-window bias, independently by both authors.

Given the expected heterogeneity due to differences in study populations, definitions of statin exposure, follow-up length, and other factors, the study-specific results were pooled using the random effects model and stratified by the type of biases identified [13]. R version 4.1.0 (The R Foundation for Statistical Computing, 2021) was used to analyze the data and generate forest plots.

Results

We identified 2,835 records from the database search (Figure 1). Of these, 447 duplicates were excluded and an additional 2373 were deemed not relevant or inappropriate and excluded, resulting in a final analysis of 15 observational studies of the effect of statin use compared to nonuse on the risk of mortality in COPD patients (Supplemental Table S1) [14–28]. The pooled relative risk (PRR) of mortality with statin use across all 15 studies was 0.66 (95% CI: 0.59-0.74) (Figure 2).

Figure 1. Flowchart of the study selection process.

Figure 2. Forest plot of the reported relative risks of mortality associated with statin use in COPD patients from the 15 observational studies included in the review. The relative risks have been pooled according to the biases identified in the studies.

Of the 15 studies included, we found that 7 were affected by immortal time bias (PRR 0.62; 95%: 0.53-0.72), 2 by collider-stratification-bias (PRR 0.60; 95% CI: 0.45-0.80), 1 by immeasurable time bias (RR 0.50; 95% CI: 0.40-0.62), and 1 by time-window-bias (RR 0.61; 95% CI: 0.38-0.98). One of the studies used a study design that may have been affected by exposure misclassification [20], as previously described [29]. The three remaining studies, that addressed these biases, had a pooled relative risk ratio of 0.77 (95% CI: 0.61-0.98). Figure 2 displays these studies according to the form of bias.

Immortal time bias

Seven of the fifteen studies included were subject to immortal time bias with pooled relative risk ratio of 0.62 (95% CI 0.53-0.72) [15, 17, 19, 22–24, 28]. Immortal time bias can arise when exposure groups are defined based on prescriptions given during follow-up, creating an immortal time in follow-up during which the outcome of interest cannot occur [30].

An example of immortal time bias is the prospective cohort study by Citgez et al. which evaluated the association between statin use and the risk of mortality in patients with COPD [24]. The study included 795 patients enrolled in the Cohort of Mortality and Inflammation in COPD (COMIC) study from December 2005 until April 2010. Patients were followed from study entry until death or end of 3-year follow-up. Exposure to statin was determined through a search of patients’ pharmacy records, such that statin exposure could have also included new statin use after study entry. In this case, the time between study entry and first statin prescription is immortal because the patient had to remain alive until the first prescription to be grouped as exposed (Figure 3). If this “immortal time” is excluded, misclassified, or not properly accounted for in the data analysis, the statin group will receive an artificial survival advantage while the unexposed group will experience an inflated event rate. This will in turn lead to the treatment under study having an overestimated protective effect in preventing the outcome of interest [30, 31].

Figure 3. Illustration of immortal time bias introduced when patients are enrolled according to drug used at any time during the observation period i.e. the ever-use approach. The time from cohort entry to first statin prescription is immortal, as the patient must survive to redeem this prescription (grey line).

As a result of misclassifying the immortal time, after adjusting for potential confounders, statin use was associated with a 30% reduced hazard for all-cause mortality (hazard ratio [HR] 0.70; 95% CI: 0.51 − 0.96). Citgez et al. acknowledged the possibility of introducing immortal time bias with the exposure definition and performed a sensitivity analysis that excluded 62 patients who started statins after study entry. After the exclusion, statin use was no longer significantly associated with all-cause mortality (HR 0.82; 95% CI: 0.59 − 1.16). However, excluding the patients that initiated statins during follow-up did not eliminate the immortal time bias introduced into their analysis. If these 62 patients had died before initiating statins, by definition, they would have been classified as unexposed. Thus, the immortal time between study entry and statin initiation is excluded from the analysis which would also result in the inflation of the death rate in the unexposed group and an overestimation of the protective effect of the treatment [31].

Six other studies also introduced immortal time bias into their analysis by misclassifying [15, 19, 22, 23, 28], or excluding [17], the immortal time created with their study design. Immortal time bias can simply be eliminated with a time-dependent analysis or proper study design [30, 31].

Collider-stratification bias

Two of the fifteen studies, with pooled relative risk ratio of 0.60 (95% CI 0.45-0.80), were possibly affected by the collider stratification bias [16, 18]. Collider stratification bias, also known as collider bias, is a form of selection bias that is caused by conditioning on the common effects of the exposure and outcome [29, 32]. Conditioning on these otherwise unrelated variables may create associations that are not present in the study’s source population, leading to biased results [33]. It can also reverse the direction of the association, making a seemingly harmful exposure appear protective and creating a paradoxical relationship [34, 35].

An example of collider-stratification bias is the retrospective cohort study by Lawes et al. which investigated if statin use was associated with reduced mortality in patients with COPD in New Zealand [18]. The study enrolled 1,687 patients admitted to any public hospital in New Zealand with a discharge code indicating COPD in 2006. Patients who had redeemed a prescription for statins at least once in the six months prior to their hospitalization were classified as "statin users," while those who had not redeemed any statin prescriptions during that time were classified as "nonusers." The other study that possibly introduced collider-stratification bias is the retrospective cohort study by Mortensen et al. which enrolled 11,212 patients who had been hospitalized for COPD in 2000 [16]. Patients whose most recent prescription of statins would have lasted until their date of hospitalization were considered "current users," while all others were classified as nonusers.

By defining their cohorts based on hospitalization for COPD, these two studies introduced the potential collider-stratification bias. This means that hospitalization, an important predictor of mortality in COPD, acted as a potential collider in the relationship between statin use and mortality. Previous research has shown that hospitalizations due to COPD exacerbations can worsen the disease prognosis and increase the risk of mortality [36], while statin use is associated with COPD hospitalization, even indirectly [9, 37–39]. Therefore, the use of COPD hospitalization to define the cohort in both studies led to the creation of a collider and conditioning on it may have introduced an association between statin use and mortality that is not present in the source population.

The directed acyclic graph (DAG) presented in Figure 4 illustrates the collider-stratification bias resulting from the associations between statin use, COPD hospitalizations, and the risk of mortality. There are various unmeasured risk factors, such as genetics and lifestyle factors, that can contribute to both COPD hospitalization and death. When studies report that statin use leads to a decrease in COPD hospitalization and death, it indicates a potential link between statins and hospitalization for COPD, with COPD hospitalization becoming a collider. Therefore, conditioning on COPD hospitalization through restriction, stratification, or adjustment can introduce a non-causal association between statin use and unmeasured risk factors [33]. These non-causal associations create a new pathway between statin and mortality that can result in a spurious protective effect of statin use on mortality. As both studies reported a decrease of up to 49% in the risk of death with statin use, the bias may have led to the overestimation of the protective effect of statins. Adjusting for other potential confounders in the analysis are insufficient to bypass the bias introduced [29]. The best way to avoid the collider-stratification bias is to address it at the study design stage.

Figure 4. Illustration of collider-stratification bias resulting from associations between statin use, COPD hospitalizations, and the risk of mortality, with the cohort defined by the COPD hospitalisation.

Time-window bias

One of the included studies with a relative risk of 0.61 (95% CI: 0.38-0.98) may have introduced time-window bias into their analysis [21]. Time-window bias refers to the type of bias that occurs in case-control studies when controls are selected by a time-independent approach so that the duration of exposure opportunity is not taken into account [36, 40]. This selection approach can lead to differential exposure time opportunity between cases and controls and bias the results of the study (Figure 5) [41].

Figure 5. Illustration of time-window bias resulting from the time-independent approach of selecting controls in the nested case-control analysis.

The study with time-window bias used a nested-case control design within the Rotterdam Study cohort [21]. The 363 patients who died between April 1st, 1991 and January 1st, 2008 made up the case series and were matched by sex and age with 2345 controls “who were still alive on the same day of follow-up as their matched case”. Statin exposure was ascertained through a retrospective search of each patient’s pharmacy record and was defined as any exposure to statins from COPD diagnosis to the case’s mortality date. As a result of this study design, cases had longer observation period (mean 7 years) compared to the controls (mean 5 years). By matching on calendar date rather than length of observation period, the cases and controls had differential exposure time opportunity which biased the results of the analysis [36, 40, 41]. It is difficult to predict the direction of time-window bias as it is determined by the distribution of treatment observation periods between cases and controls but the bias can be avoided by ensuring equal time windows between exposure groups [40].

Immeasurable time bias

One of the studies in the review, with a relative risk of 0.50 (95% CI: 0.40-0.62), was at risk of immeasurable time bias [14]. Immeasurable time bias occurs when there is a period of time under study during which a subject cannot be recognized as exposed [10, 31, 41]. This bias is common to observational studies that use computerized databases and is caused by the inability to identify in-hospital medication use.

The nested case-control study by Mancini et al. investigated the effect of statins and other cardiovascular medications on hospitalization, myocardial infarction, and mortality in COPD patients. The index date for the cases was defined according to the outcome under investigation, and statin use was measured in the 60 days prior to the index date. The administrative database used for the study recorded all physician visits, procedures, hospitalizations, and outpatient prescription drugs [14]. However, like many other computerized databases it may not cover in-patient medication use. As a result, cases hospitalized just before death may have been considered unexposed due to the lack of information on in-hospital prescriptions, introducing immeasurable time bias. For chronic illnesses such as COPD, it is common for severe patients to have frequent and lengthy hospitalizations just before death [10]. Therefore, mortality cases such as those in this study will spend more time hospitalized and will have less time to be considered exposed when compared to the controls, leading to biased results. The study reported a 50% reduction in the risk of mortality with statin use. With immeasurable time bias, the risk of death will be underestimated for the cases, resulting in spurious or overestimated protective effects for statin use.

More detailed discussion on how this study may have introduced immeasurable time bias to their analysis and possible approaches to circumvent the bias have been previously described [10, 31].

Confounding bias

One of the primary difficulties in analyzing treatment effects using observational data is to account for confounding variables [42]. All fifteen studies included in the review were at risk of substantial confounding bias [14–28]. While some respiratory and cardiovascular factors were adjusted for in the studies, important confounding variables such as comorbidities and markers of COPD severity were not (Supplemental Table S2). These confounders are significant predictors of mortality in COPD, and failing to account for them can lead to biased results if the statin users differ significantly from nonusers.

The three studies that avoided the time-related and other biases described earlier in this review were, however, subject to significant confounding bias [25, 26, 27]. The Raymakers et al. study used confounders, such as comorbidities and prior hospitalizations, that were measured in the same 1-year window used to assess statin exposure [25]. By measuring statin exposure and the confounders over the same time period, it is unclear which came first. It is then challenging to establish whether the “confounders” are truly confounders (measured prior to exposure start) or intermediate factors influenced by the exposure (measured after exposure start). Establishing the temporal relationship between exposure and measured confounders is important to reduce the risk of bias [43].

The studies by Damkjaer et al. and Smith et al. avoided immortal time bias by defining statin use as a time-varying exposure, thus properly classifying the start of exposure [26, 27]. However, adjustment for some key confounders was based on covariates measured at study enrollment and not updated at the time of statin initiation or the similar time point for nonusers. Indeed, while the study by Smith updated covariates of diabetes, hypertension and cardiovascular disease, key COPD-related covariates such as exacerbations and hospitalizations were not. Given that statin use was treated as a time-varying exposure in the analysis, it was necessary to update these key confounding variables at the time of statin initiation as well to reduce the impact of confounding [44].

Discussion

Observational studies are now widely used as a source of real-world evidence for the effectiveness and safety of medications, including the identification of new indications for drugs beyond their approved and prescribed uses [45]. Our methodological assessment of all 15 observational studies reporting on statin use and mortality in COPD revealed that all fifteen studies included were affected by major biases. Indeed, immortal time bias, collider stratification bias, and immeasurable time bias are known to overestimate the protective effects of a drug, while the direction of time-window bias, exposure misclassification, and confounding bias is unpredictable.

The link between COPD and CVD may be attributed, to some extent, to common risk factors such as smoking, age, gender, and physical inactivity. However, it has also been suggested that systemic inflammatory changes related to COPDs may also contribute to this association [2]. Thus, the interest in the potential benefits of cardiovascular medications for COPD patients, such as statins which have been suggested to possess anti-inflammatory and immunomodulating properties, has been growing [3]. However, the exact mechanism by which statins may benefit individuals with COPD is not yet fully understood.

The two RCTs of statins in COPD found divergent results on the risk of COPD exacerbations relative to placebo, with one reporting no effect and the other a significant reduction [8, 9]. With respect to mortality, a meta-analysis of the RCTs found that statins were not associated with lower mortality relative to placebo (pooled odds ratio 1.03; 95% CI: 0.61 − 1.74) [46], as did the more recent trial [9]. However, all RCTs had relatively small sample sizes and were thus not adequately powered to detect a reduction in all-cause mortality [8, 9, 47]. Our methodological review revealed that the observational studies, most of which report lower mortality with statins in COPD, were affected by significant biases, which limited the ability to draw meaningful conclusions from their findings.

Future observational studies should employ study designs and data analyses that avoid time-related biases and other biases known to impact pharmacoepidemiologic research. For example, the prevalent new user design which emulates a RCT, while permitting comparisons with nonuse, will mitigate potential selection bias, immortal time bias, and time-related confounding [48]. Moreover, one should be cognizant of healthy user and healthy adherer biases, which have been suggested as possible explanations for the overestimated or spurious protective effects of preventative therapies, such as statins or aspirins, on the outcome of interest [49]. In these studies, a cause-specific mortality outcome should also be considered because using all-cause mortality could itself dilute the hazard ratios due to causes that are entirely unrelated to the disease.

In summary, several time-related biases, such as from immortal and immeasurable time, collider-stratification, exposure misclassification and confounding have affected the observational studies examining the effectiveness of statins on mortality in patients with COPD. These biases can easily be avoided with proper methods. Since observational studies are now accepted as evidence of the effects of medications in real-world clinical settings, it is thus crucial that they are designed and analyzed specifically to avoid these major biases, though some residual confounding can remain such as from healthy user and healthy adherer biases. Adequately designed observational studies that address these avoidable biases will need to be conducted to resolve the uncertainty regarding the potential beneficial role of statin use on mortality in patients with COPD.

Disclosure statement

SS attended scientific advisory committee meetings or received speaking fees from AstraZeneca, Atara, Boehringer-Ingelheim, Bristol-Myers-Squibb, Merck, Novartis, Panalgo, Pfizer and Seqirus.

This study is not funded. Pr. Suissa is the recipient of the Distinguished James McGill Professorship award.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.