Abstract
Objective To examine the association between statin exposure in a dose-dependent manner and intracerebral hemorrhage (ICH) in a large nationwide study.
Methods The computerized database of the largest health care provider in Israel was used to identify diagnosed ICH among new users of statins, who started statin treatment between 2005 and 2010. We assessed a dose–response relationship between ICH and statins, using the average atorvastatin equivalent daily dose (AAEDD). Multivariable Cox proportional hazard regression models, adjusted for baseline disease risk score, were applied to estimate the hazard ratio of ICH.
Results Of the 345,531 included patients, 1,304 were diagnosed with ICH during a median follow-up of 9.5 years (interquartile range 7.6–11.0). Overall, 75.3% of patients had AAEDD <10 mg/d, 19.0% had AAEDD 0–19.9 mg/d, and 5.7% had AAEDD ≥20 mg/d. The corresponding proportions were 81.0%, 15.0%, 4.0% among ICH cases, and 75.3%, 19.0%, 5.7% among non-ICH cases. Compared to those with AAEDD <10 mg/d (reference), the adjusted hazard ratio (HR) for ICH was 0.68 (95% confidence interval [CI] 0.58–0.79) in those with AAEDD 10–19.9 mg/d, and 0.62 (0.47–0.81) in those with AAEDD ≥20 mg/d. Compared to the lowest baseline total cholesterol quartile, the adjusted HR for ICH was 0.71 (95% CI 0.62–0.82), 0.55 (0.47–0.64), and 0.57 (0.49–0.67) in those in the second, third, and highest quartiles, respectively. The results were similar and robust among highly persistent statin users and after controlling for the change in cholesterol level.
Conclusions This study confirms that the risk of ICH decreases with increasing cholesterol levels, but suggests that statin use might be associated with decreased risk of ICH.
Glossary
- AAEDD=
- average atorvastatin equivalent daily dose;
- CHS=
- Clalit Health Services;
- CI=
- confidence interval;
- DRS=
- disease risk score;
- HMO=
- health maintenance organizations;
- HPS=
- Heart Protection Study;
- HR=
- hazard ratio;
- ICD-9=
- International Classification of Diseases–9;
- ICH=
- intracerebral hemorrhage;
- LDL=
- low-density lipoprotein;
- MI=
- multiple imputation;
- PDC=
- proportion of follow-up days covered;
- SPARCL=
- Stroke Prevention by Aggressive Reduction in Cholesterol Levels
Patients with cardiovascular risk factors have elevated risk for both ischemic stroke and intracerebral hemorrhage (ICH), a severe type of stroke resulting in high rates of mortality and disability.1 HMG-CO reductase inhibitors (statins) are lipid-lowering drugs widely prescribed for primary and secondary prevention of vascular events.2 Concerns have been raised, however, for potential increased risk of ICH among patients treated with statins. These concerns mainly derived from the Stroke Prevention by Aggressive Reduction in Cholesterol Levels (SPARCL) trial,3 demonstrating more ICH in patients randomized for statin therapy. Similar effect, among a subgroup of patients with a history of cerebrovascular disease, was found in the Heart Protection Study (HPS), a large multicenter randomized trial.4 In addition, some epidemiologic and case report studies found that patients with lower lipid levels are at increased risk for ICH.5,–,8 As low lipid levels weaken endothelial cells, it has been hypothesized that fragile endothelium promote vessels rapture and ICH in patients with low cholesterol. This mechanism, however, remains poorly understood.7
We set up to examine the association between statin exposure in a dose-dependent manner and ICH in a large nationwide study.
Methods
Source of data
This study is based on data from the computerized database of Clalit Health Services (CHS), which provides inclusive health care for more than half of the Israeli population. Health care coverage in Israel is mandatory and is provided by 4 groups akin to not-for-profit health maintenance organizations (HMO). All members of the different HMOs have a similar health insurance plan and similar access to health services, including low medication copayment. The electronic medical record database of CHS includes data from multiple sources: records of primary care physicians, community specialty clinics, hospitalizations, laboratories, and pharmacies.9,10 A registry of more than 100 diagnoses of chronic diseases is compiled from these data sources. Diagnoses are captured in the registry by diagnosis-specific algorithms, employing ICD-9 code reading, text reading, laboratory test results, and disease-specific drug usage. A record is kept of the data sources and dates used to establish the diagnosis, with the earliest recorded date, from any source, considered to be the defining date of diagnosis.
Study design
We performed a retrospective cohort study using prospectively and routinely collected data in the setting of a large population-based HMO in Israel.
Selection of study population
Using the CHS computerized database, we identified all adult (age ≥50 years) new users of statins, who started statins between January 1, 2005, and December 31, 2010, and were not on statin treatment in 2004 (n = 376,030). The first filled statin prescription date was defined as the date of entry in the study (index date). Participants also had to have at least 1 test result for total cholesterol level in the year before the index date (n = 351,987). Of them, we included participants who had at least 1 year of follow-up after the index date (n = 345,531).
Study variables
The outcome of interest was ICH, defined as primary discharge diagnosis with ICH (ICD-9 code 431). We conducted a validation study to evaluate the accuracy of using inpatient ICD-9 code 431 to identify ICH in the CHS database. An experienced neurologist reviewed the medical files of 365 cases of ICH that were detected in a subgroup of 97,071 patients considered to be highly persistent statin users. All patients were followed from the index date until outcome event (ICH), death, loss to follow-up, or June 30, 2017, whichever came first.
In order to increase comparability between statin users, a uniform scale for statins treatment was adopted by transforming the different types of statins into equivalent atorvastatin potency.11 For the purpose of this study, exposure to statins was defined by means of the average atorvastatin equivalent daily dose (AAEDD). AAEDD was calculated using data on statin type, dosage, and number of filled pills during the entire follow-up period. If the number of pills in the last filled prescription was higher than the remnant days of follow-up, we included only the corresponding number of pills in the calculation of AAEDD. For each patient, we estimated the cumulative atorvastatin equivalent dose (during the entire follow-up period), which in turn was divided by the number of days of follow-up to obtain the AAEDD. The AAEDD was classified into 3 predefined categories: (1) <10 mg/d, (2) 10–19.9 mg/d, and (3) ≥20 mg/d, with lowest category used as reference. Furthermore, we calculated the proportion of follow-up days covered (PDC) with statins by dividing the number of statin pills filled by the number of the entire follow-up days, which was used to assess the association of AAEDD with ICH among highly persistent statin users (PDC ≥80%).12
The most recent total cholesterol level test performed in the year before the index date was used to assess the association between baseline total cholesterol and ICH. Baseline total cholesterol level was classified into quartiles and the lowest quartile was used as reference category. The same method was applied to classify low-density lipoprotein (LDL) levels at baseline.
In addition, for each patient, the following data were retrieved from the computerized database of the CHS: demographic and other descriptive variables, health habits, presence of selected chronic medical conditions and ICH risk factors at baseline, medication use of selected drug categories, and medical services utilization, as shown in table 1.
Baseline characteristics of the study population stratified by categories of atorvastatin equivalent daily dose
Standard protocol approval, registration, and patient consent
This retrospective cohort study was based on automatically and routinely collected administrative and clinical data in a coded database. As such, approval was not sought from an ethics review board.
Statistical methods
Cox proportional hazard regression models were used to estimate the crude and the adjusted hazard ratio (HR) for the association between AAEDD, cholesterol, and ICH. AAEDD and cholesterol were tested as categorical variables, using the lowest category as reference, and as continuous variables.
Because of the large number of confounders, we performed adjustment for a disease risk score (DRS), a summary measure of disease probability. In comparison with conventional multivariate analyses, adjustment for the single variable DRS increases the efficiency of the analyses when controlling for the potential confounders.13,14 Previous simulation studies have shown that the DRS and propensity score had comparable performance. Furthermore, DRS has advantage when multiple comparison groups are studied.13,14 The DRS was estimated using a logistic regression model that included clinically most relevant ICH risk factors (age, sex, hypertension, diabetes, previous stroke, vascular disease, chronic kidney diseases, antiaggregants, and anticoagulant medications use) and other clinical covariates likely to be correlated with ICH (table 1). Adjusting for a large number of covariates for DRS may improve control of confounding, as these variables may collectively be proxies for unmeasured factors. In the main analysis, DRS was modeled as a continuous variable in the Cox proportional hazard regression models.
In addition, we performed the following sensitivity analyses: (1) adjusting for absolute change in total cholesterol after treatment (time-dependent variable), (2) adjusting for baseline LDL cholesterol instead of total cholesterol, (3) we repeated the analysis by adjusting for the clinically most relevant variables (i.e., standard multivariable regression), (4) to avoid assuming linearity between DRS and ICH, DRS was categorized into quintiles and were used in the Cox proportional hazard regression models once as stratifying variable and once modeled as categorical variable, (5) restricting the analysis to highly persistent statin users (PDC ≥80%). We performed an additional sensitivity analysis to assess the dose–response relationship between statin use and ICH type (deep vs lobar). This analysis was restricted to highly persistent statin users with PDC ≥80% (n = 97,071), in whom the diagnosis of ICH was validated and classified into lobar and deep ICH by reviewing the patients' files, as described earlier. To account for missing type of ICH, multiple imputation (MI) method (SAS [Cary, NC] MI procedure) was performed. Five imputation datasets were created, where all model variables were included in the imputation process.
An interaction was examined between AAEDD and relevant ICH risk factors (sex, age, hypertension, diabetes, and history of stroke/TIA).
Based on the distribution of AAEDD categories and study duration, this study has more than 90% power to detect HR of 0.8, comparing AAEDD >10 mg/d vs <10 mg/d, using 2-sided 5% significance level.
All statistical analyses were performed using SPSS Statistics 22.0 (IBM, New York, NY) and SAS version 9.3 software. For all analyses, p < 0.05 for the 2-tailed tests was considered statistically significant.
Data availability
Anonymized data will be shared by request from any qualified investigator.
Results
Descriptive statistics
A total of 345,531 patients with mean age 63.8 years (SD 10.0) were included in the study; 195,755 (56.7%) were female. Table 1 shows the distribution of baseline demographic and clinical characteristics of the study population stratified by the 3 AAEDD categories. Compared to the lowest category of AAEDD, patients in the highest category were younger (p < 0.001) and more likely to be male (p < 0.001). History of ICH was detected in 0.15% of those in the lowest AAEDD category compared to 0.13% in the highest category (p = 0.298). Anticoagulant use was more frequent among those in the lowest AAEDD category (2.6%) compared to in the highest AAEDD category (1.9%) (p < 0.001).
The median follow-up time was 9.5 years (interquartile range 7.6–11.0). Overall 1,304 patients had ICH during 3,110,593 person-years of follow-up, representing an ICH crude incidence rate of 41.9 per 100,000 person-years.
Among the subset of 365 patients selected for the validation of ICH diagnosis in the CHS database, 16 files could not be reached. Of the remaining 349 files, there was convincing evidence for ICH in 307 cases, yielding a positive predictive value of 88%. Among them, 166 (54.0%) were deep ICH, 109 (35.5%) were lobar ICH, and in 32 (10.5%) patients the location could not be determined.
Risk factors for ICH
The univariate and multivariate HRs for the association of each variable included in table 1 with ICH are shown in table e-1 (links.lww.com/WNL/A595). The strongest risk factors for ICH were prior ICH, with adjusted HR of 4.45 (95% confidence interval [CI] 2.80–7.10); prior stroke/TIA, 1.79 (1.53–2.01); alcohol abuse, 1.80 (1.15–2.81); use of anticoagulation, 1.58 (1.19–2.10); diabetes, 1.57 (1.40–1.77); hypertension, 1.36 (1.18–1.57); male sex, 1.48 (1.32–1.66); and older age, 1.03 (1.02–1.03) for each 1-year increase.
Statins dose and ICH
The crude incidence rate of ICH decreased in a dose–response manner with increasing AAEDD: 46.1 per 100,000 person-years in those with AAEDD <10 mg/d, 31.1 per 100,000 person-years in those with AAEDD 10–19.9 mg/d, and 27.3 per 100,000 person-years in those with AAEDD ≥20 mg/d (table 2). The distribution of time to ICH by AAEDD categories showed similar results, as depicted in the figure, A.
Descriptive statistics, incidence density rate, and crude hazard ratios (HRs) for the association between average atorvastatin equivalent daily dose, baseline total cholesterol, and intracerebral hemorrhage (n = 345,531)
(A) By average atorvastatin equivalent daily dose categories. (B) By baseline total cholesterol quartiles.
Compared to those with AAEDD <10 mg/d (reference), the adjusted HR for ICH was 0.68 (95% CI 0.58–0.79) and 0.62 (0.47–0.81) in those with AAEDD 10–19.9 mg/d and those with AAEDD ≥20 mg/d, respectively (table 3). The results were similar when AAEDD was examined as continuous variable, with adjusted HR of 0.82 (95% CI 0.76–0.90) for each 10 mg/d increase in AAEDD (table 3). The results did not change after further adjustment for the absolute change in total cholesterol after starting treatment, included as time-dependent variable, with adjusted HR of 0.86 (95% CI 0.79–0.94) for each 10 mg/d increase in AAEDD (table 4). The results were similar and robust with adjustment to baseline LDL cholesterol instead of total cholesterol (tables 5 and 6).
Adjusteda hazard ratios (HRs) (95% confidence intervals) for the association between average atorvastatin equivalent daily dose, baseline total cholesterol, and intracerebral hemorrhage examined as categorical variable (upper part of the table) and as continuous variable (lower part of the table), presented separately for all included patients and for highly persistent statin users (proportion of follow-up days covered [PDC] ≥80%)
Adjusteda hazard ratios (HRs) (95% confidence intervals) for the association between average atorvastatin equivalent daily dose, baseline total cholesterol, absolute change in total cholesterol (time-dependent variable), and intracerebral hemorrhage with exposure variables examined as continuous variables, and presented separately for all included patients and for highly persistent statin users (proportion of follow-up days covered [PDC] ≥80%)
Adjusteda hazard ratios (HRs) (95% confidence intervals) for the association between average atorvastatin equivalent daily dose, baseline low-density lipoprotein (LDL) cholesterol, and intracerebral hemorrhage examined as categorical variable (upper part of the table) and as continuous variable (lower part of the table), presented separately for all included patients and for highly persistent statins users (proportion of follow-up days covered [PDC] ≥80%)
Adjusteda hazard ratios (HRs) (95% confidence intervals) for the association between average atorvastatin equivalent daily dose, baseline low-density lipoprotein (LDL) cholesterol, absolute change in LDL cholesterol (time-dependent variable), and intracerebral hemorrhage with exposure variables examined as continuous variables, and presented separately for all included patients and for highly persistent statin users (proportion of follow-up days covered [PDC] ≥80%)
Adjusting for the clinically most relevant variables by means of traditional multivariable regression yielded similar results; adjusted HR for ICH 0.79 (95% CI 0.72–0.87) for each 10 mg/d increase in AAEDD. Stratifying by DRS quintiles showed that the adjusted HR for ICH was 0.76 (0.70–0.83) for each 10 mg/d increase in AAEDD. Modelling DRS quintiles as categorical variables showed consistent results with an adjusted HR for ICH of 0.76 (0.69–0.83) for each 10 mg/d increase in AAEDD.
Restricting the analyses to highly persistent statin users (PDC ≥80%) yielded similar results (tables 3–6). When looking at ICH type in this group, for each 10 mg/d increase in AAEDD, adjusted HR was 0.77 (95% CI 0.62–0.96) for deep ICH and 0.86 (95% CI 0.67–1.10) for lobar ICH (pfor interaction = 0.523).
No significant interaction was found between AAEDD and sex, age, hypertension, diabetes, or previous stroke/TIA (all pfor interaction >0.1), suggesting that the association between AAEDD and ICH is not modified by these factors.
Cholesterol level and ICH
The crude incidence rate of ICH decreased in a dose–response manner with increasing baseline total cholesterol quartiles: 64.4 per 100,000 person-years in the lowest quartile, 41.8 in the second quartile, 31.2 in the third quartile, and 31.6 per 100,000 person-years in the highest quartile (table 2). The distribution of time to ICH by baseline total cholesterol quartiles showed similar results, as depicted in the figure, B.
Compared to the lowest baseline total cholesterol quartile, the adjusted HR for ICH was 0.71 (95% CI 0.62–0.82), 0.55 (0.47–0.64), and 0.57 (0.49–0.67) in those in the second, third, and highest quartiles, respectively (table 3). The results were similar when baseline total cholesterol was examined as continuous variable with adjusted HR of 0.94 (0.92–0.95) for each 10 mg/d increase in baseline total cholesterol (table 3). The results did not change after further adjustment for the absolute change in total cholesterol, adjusted HR of 0.95 (0.93–0.96) for each 10 mg/d increase in baseline total cholesterol (table 4). Interestingly, in contrast to baseline total cholesterol, the absolute change in total cholesterol during follow-up showed opposite results, with HR of 0.98 (0.96–0.99) for each 10 mg decrease in total cholesterol level (table 4). The results were similar when an association was tested between ICH risk and baseline LDL cholesterol instead of total cholesterol (tables 5 and 6). Restricting the analyses to highly persistent statin users (PDC ≥80%) yielded similar results (tables 3 and 4).
Stratified analysis showed that the HR for ICH was 0.94 (0.92–0.96) for each 10 mg increase in total cholesterol level among those without prior stroke/TIA compared to 0.97 (0.94–1.0) in those with prior stroke/TIA (pfor interaction = 0.038), suggesting that the association might be slightly stronger among those without prior stroke/TIA.
Discussion
In this large nationwide study, we looked at the association between statin use and ICH encompassing cholesterol levels at baseline, on follow-up, and statin daily dose exposure. In line with previous studies,5,–,8 this study shows that the risk of ICH decreases with increasing cholesterol levels, reinforcing the validity of our results. Whether the association between ICH risk and cholesterol levels can be applied to patients treated with statins is still a matter of investigation. We also showed that statin use, in addition to its proven effectiveness in reducing cardiovascular mortality,3,15 may prevent ICH.
As yet, no consensus has emerged concerning statins and ICH risk, as results across studies ranged from elevated to no effect and decreased risk. Concerns raised following post hoc analysis of SPARCL3 and HPS4 studies, robust by a Cochrane review,16 were not found in other studies. Two meta-analyses, one17 of 31 randomized controlled trials with a total of 182,803 patients and another18 from 23 randomized trials and 19 observational studies comprising 248,391 patients, did not find a significant association between statin therapy and ICH. Some previous case-control studies support our results.19,20 It has been shown, in a multivariable model, that treatment with statins were associated with decreased risk of ICH as compared with a history of high cholesterol alone.20
This inconsistency across trials has led to uncertainty in decision-making such as balancing the risks of hemorrhagic vs ischemic consequences when prescribing statins, especially in patients with known cerebrovascular disease or previous ICH. Using a Markov decision model analysis,21 it was recommended to consider for patients with a history of ICH, particularly those with a lobar location, associated with cerebral amyloid angiopathy,22 to avoid the use of statins. However, our study suggests that the reduced risk of ICH associated with statins did not differ across ICH types, although it did not reach statistical significance in lobar ICH. Large prospective studies are still needed to conclusively confirm the safety of post lobar ICH statin use.
It is puzzling that despite promoting reduction in cholesterol levels, a known risk factor for ICH, also in our cohort, statins still protect from ICH in a dose-dependent manner. Of note, the clinical effect of statins is not limited to lipid-lowering but also derived from other2 pleotropic cholesterol-independent effects such as inhibition of inflammatory processes by attenuation of important immunologic mediators. Other plausible mechanisms that may explain our results include neuroprotection by increased nitric oxide levels mediated by upregulation of endothelial nitric oxide synthase, an antioxidative effect mediated by reduced lipoprotein oxidation, and improving cerebral blood flow.23
As relation between statins and ICH risk has not been definitely established, an antecedent statin use was associated with no harm and possibly better prognosis following ICH.7,24 Although different mechanisms participate to determine ICH risk and ICH outcome, previous studies, taken together with our results, suggest a role for statins in the complex pathophysiology of ICH and ICH recovery.
A notable strength of this study is being a population-based study with a large number of new statin users with all the advantages that such design confers compared to studies that include prevalent statin users. In addition, we showed that the positive predictive value of ICH diagnosis was 88%, in line with a previous systematic review.25 Including only statin users in our study decreases the effect of confounding by indication compared to studies looking at statin users relative to nonstatin users. In addition, restricting the analyses to highly persistent statin users decreases the influence of healthy user effect. Unlike most similar previous studies, this study has data on total and LDL cholesterol levels, both at baseline and following treatment, allowing us to report more accurate risk estimates. Our study has the potential of controlling for important risk factors like smoking and alcoholism that were missed in some previous studies. Controlling for DRS as single variable increases the efficiency of the analysis as compared to conventional multivariate analyses. Yet this study has some limitations; first, we rely solely on an administrative computerized database that was not specifically designed for the present study. Anticoagulant use was significantly more frequent among those in the lowest AAEDD category compared to the highest AAEDD category, raising concerns about confounding. Despite adjustment for anticoagulants and many other variables, some residual confounding might still exist in this observational study. We also looked at ICH location only among a subgroup of highly persistent statin users. Finally, misclassification of ICH may exist between the statin users group; however, this misclassification is likely to have been nondifferential and is likely to bias the results toward the null.
Our study suggests that statin use might be associated with decreased risk of ICH in a dose–effect manner. Fear of ICH should not discourage prescription of statins for primary and secondary prevention.
Author contributions
Walid Saliba: study design, data collection, data analysis, writing of the paper. Hedy S. Rennert: data analysis, provided critical review of the manuscript. Ofra Barnett-Griness: data analysis, provided critical review of the manuscript. Naomi Gronich: study design, provided critical review of the manuscript. Jeremy Molad: study design, data collection, provided critical review of the manuscript. Gad Rennert: study design, provided critical review of the manuscript.
Study funding
No targeted funding reported.
Disclosure
The authors report no disclosures relevant to the manuscript. Go to Neurology.org/N for full disclosures.
Footnotes
Go to Neurology.org/N for full disclosures. Funding information and disclosures deemed relevant by the authors, if any, are provided at the end of the article.
Editorial, page 197
CME Course: NPub.org/cmelist
- Received November 6, 2017.
- Accepted in final form April 19, 2018.
- © 2018 American Academy of Neurology
References
Disputes & Debates: Rapid online correspondence
- Author response to Vilaniam et al.
- Author response to Prof. Goldstein
- Reader response: Association of statin use with spontaneous intracerebral hemorrhage
- Reader response: Association of statin use with spontaneous intracerebral hemorrhage: A cohort study
NOTE: All authors' disclosures must be entered and current in our database before comments can be posted. Enter and update disclosures at http://submit.neurology.org. Exception: replies to comments concerning an article you originally authored do not require updated disclosures.
- Stay timely. Submit only on articles published within the last 8 weeks.
- Do not be redundant. Read any comments already posted on the article prior to submission.
- 200 words maximum.
- 5 references maximum. Reference 1 must be the article on which you are commenting.
- 5 authors maximum. Exception: replies can include all original authors of the article.
- Submitted comments are subject to editing and editor review prior to posting.