Preadmission Use of Glucocorticoids and 30-Day Mortality After Stroke
Background and Purpose—The prognostic impact of glucocorticoids on stroke mortality remains uncertain. We, therefore, examined whether preadmission use of glucocorticoids is associated with short-term mortality after ischemic stroke, intracerebral hemorrhage (ICH), or subarachnoid hemorrhage (SAH).
Methods—We conducted a nationwide population-based cohort study using medical registries in Denmark. We identified all patients with a first-time inpatient diagnosis of stroke between 2004 and 2012. We categorized glucocorticoid use as current use (last prescription redemption ≤90 days before admission), former use, and nonuse. Current use was further classified as new or long-term use. We used Cox regression to compute 30-day mortality rate ratios with 95% confidence intervals (CIs), controlling for confounders.
Results—We identified 100 042 patients with a first-time stroke. Of these, 83 735 patients had ischemic stroke, 11 779 had ICH, and 4528 had SAH. Absolute mortality risk was higher for current users compared with nonusers for ischemic stroke (19.5% versus 10.2%), ICH (46.5% versus 34.4%), and SAH (35.0% versus 23.2%). For ischemic stroke, the adjusted 30-day mortality rate ratio was increased among current users compared with nonusers (1.58, 95% CI: 1.46–1.71), driven by the effect of glucocorticoids among new users (1.80, 95% CI: 1.62–1.99). Current users had a more modest increase in the adjusted 30-day mortality rate ratio for hemorrhagic stroke (1.26, 95% CI: 1.09–1.45 for ICH and 1.40, 95% CI: 1.01–1.93 for SAH) compared with nonusers. Former use was not substantially associated with mortality.
Conclusions—Preadmission use of glucocorticoids was associated with increased 30-day mortality among patients with ischemic stroke, ICH, and SAH.
Glucocorticoids are potent immunosuppressive drugs used in the treatment of an array of inflammatory and autoimmune disorders.1 The prevalence of use is high, and ≈3% of all Danes redeem at least one systemic glucocorticoid prescription each year.2 The annual incidence of first-time stroke is ≈600 000 cases in the United States3 with a 30-day mortality of ≈11% for ischemic stroke,4 34% for intracerebral hemorrhage (ICH),4 and 29% for subarachnoid hemorrhage (SAH).5 Patients using high doses of glucocorticoids have a 1.7-fold increased risk of stroke,6 but whether their outcome after a stroke also is affected remains unclear.
Stroke is often complicated by cerebral edema with subsequent clinical deterioration or death.7 The edema exerts pressure on the surrounding structures, causing direct vascular occlusion and vasospasm through the inflammatory process.7 In theory, glucocorticoids could help to reduce edema following stroke and thus improve clinical outcome. A possible negative effect on mortality could be mediated through adrenal suppression compromising the cortisol response to critical illness like stroke.8 Glucocorticoids also cause immunosuppression at supraphysiological doses,8 possibly leading to infections9 that can be fatal in the course of stroke.
Two small meta-analyses (<500 patients) have examined the effect of glucocorticoid administration in the acute phase of stroke on 1-month mortality and reported neutral results for ischemic stroke,10 ICH,11 and SAH.11 However, the results were inconsistent between trials included in the studies with reports of both increased and decreased mortality.
Given the high prevalence of glucocorticoid use, any association with mortality after stroke would have important public health and clinical implications. We, therefore, investigated whether preadmission glucocorticoid use is associated with short-term mortality after stroke.
We conducted a nationwide population-based cohort study in Denmark during the period July 1, 2004, to December 31, 2012. During this time period, the total cumulative underlying population at risk was 6 379 918 inhabitants. The study period started 6 months after the initiation of the Danish National Health Service Prescription Database on January 1, 2004, to ensure that at least 6 months of preadmission prescription history was available for all participants.12 The Danish health care system provides free and unfettered access to general practitioners and hospitals for the entire population and refunds a variable proportion of the costs of prescribed medication, including systemic glucocorticoids.12 In Denmark, care for patients with medical emergencies, including stroke, is provided by public hospitals.13 We linked medical registries using the unique 10-digit central personal registry number assigned to each Danish citizen at birth and to residents upon immigration.14
Patients With Stroke
We used the Danish National Patient Registry (DNPR), covering all Danish hospitals,15 to identify all persons in the Danish population with a first-time inpatient hospitalization for stroke during the study period. As approximately two thirds of all unspecified strokes are known to be ischemic strokes, we classified unspecified strokes (40% of all stroke diagnoses) as ischemic strokes.16 Follow-up was virtually complete except for one patient who emigrated within the first 30 days of follow-up. The DNPR contains data on admission and discharge dates and discharge diagnoses from all Danish nonpsychiatric hospitals since 1977 and on emergency room and outpatient clinic visits since 1995.15 Each hospital discharge is assigned one primary diagnosis and ≤19 secondary diagnoses classified according to the International Classification of Diseases, 8th revision until the end of 1993 and 10th revision thereafter.15 Both primary and secondary diagnoses were used to identify patients with ischemic stroke, ICH, and SAH.
We used the prescription database to identify all prescriptions redeemed for systemic glucocorticoids by study participants.12 Pharmacies in Denmark are equipped with electronic accounting systems, primarily used to provide reimbursement from the National Health Service.12 For each redeemed prescription, the patient’s central personal registry number, the amount and type of drug prescribed according to the Anatomical Therapeutic Chemical classification system, and the date the drug was dispensed are transferred electronically from the pharmacy to the prescription database.12
Based on methods used previously,17 we defined current users as patients whose most recent prescription redemption was within 90 days before hospital admission for stroke. As the drugs’ anti-inflammatory effects may be less pronounced and side effects more pronounced in long-term users,8 there could be differences in the association with mortality according to the length of use.18 Current users, therefore, were categorized further into new users, who redeemed their first-ever prescription within 90 days before admission, and long-term users, who redeemed their first-ever prescription >90 days before admission. We defined former users as patients whose last prescription redemption was between 90 and 180 days before admission. Nonusers (ie, individuals who filled no prescriptions for any glucocorticoids within 180 days before admission) constituted the reference cohort in all comparisons.
We computed prednisolone-equivalent cumulative doses as described previously.17 Briefly, the prednisolone-equivalent cumulative dose was obtained by multiplying number of pills/injections×dose per pill/injection×prednisone conversion factor for each prescription and then summing across all prescriptions redeemed within 90 days of admission (current use).
We used the Danish Civil Registration System to obtain information on all-cause mortality.14 The Civil Registration System has recorded all changes in vital status and migration for the entire Danish population since 1968, with daily electronic updates.14
We used the complete inpatient and outpatient medical history available in the DNPR15 to ascertain presence of potentially confounding comorbidities. We categorized the severity of comorbidity using the Charlson Comorbidity Index (CCI), a scoring system that has been adapted for use with hospital discharge data.19–21 The CCI assigns between 1 and 6 points to a range of diseases, depending on the strength of their relation with mortality.19,20 We computed the total CCI score for each patient (excluding cerebrovascular disease and hemiplegia from the score) and defined 3 categories of comorbidity based on scores of 0 (low), 1–2 (moderate), and ≥3 (high).
The complete inpatient and outpatient medical history available in the DNPR provided information on known prognostic factors (myocardial infarction, atrial fibrillation or flutter, intermittent arterial claudication, and diabetes mellitus)22 and other potential confounders with a prognostic impact on stroke or an association with glucocorticoid use, including angina pectoris, venous thromboembolism, hypertension, rheumatoid arthritis and related diseases, connective tissue disease, chronic obstructive pulmonary disease, inflammatory bowel disease, obesity, chronic kidney disease, alcoholism-related diseases, and cancer. To increase the sensitivity of diagnoses of diabetes mellitus, chronic obstructive pulmonary disease, and alcoholism-related diseases, we searched the prescription database for any previous prescriptions for diabetic medications, respiratory medications, and alcohol deterrents.
We obtained information from the prescription database12 on concurrent use (ie, ≤90 days before admission) of angiotensin-converting enzyme inhibitors/angiotensin receptor blockers, beta-blockers, calcium channel blockers, diuretics, nitrates (if ≥2 prescriptions were registered), statins, acetylsalicylic acid, clopidogrel, vitamin K antagonists, selective serotonin reuptake inhibitors, and nonaspirin nonsteroidal anti-inflammatory drugs.
We characterized the stroke cohort according to sex, age group (<60, 60–69, 70–79, ≥80 years), individual comorbidities, and comedication use. We followed all patients from their hospital admission date until death, emigration, December 31, 2012, or 30 days of follow-up, whichever came first.
We used a Cox proportional-hazards regression analysis to compute the hazard ratio as a measure of the mortality rate ratio (MRR) within 30 days of hospital admission for current, new, long-term, and former use compared with nonuse. In the analysis using prednisolone-equivalent cumulative doses, we assessed systemic glucocorticoids according to 3 categories (≤150, >150–500, and >500 mg) based on the dose distribution in the study population.
The analysis was adjusted for sex, age groups, and the individual comorbidities and comedications listed in Table 1. The proportional hazard assumption was assessed using log–log plots and found valid. We repeated the analysis stratifying by sex, age group, CCI score, and presence/absence of cardiovascular variables that may relate to exposure (rheumatoid arthritis and related diseases, connective tissue disease, chronic obstructive pulmonary disease, inflammatory bowel disease, cancer, diabetes mellitus, chronic kidney disease, and current use of nonsteroidal anti-inflammatory drugs) or outcome (myocardial infarction, congestive heart failure, atrial fibrillation or flutter, angina pectoris, and hypertension).
We performed 5 sensitivity analyses. First, to increase the positive predictive value of the stroke diagnosis, we restricted the analyses to patients who had a computed tomography or magnetic resonance imaging scan registered in the DNPR during the stroke admission. Second, to examine the impact of differences in exposure definitions, we repeated the analysis substituting the 90-day exposure window with 60-day and 30-day exposure windows. Third, to reduce the potential for confounding by indication for glucocorticoid use, we directly compared current users with former users and former users with nonusers. Fourth, to examine the effect among severe stroke cases, we stratified by intensive care unit (ICU) admission and length of hospital stay as proxy measures of severity. Fifth, we divided the follow-up period into 4 periods by weeks since admission. Analyses were performed using SAS version 9.2 (SAS Institute Inc, Cary, NC). The study was approved by the Danish Data Protection Agency (record number 2011-41-5755). Because this study did not involve contact with patients or any intervention, it was not necessary to obtain permission from the Danish Scientific Ethical Committee or to obtain patient consents. Relevant International Classification of Diseases and Anatomical Therapeutic Chemical codes along with the prednisone conversion factors and the equivalent glucocorticoid dose for each type of glucocorticoid are provided in Table I in the online-only Data Supplement.
Patient characteristics are shown in Table 1. We identified 100 042 patients with a first-time stroke between 2004 and 2012. Of these, 83 735 (83.7%) had ischemic stroke (median age: 74 years), 11 779 (11.8%) had ICH (median age: 72 years), and 4528 (4.5%) had SAH (median age: 58 years). A total of 4645 patients (4.6%) were current users of glucocorticoids, 1535 (1.5%) were former users, and 93 862 (93.8%) were nonusers. All cardiovascular diseases, obesity, diabetes mellitus, chronic kidney disease, cancer, and use of all cardiovascular drugs were more common among glucocorticoid users than among nonusers across all stroke subtypes. Current and former users of glucocorticoids had substantially higher Charlson Comorbidity Index scores than nonusers among all stroke patients. Among current and former users of glucocorticoids, women were overrepresented in all stroke subtypes.
Mortality estimates are provided in Tables 2 and 3. For patients with ischemic stroke, the adjusted 30-day MRR was 1.58 (95% confidence interval (CI): 1.46–1.71) for current users compared with nonusers, mainly driven by the effect of new use (MRR=1.80, 95% CI: 1.62–1.99). Mortality estimates were lower for hemorrhagic strokes. Among ICH patients, the MRR was 1.26 (95% CI: 1.09–1.45) for current users compared with nonusers, with no notable difference between new and long-term users. Among SAH patients, the MRR was 1.40 (95% CI: 1.01–1.93) for current users compared with nonusers, with the strongest association for new users (MRR=1.54, 95% CI: 1.03–2.31).
Comparing former users with nonusers, no substantial association with 30-day mortality was present among patients with ICH (MRR=1.04, 95% CI: 0.81–1.34), SAH (MRR=1.26, 95% CI: 0.81–1.95), or ischemic stroke (MRR=1.17, 95% CI: 1.01–1.36).
When current users were compared directly with former users, the adjusted MRRs were similar to the MRRs in the primary analysis for ischemic stroke and ICH patients (Table II in the online-only Data Supplement). However, there was no clear association with mortality for SAH patients (MRR=0.73 [95% CI: 0.39–1.35]).
The results of the primary analysis of current versus nonuse were supported by the analyses of prednisolone-equivalent cumulative doses versus nonuse. These revealed a dose–response relation for both ischemic and hemorrhagic strokes. The dose–response relation was most pronounced for ischemic stroke (MRR increased from 1.43 [95% CI: 1.24–1.65] to 1.74 [95% CI: 1.52–1.98]) and ICH (from 1.13 [95% CI: 0.86–1.48] to 1.52 [95% CI: 1.17–1.96]), whereas the increase in MRR with increasing dose was more subtle in the SAH group (from 1.40 [95% CI: 0.85–2.32] to 1.58 [95% CI: 0.81–3.08]; Table 3). The results were robust in analyses restricted to new and long-term users (data not shown).
There was no notable modification of the effect within strata of sex or age. Within strata of comorbidity level and relevant diseases, the MRR was consistently higher among patients without disease, presumably because of lower baseline hazard (Table III in the online-only Data Supplement).
The results remained robust in the analysis restricted to computed tomography- or magnetic resonance imaging-confirmed diagnoses (Table IV in the online-only Data Supplement) and when using a 60-day or 30-day exposure window (Tables V and VI in the online-only Data Supplement). Among patients with an ICU admission, the mortality estimates were close to unity, whereas stratification by length of hospital stay did not change the results substantially (Table VII in the online-only Data Supplement). Although the relative impact associated with glucocorticoid use increased slightly with time since stroke, the vast majority of stroke-related deaths occurred during the first week and drove the overall results (Table VIII in the online-only Data Supplement).
In this large population-based cohort study including >100 000 patients hospitalized for first-time stroke, we found that preadmission use of glucocorticoids was associated with increased 30-day mortality. The association was most pronounced for ischemic stroke. For all stroke subtypes, we found a dose–response relation between prednisolone-equivalent cumulative doses and 30-day mortality. Among patients with an ICU admission, we observed a weakened association, which likely represents confounding by the higher baseline mortality rate among ICU patients. Among former users, who recently discontinued their glucocorticoid treatment, a small increase in mortality was observed among ischemic stroke and ICH patients which may relate to adrenal insufficiency and after impaired cortisol response to critical illness like stroke.8
This is the first population-based cohort study to examine the effect of preadmission use of glucocorticoids on short-term mortality after stroke. However, the effect of glucocorticoids in acute ischemic stroke was investigated in a recent meta-analysis of 8 randomized trials, including 466 patients. No difference in mortality was found at 1 month (odds ratio =0.97, 95% CI: 0.63–1.47).10 Another meta-analysis investigated the effect of glucocorticoids on mortality in acute ICH and SAH and included 8 randomized trials with a total of 326 patients (140 patients in 1 SAH trial and 186 patients in 4 ICH trials).11 As point estimates in this study were comparable with our results (SAH: relative risk =1.49, 95% CI: 0.85–2.61; ICH: relative risk =1.14, 95% CI: 0.91–1.42), the small sample size may be the limiting factor for statistical significance. Furthermore, as most adverse effects of glucocorticoids are notoriously known to arise after longer time of use,8 the one-time administration of glucocorticoids in the acute phase of stroke in the meta-analyses may not leave enough time for adverse effects to develop within 1 month of follow-up.
The pathophysiological mechanism responsible for a potential negative influence of glucocorticoids on stroke mortality may be multifactorial. Glucocorticoids induce adrenal suppression that compromises the cortisol response to critical illness like stroke.8 Moreover, long-term glucocorticoid use is associated with increased risk of atrial fibrillation, diabetes mellitus, and hypertension,8,23 which are all well-known predictors of stroke mortality.24,25 At supraphysiological doses, glucocorticoids cause immunosuppression, possibly leading to infections like pneumonia, which can be deadly in the course of stroke.24 Finally, because of the mineralocorticoid effect of glucocorticoids, sodium and water retention can occur, resulting in exacerbation of heart failure.8
The validity of the study results depends on several factors. We conducted this population-based study within a tax-supported uniformly organized health care system, which limited the risk of selection bias.26
The positive predictive value of acute stroke diagnoses in the DNPR has been estimated at ≈97% for ischemic stroke, 74% for ICH, and 67% for SAH.16 As it is unlikely that coding errors would be associated with glucocorticoid use, any misclassification would be nondifferential and bias the results toward the null. Because we classified unspecified strokes as ischemic strokes, a few ICHs (≈6%) were inevitably misclassified as ischemic strokes.16 This misclassification was most likely nondifferential and would bias the estimates toward unity.
Data on reimbursed medications are virtually complete in the prescription database,12 and systemic glucocorticoids are not sold over-the-counter in Denmark. We, therefore, identified all patients with redeemed prescriptions for systemic glucocorticoids. As glucocorticoids have a direct beneficial effect on a wide range of symptoms, this is likely to increase adherence. Any misclassification of drug exposure because of nonadherence would attenuate the estimate of association toward unity and thus cannot explain our findings of increased mortality.
Despite extensive confounder adjustment, our study is vulnerable to residual and unmeasured confounding. Specifically, residual confounding of nonsteroidal anti-inflammatory drug use cannot be ruled out as low-dose (≤200 mg) nonsteroidal anti-inflammatory drugs are sold over-the-counter. Furthermore, confounding by indication of the underlying diseases that lead to glucocorticoid use may bias our results. Despite the establishment of a dose–response relation between increasing prednisolone-equivalent cumulative doses and mortality, we cannot rule out at least some influence of temporal variations in disease severity. Thus, exacerbation of the underlying diseases for which glucocorticoids were prescribed may be associated with both increased use of glucocorticoids and mortality after stroke, especially for new users. However, we have adjusted for the most important of these diseases, including cancer, chronic pulmonary disease, inflammatory bowel disease, rheumatoid arthritis, and other connective tissue diseases.
The observational nature of our study design and lack of detailed clinical data prevent us from suggesting specific guidelines for the clinical care of stroke patients. We have, however, identified a high-risk group for a poor outcome, and efforts should be done to prevent common complications to stroke, such as venous thromboembolism, infections, and fractures among these patients. Furthermore, by integrating our results with established risk factors for in-hospital mortality (ie, age, stroke severity, diabetes mellitus, atrial fibrillation, hypertension,22,24,25 and use of selective cyclooxygenase [COX]-2 inhibitors27), a more comprehensive identification of high-risk stroke patients is possible.
In conclusion, this study showed that current use of systemic glucocorticoids is associated with increased risk of short-term mortality after ischemic stroke, ICH, and SAH. Although residual confounding may partly explain these findings, we consider a causal relation possible because the association followed a dose gradient and persisted after adjustment for CCI and diseases for which glucocorticoids are prescribed. Hence, our observations merit clinical attention when prescribing glucocorticoids to patients at risk of stroke.
We thank Dr Jan Vandenbroucke for useful critical comments.
Sources of Funding
The study was supported by (1) the Program for Clinical Research Infrastructure (PROCRIN) established by the Lundbeck Foundation and the Novo Nordisk Foundation and (2) the Aarhus University Research Foundation. Neither funding source had a role in the design, conduct, analysis, or reporting of the study.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.115.012231/-/DC1.
- Received December 2, 2015.
- Revision received December 21, 2015.
- Accepted December 22, 2015.
- © 2016 American Heart Association, Inc.
- Go AS,
- Mozaffarian D,
- Roger VL,
- Benjamin EJ,
- Berry JD,
- Borden WB,
- et al
- Thorvaldsen P,
- Davidsen M,
- Brønnum-Hansen H,
- Schroll M.
- Lynge E,
- Sandegaard JL,
- Rebolj M.
- Ray WA.
- Goldstein LB,
- Samsa GP,
- Matchar DB,
- Horner RD.
- Thygesen SK,
- Christiansen CF,
- Christensen S,
- Lash TL,
- Sørensen HT.
- Andersen KK,
- Andersen ZJ,
- Olsen TS.
- Carter AM,
- Catto AJ,
- Mansfield MW,
- Bamford JM,
- Grant PJ.
- Heuschmann PU,
- Kolominsky-Rabas PL,
- Misselwitz B,
- Hermanek P,
- Leffmann C,
- Janzen RW,
- et al