Antibiotics in Primary Prevention of Stroke in the Elderly
Background and Purpose— An increasing number of reports have linked infections to atherosclerosis and thrombosis. Thus, use of antibiotics may lower the risk of developing cerebrovascular disease. We investigated whether antibiotic use is associated with the risk of stroke in elderly individuals treated for hypertension.
Methods— A cohort of 29 937 elderly subjects initiating antihypertensive therapy between 1982 and 1995 was formed from the Quebec healthcare insurance database. A nested case-control design was used in which each subject hospitalized with a primary discharge diagnosis of stroke between 1987 and 1995 was matched on calendar time to 5 randomly selected controls from the cohort. Conditional logistic regression was used to estimate odds ratios of stroke after adjustment for predisposing factors.
Results— We identified 1888 cases and 9440 controls. The overall adjusted odds ratio for current antibiotic use was 0.80 (95% confidence interval, 0.63 to 1.01), and that for recent use was 0.81 (95% confidence interval, 0.70 to 0.94). Penicillin was the only individual antibiotic class that showed a protective association across different time windows. No significant association was found between stroke risk and the use of fluoroquinolones, macrolides, tetracyclines, or cephalosporins.
Conclusions— Although no clear, consistent associations between overall antibiotic use and cerebrovascular disease could be found, an intriguing association between penicillin use and stroke should be explored further.
An increasing number of reports are linking infections to atherosclerosis and thrombosis. The proposed mechanisms are their common interaction with the inflammatory process, involving endothelial injury, macrophage transformation, and perhaps antigenic mimicry. An intriguing hypothesis sees arteriosclerosis as a chronic inflammatory condition with a possible treatable infective component that may be responsible either for disease initiation or progression or as a precipitating event contributing to plaque rupture resulting in acute ischemic events.1–3 Under that assumption, subjects treated with antibiotics may be at lower risk of developing clinical manifestations such as cerebrovascular disease than untreated subjects.
To test this hypothesis, we evaluated the effect of antibiotic use, regardless of the clinical indication or specific antibacterial activity, in the primary prevention of stroke in a general population of elderly patients initiating pharmacological treatment for hypertension.
Sources of Data
This study was conducted through the use of the computerized administrative databases of the Quebec Health Ministry (Régie del’Assurance Maladie du Québec [RAMQ]), developed in the context of the universal insurance program provided to the elderly population of the Province of Quebec. Information for each outpatient prescription includes the nature, quantity, strength, and dosage of the drug, as well as the dispensing date and prescribed duration of the prescription. The RAMQ databases also provide data on all medical services offered to the study population. Data on admission and discharge dates for all hospitalizations, along with up to 9 discharge diagnoses coded using the ninth revision of the International Classification of Diseases (ICD-9), are available from the hospital discharge database. The coding accuracy of discharge diagnoses in these databases has previously been shown.4
See Editorial Comment; page e166
A case-control analysis of a cohort of newly diagnosed hypertensive patients was conducted, with cases defined as any hospitalization with a primary discharge diagnosis of stroke (ICD-9 code, 430 to 438). Admission date for stroke was labeled the index date. For each stroke hospitalization, a risk set of all subjects with the same month and year of cohort entry and still at risk for an event at the case’s index date was identified. Five controls, selected at random within these risk sets through incidence-density sampling, were matched to each case accordingly. A detailed description of the cohort is available.5
Exposure history for antibiotics was available for up to 14 years before the index date. Because the latency period for stroke in relation to antibiotic use is not known, several a priori–defined time windows were used to document exposure: current (within 30 days of index date), recent (within 30 days to 1 year), and past (>1 year before) use. The prespecified hierarchy assignment order for simultaneous exposure to >1 of these time windows was the following: current>recent>past use. For each of these time windows, indicator variables were created for any antibiotic class (our primary analysis) and each individual antibiotic class (fluoroquinolones, tetracyclines, macrolides, cephalosporins, penicillins, and others) and compared with no antibiotic use.
Adjustment Factors and Statistical Analyses
Cases of stroke were contrasted to controls regarding antibiotic use through multivariate conditional logistic regressions to account for the effect of matching. Hospital diagnoses or the use of prescribed medications at any time during follow-up was used to identify the following potential confounders: diabetes, prior coronary heart disease, cardiac arrhythmias (including atrial fibrillation), cardiomyopathy (including congestive heart failure), hyperlipidemia, and use of anticoagulants, corticosteroids, aspirin, and nonsteroidal anti-inflammatory drugs. We adjusted for all these factors in the analyses.
Data are not available on smoking status in these administrative databases. We attempted to circumvent the unavailability of such data by using a surrogate measure of the long-term complication of smoking composed of theophylline use or any hospitalization for pulmonary diseases. This measure was included in the multivariate analyses.
Adjusted odds ratios (ORs), approximations to the rate ratios, and 95% confidence intervals (CIs) are presented throughout this article.
Among a total of 29 937 subjects, we identified 1888 cases of stroke and 9440 matched controls (Table 1). On average, cases and controls were followed up for 5 years (median, 5 years; range, 1 day to 12.6 years) from cohort entry to index date. Table 2 shows the associations of antibiotics and risk of stroke in the various time windows. Any antibiotic use showed a protective association that attenuates with time. Only recent use showed a statistically significant association with stroke (OR, 0.81; 95% CI, 0.7 to 0.94). At least 24% of the strokes (n=459) represented recurrent events. Sensitivity analyses revealed no difference in results when these “prevalent” cases were excluded from the models (any recent use: OR, 0.81; 95% CI, 0.7 to 0.96) or when only ischemic strokes were used (any recent use: OR, 0.85; 95% CI, 0.7 to 1.01), which represented 80.4% of the cases.
Among individual antibiotic classes, only penicillin use was found to be associated with stroke, with stronger effects for current (OR, 0.53; 95% CI, 0.33 to 0.86) than recent (OR, 0.73; 95% CI, 0.62 to 0.86) and past (OR, 0.86; 95% CI, 0.77 to 0.97) use. Although macrolides and fluoroquinolones showed a similar trend, no statistically significant effect on stroke risk was found. Use of tetracyclines, cephalosporins, or others combined was also not associated with stroke.
Although our analysis showed a protective trend for overall antibiotic use, it failed to demonstrate a consistent significant association between antibiotic use and stroke. As for individual antibiotic classes, only penicillin showed a consistent and significant association with the occurrence of stroke in a population-based study of elderly hypertensive subjects.
Other studies have examined whether antibiotics effective against specific bacteria such as Chlamydia pneumoniae (fluoroquinolones, macrolides, or tetracyclines), regardless of their clinical indications, could lead to a decreased risk of ischemic events such as myocardial infarction.6–10 These studies have shown conflicting results. Our own experience failed to demonstrate a consistent protective association across antibiotic classes, although a nonsignificant protective trend diminishing with time was observed for antibiotics with antichlamydial activity (OR, 0.68; 95% CI, 0.46 to 1.00 for use in the previous 3 months; OR, 0.80; 95% CI, 0.58 to 1.11 for use in the previous 6 months).5 In contrast, only 1 prior study has examined the putative association between antibiotic use and the occurrence of ischemic stroke, with no significant associations.11 One reason for the negative result might be related to the particularly long period between their defined exposure and stroke outcome (mean time, 646 days).
A number of limitations pertaining to our study should be kept in mind. First, antibiotic use is only an indirect marker of the presence of infection, on which we have no clinical or serological data and which thus cannot be directly linked to the occurrence of stroke. We are also uncertain about patient compliance because actual antibiotic intake was not measured.
Because information on important risk factors is sometimes lacking, observational studies such as this one are especially sensitive to confounding and systematic biases. Smoking is one potentially important confounder of the association between antibiotic use and stroke, and unfortunately, no direct information relating to individual smoking habits is available in the Quebec administrative databases. Because it would be reasonable to think that smokers are more likely to be dispensed antibiotics, which are indicated, among other things, for exacerbation of chronic bronchitis, insufficient correction of cigarette use could bias our study results by decreasing any measured protective effect of antibiotics. Another potential confounding factor not controlled for in the present study was socioeconomic status. Because infectious diseases could be unequally distributed among different socioeconomic categories, antibiotic use could thus vary accordingly.
Another limitation pertains to the study source population. Because we had incomplete information on potential strokes occurring before our study period, it might have led to a lack of comparability of cases and controls with respect to preexisting atherosclerotic disease. However, our sensitivity analysis showed no differences in our findings when patients with multiple events occurring during the study period were excluded or when hemorrhagic strokes were excluded.
In addition to possible systematic biases, one should consider the role of random error or chance in our study findings. Because no attempt was made to account for multiple testing, type I error is a possible explanation for the statistically significant associations found. The relatively small number of cases and controls found in some of the current time windows could also have hindered our ability to demonstrate significant associations.
In conclusion, no clear, consistent effect of antibiotics use was found in relation to stroke in a high-risk elderly population. Further confirmation studies for penicillin use in other populations and an evaluation of the clinical significance of this finding remain to be performed.
This work was funded by the Réseau d’utilisation des médicaments du Fonds de la Recherche en Santé du Québec (FRSQ). This study used data provided by RAMQ and Ministère de la santé et des services sociaux du Québec. At the time this study was undertaken, Dr Bourgault was the recipient of a postdoctoral fellowship from the MRC of Canada at Yale University School of Internal Medicine (Boston, Conn). Dr Brophy receives support from FRSQ. Dr Suissa is the recipient of a distinguished scientist award from the Canadian Institutes of Health Research. The statements contained herein do not necessarily represent those of the Quebec government.
Results of this study were presented at the 17th International Conference on Pharmacoepidemiology, Toronto, Canada, August 23–26, 2001.
- Received March 17, 2003.
- Revision received April 3, 2003.
- Accepted April 29, 2003.