1625

Risk of Hospitalization for Cardiovascular Disease after Use of Macrolides and Penicillins: A Comparative Prospective Cohort Study Lars Østergaard,1 Henrik Toft Sørensen,2 Jes Lindholt,5 Tom E. Sørensen,3 Lars Pedersen,2 Thore Eriksen,4 and Paul Lehm Andersen4

1

Infectious Disease Unit, Alfred Hospital, Melbourne, Australia; Department of Clinical Epidemiology, Aarhus University and Aalborg Hospitals, 3Department of Computer Science, University of Aarhus, and 4Department of Infectious Disease and Internal Medicine, Aarhus University Hospital, Skejby Sygehus, Aarhus, and 5Department of Vascular Surgery, Viborg Sygehus, Viborg, Denmark

2

Chlamydia pneumoniae has been associated with cardiovascular diseases, and C. pneumoniae infection is treatable with macrolides. In this comparative cohort study, 634 users of macrolides and 3827 users of penicillins were identified from the Danish Health Service Registry of Prescriptions and followed up for an average of 6 months. The patients were then linked to the Regional Hospital Discharge Registry to assess the outcome of hospitalization for cardiovascular disease. In the first 3 months, the relative risk (RR) of admission for a cardiovascular disease was 0.48 (95% confidence interval, 0.27–0.88) in users of macrolides compared with users of penicillins. No difference was seen after 3 months. Interaction analyses indicated that the lower risk seen in users of macrolides could be more pronounced in patients without versus those with a previous cardiovascular disease (RR, 0.39 vs. 0.52), in patients ⭓60 versus !60 years old (RR, 0.39 vs. 0.64), and in men versus women (RR, 0.35 vs. 0.67).

Risk factors for cardiovascular disease, such as age, sex, hypercholesterolemia, and hypertension, can hardly alone explain the variation in risk of this disease [1]. During the past decade, inflammation and infection have received attention as possible additional risk factors in the pathogenesis of cardiovascular disease. It has been observed that the levels of inflammatory markers, for example, C-reactive protein, are increased in patients with cardiovascular disease [2, 3]. Also, pathoanatomic studies have shown a significant amount of inflammatory cells in atherosclerotic lesions [4]. The factors that initiate or precipitate this inflammation are not known, but in several serologic, pathologic, and animal studies [5–7], Chlamydia pneumoniae has been linked to the inflammatory process involved in cardiovascular disease. C. pneumoniae has also been associated with acute and chronic coronary syndromes [8], strokes [9], and aortic aneurysms [10]. Macrolide treatment of patients with acute coronary syndromes reduces the risk of secondary ischemic events within the first 30 days of follow-up in some studies [11, 12], but the

effect seems to be reduced after a follow-up of 180 days [13]. Other studies have not been able to show an effect of macrolide treatment on acute coronary syndromes after 6 months of follow-up [14]. Two case-control studies of the effect of macrolides on risk of first-time myocardial infarction have been published. Meier et al. [15] found no association when macrolides were prescribed in the previous 3 years, nor did Jackson et al. [16], who assessed macrolide treatment in the year preceding myocardial infarction. Macrolides, in contrast to penicillins, have an antimicrobial effect on C. pneumoniae. The optimal treatment regimen is, however, not known, and it could be hypothesized that the current recommended courses of macrolide therapy may have an effect of only short duration on cardiovascular diseases. In this prospective comparative cohort study, we therefore assessed the time-dependent effect of macrolide therapy compared with penicillin therapy on the risk of hospitalization due to cardiovascular disease in subjects aged 50–69 years.

Received 9 January 2001; revised 27 February 2001; electronically published 9 May 2001. Presented in part: 4th meeting of the European Society for Chlamydia Research, Helsinki, 20–23 August 2000 (abstract 436). The study was approved by the regional ethics committee of Aarhus County and the Danish Data Protection Agency. Financial support: Scandinavian Society of Antimicrobial Chemotherapy; NASTRA (Det Nationale Strategiudvalg for Sundhedsvidenskab; grant 97 00 677). Reprints or correspondence: Dr. Lars Østergaard, Alfred Infectious Disease Unit, Alfred Hospital, PO Box 315, Prahran, Melbourne, Victoria 3181 Australia ([email protected]).

Population and Methods

The Journal of Infectious Diseases 2001; 183:1625–30 䉷 2001 by the Infectious Diseases Society of America. All rights reserved. 0022-1899/2001/18311-0010$02.00

Study population. Aarhus County, Denmark, has ∼600,000 inhabitants. Those 50–69 years old (n p 151,000) were chosen as the study population. Exposure data. The Danish National Health Service provides tax-supported health care to all inhabitants, ensuring free access to general practitioners and hospitals and refunding a variable portion of the costs of drugs prescribed by physicians. The National Health Service, Aarhus County, retains key information about prescriptions for refundable drugs dispensed from all of the pharmacies in the county. This information includes the personal identification number of the customer, type of drug prescribed

1626

Østergaard et al.

according to the anatomic therapeutic chemical classification system, and the date of prescription. The personal identification number comprises 10 digits that encode sex and date of birth. In the present study, information on prescriptions of macrolides and penicillins was collected from 1 March to 1 November 1998. During these 8 months, we identified 634 users of macrolides and 3827 users of penicillins. These were the 2 cohorts of the study. Antibiotics are prescribed only by doctors in Denmark. Cephalosporins, tetracyclines, or quinolones are not subsidized by the Danish government, and the effect of these antibiotics could therefore not be assessed. Outcome data. The 2 cohorts were linked to the regional Hospital Discharge Registry (HDR) in Aarhus County, which retains on a permanent basis key information on all patients discharged from all hospitals in the county since 1994. The files of the HDR include the identification number of the patient, date of discharge, and up to 20 discharge diagnoses, coded according to the Danish version of the International Classification of Diseases, 10th revision (ICD-10). By use of the HDR, we collected information about any discharge diagnosis of cardiovascular disease before and after the prescription of antibiotics. The codes for discharge diagnoses of cardiovascular diseases were the following: ICD-10 I20–I20.9, angina pectoris; ICD-10 I21–I22.9, acute myocardial infarction; ICD10 I23–I23.8, complications of acute myocardial infarction; ICD10 I46–I46.9, cardiac arrest; ICD-10 I50–I50.9, heart failure; ICD-10 I63–I63.9, cerebral infarction; ICD-10 I64–I64.9, stroke;

JID 2001;183 (1 June)

ICD-10 I65–I66.9, occlusion and stenosis of precerebral arteries; ICD-10 I70–I70.9, atherosclerosis of arteries of extremities; ICD10 I71–I71.9, aortic aneurysms; ICD-10 I74–I74.9, arterial thrombosis; and ICD-10 I80–I80.9, thrombophlebitis. The diagnoses of interest with regard to confounders and independent risk factors included hypertension, diabetes mellitus types 1 and 2, hypercholesterolemia, and chronic obstructive pulmonary disease (COPD). The latter was used as a proxy measure for smoking. Information on these risk factors was obtained from 2 sources. One was by “looking back” in the HDR for a period of 4 years before the prescription of macrolides or penicillins to see if the patient had a discharge diagnosis for any of the risk factors. The other source was the registry of prescriptions, which was searched for prescriptions for any drugs being used to treat any of the risk factors. Thus, information on the risk factors was also sought among patients being treated for their risk factor in an outpatient setting. Statistical analyses. The follow-up period regarding hospitalization began on the date of the prescription and ended at the day of admission to hospital, the date of death, or the end of the study (19 January 1999). Patients treated with both macrolides and penicillins during the study period were excluded from analyses. Patients were included only once in the study, and only the cardiovascular diagnosis that first was obtained was used as the outcome. A Cox proportional hazard model was used to estimate the relative risk of hospitalization due to cardiovascular disease, adjusted for confounders (sex, previous cardiovascular disease, hyperten-

Table 1. Characteristics of 4461 Danish patients in a study of the effect of macrolides versus penicillins on the risk of cardiovascular disease.

Characteristic Sex Male Female Age, years ⭓60 !60 Previous cardiovascular disease Yes No Hypercholesterolemia Yes No Hypertension Yes No Chronic obstructive pulmonary disease Yes No Type 1 diabetes Yes No Type 2 diabetes Yes No Discharge diagnosis of cardiovascular disease during follow-up Yes No

Percentage of total patients (n p 4461)

P

Macrolides

Penicillins

Total no. of patients

272 362

1626 2201

1898 2563

42.5 57.5

.88

338 296

2004 1823

2342 2119

52.5 47.5

.69

146 488

872 2955

1018 3443

22.8 77.2

.93

27 607

196 3631

223 4238

5.0 95.0

.41

223 411

1359 2468

1582 2879

35.5 64.5

.99

180 454

747 3080

927 3534

20.8 79.2

!.01

18 616

157 3670

175 4286

3.9 96.1

.16

22 612

152 3675

174 4287

3.9 96.1

.62

32 602

303 3524

335 4126

7.5 92.5

.01

JID 2001;183 (1 June)

Macrolides and Risk of Hospitalization

1627

Table 2. Discharge diagnoses previously shown to be associated with Chlamydia pneumoniae among patients with and without previous cardiovascular disease treated with macrolides versus penicillins. Previous cardiovascular disease (secondary prophylaxis) Diagnosis Angina pectoris Acute myocardial infarction Cerebral infarction Stroke (infarct not proven) Aortic aneurysms Other cardiovascular diagnoses Total

ICD-10 code

Total no. of events

I20–I20.9 I21–I21.9 I63–I63.9 I64–I64.9 I71–I71.9

145 49 28 21 11 81 335

Macrolides (n p 146) 14 4 1 1 0 2 22

(9.6) (2.7) (0.7) (0.7) (1.4) (15.1)

Penicillins (n p 872) 111 31 8 5 9 12 176

(12.7) (3.6) (0.9) (0.6) (1.0) (1.4) (20.2)

No previous cardiovascular disease (primary prophylaxis) Macrolides (n p 488) 1 2 0 1 0 6 10

(0.2) (0.4) (0.2) (1.2) (2.0)

Penicillins (n p 2955) 19 12 19 14 2 61 127

(0.6) (0.4) (0.6) (0.5) (0.1) (2.1) (4.3)

NOTE. Data are no. (%) of events, unless otherwise specified. ICD-10, International Classification of Diseases, 10th revision.

sion, hypercholesterolemia, type 1 diabetes, type 2 diabetes, and COPD). In all models, the age was the basic time scale; that is, persons in the study cohort were first at risk from their age at inclusion into the study and until a cardiovascular event or censoring due to death or study end. By use of the PROC PHREG procedure in SAS 6.12 (SAS Institute), the proportional hazards assumption was evaluated by assessing the parallelism of the log integrated hazard function for each of the study covariates. Models were stratified by variables that did not satisfy this assumption. We examined interaction by use of product terms in the Cox models and by stratified analyses. To examine trends in the relative risks more closely, time-dependent covariates were incorporated into the models, reflecting the relative risk in disjoint time intervals: 0–3 months after treatment with macrolide (exposure), 3–6 months after treatment, and 1 6 months after treatment.

Results During the study period, the 4461 users of the 2 antibiotics contributed a total of 2385 person-years of observation. A total of 335 patients had an admission with a discharge diagnosis of cardiovascular disease during the study period, corresponding to an overall cumulative incidence of 14.0%. The baseline characteristics of the study population are shown in table 1. As seen, patients with a previous diagnosis of COPD were more likely to be treated with macrolides (180/927; 19.4%) than were patients with no diagnosis of COPD (454/3534; 12.8%) (table 1). No significant difference in use of macrolides and penicillins was seen among the other risk factors studied. Overall, of 634 patients who received macrolides, 32 (5.0%) were admitted to hospital for a cardiovascular disease, compared with 303 (7.8%) of 3827 patients who received penicillins (x 2 p 6.04; P ! .02). In the subgroup of patients with no previous history of cardiovascular disease, 10 admissions (2.0%) were seen in 488 users of macrolides, versus 127 admissions (4.3%) in 2955 users of penicillins (x 2 p 5.5 ; P ! .02) (table 2). The distribution of discharge diagnoses that previously has been associated with C. pneumoniae is shown in table 2. Statistical significance could not be reached for any particular cardio-

vascular disease, presumably in part because of the low number of events for each cardiovascular disease. However, the reduced risk of admission in users of macrolides compared with penicillins seems not to be associated with any particular cardiovascular disease. The relative risk (RR) of hospitalization for cardiovascular disease was dependent on the time interval after prescription of the macrolides. Thus, during the first 3 months, there was benefit from macrolide therapy compared with penicillin therapy (RR, 0.48; 95% confidence interval [CI], 0.27–0.88; P p .02), but this effect was not statistically significant after the first 3 months of follow-up (figure 1). We found interaction between the effect of macrolides and the following variables, which were therefore stratified: no previous hospitalization due to cardiovascular disease, age, and sex. The RRs during the first 3 months of follow-up for each of these strata are shown in table 3, and, as shown in figure 2, the same time-dependent pattern for the RR seen for the total group was found for each of the 3 strata. Thus, during the first

Figure 1. Relative risks (RRs) and 95% confidence intervals (bars) in 3 time intervals after treatment with macrolides versus penicillins, as determined by Cox models with time-dependent covariates.

1628

Østergaard et al.

Table 3. Relative risks for hospitalization for cardiovascular disease during first 3 months after treatment with macrolides or penicillins.

Risk factor Overall Penicillins Macrolides Stratum Previous cardiovascular disease No (primary prophylaxis) Penicillins Macrolides Yes (secondary prophylaxis) Penicillins Macrolides Age at treatment, years ⭓60 Penicillins Macrolides !60 Penicillins Macrolides Sex Male Penicillins Macrolides Female Penicillins Macrolides

Patients with event during first 3 months, no.

Total no. of patients

RR

95% CI

146 12

3827 634

1.00 0.48

Reference 0.27–0.88

0.39

0.14–1.12

58 4

2955 488 0.52

0.25–1.09

88 8

872 146 0.39

0.17–0.89

95 6

2004 338 0.64

0.27–1.51

51 6

1823 296 0.35

0.14–0.88

84 5

1626 272 0.67

0.30–1.46

62 7

2201 362

NOTE. Calculations were based on Cox models in different strata. RRs are shown for use of macrolide compared with penicillin as reference group. CI, confidence interval; RR, relative risk.

3 months of follow-up, the effect of macrolide treatment was more pronounced for patients with no previous history of cardiovascular disease (figure 2A), patients ⭓60 years old (figure 2B), and men (figure 2C).

Discussion Use of macrolides was associated with a subsequent reduced risk for hospitalization for a cardiovascular disease, compared with use of penicillins. The effect lasted for 3 months only and was associated with being male, being ⭓60 years old, and not having had a previous cardiovascular disease. The main strengths of our study are the uniformly organized Danish health care system, which allows a population-based design that avoids selection bias introduced by differential patient recruitment, as in studies from referral centers; the ability to adjust for the hospital discharge history of cardiovascular disease; and the completeness of follow-up. Selection bias is unlikely in a design that relies on computerized linkage of registries with almost complete coverage for the study area. Patients treated with penicillins were chosen as the comparator because this group is a more appropriate reference group than, for example, the background population. This is so because a group treated with penicillins is more similar with regard to comorbidity (e.g., lung disease) and underlying risk of cardio-

JID 2001;183 (1 June)

vascular disease than are groups not being treated with antibiotics. For instance, there is an increased risk of having a cardiovascular event after having respiratory tract symptoms [17], and although we do not know the indications for prescribing the antibiotics, we know from previous studies that respiratory tract infection is the major indication for prescribing macrolides and penicillins in Denmark. Thus, if a healthy control group had been chosen instead of a group of penicillin users, the effect of macrolides would have been overestimated. It is well known that hospital discharge diagnoses vary in quality, and we relied on the coding of the hospital discharge diagnoses by hospital doctors at the time of discharge [18]. The validity of a diagnosis of acute myocardial infarction has previously been found to be high [19], whereas it was lower for a diagnosis of hypertension [18]. This potential nondifferential classification tends to give conservative estimates of the risk. A strategy of analyses considering all cardiovascular diagnoses as outcomes was chosen because there is some misclassification of the discharge diagnosis. Any bias of this strategy is toward the null hypothesis, and our estimates are therefore conservative. In other words, the evidence for any association would be stronger if a difference can be shown for the whole group of cardiovascular diagnoses instead of just specific diagnoses. Although there was a trend of the effect of macrolides being found for all disease categories, we were not able to show a statistically significant difference in risk of hospitalization for any particular category of diagnosis. This may, however, be explained by the low number of events in each of the groups. It is well known that compliance is not complete for antibiotics, and we did not have information on dose or duration of the antibiotics. This may introduce bias. However, any noncompliance, low compliance, and misclassification of the exposure will bias the study toward the null hypothesis and thus underestimate the real effect. As seen from our data, macrolide prescription was associated with COPD, which, in turn, is related to smoking, which is a risk factor for cardiovascular disease. That more patients with COPD were treated with macrolides than with penicillins also tends to underestimate the observed effect of macrolides in comparison with penicillins. The lower risk of hospitalization for cardiovascular disease in macrolide users was transient, and we were not able to show a beneficial effect of macrolides 3 months after the prescription. The finding of a dependency between the time of macrolide use and the admission for a cardiovascular disease strengthens the evidence for an association between macrolides and reduced risk of cardiovascular disease. In addition, the finding of a short-term effect is in accordance with the findings of Gurfinkel et al. [12], who found an effect of roxithromycin therapy after 30 days, but not after 180 days, when given to patients with an acute cardiovascular disease (secondary prophylaxis) [13]. Meier et al. [15] showed that patients developing a myocardial infarction were less likely than patients not developing myocardial infarction to have been treated with quinolones and

JID 2001;183 (1 June)

Macrolides and Risk of Hospitalization

1629

1 year before the myocardial infarction. These findings again support our findings of only a short-term effect of macrolides. We did not investigate the effect of quinolones, tetracyclines, or cephalosporins, because these drugs are not subsidized by the Danish government, and thus data regarding these drugs are not reported to the register. The use of these drugs in Denmark is, however, negligible, and it is not likely that data on the use of these drugs would have changed the results. The mechanisms behind the beneficial effect of macrolides on cardiovascular diseases are not known. Some mechanisms that have been postulated include a stabilizing effect on asymptomatic atherosclerotic plaques that otherwise might have ruptured and led to subsequent thrombosis. Such stabilization may be due to an antichlamydial effect, whereby the growth of this organism is inhibited. This is supported by the fact that the antichlamydial drugs tetracycline and quinolones may also have a beneficial effect [15]. However, a direct anti-inflammatory effect on diseased vessels is also a possibility, because some macrolides have an anti-inflammatory effect [20], which is in part due to a direct inhibitory action on neutrophils and several cytokines [21]. Because of limitations of comparative population-based cohort studies, the data obtained in this study do not justify a change in current prescription guidelines. However, the data provide a unique background for designing blinded, randomized, and controlled clinical trials assessing the effect of macrolides on cardiovascular events in men ⭓60 years old with no previous cardiovascular disease. Such studies should also include a careful evaluation of adverse events as well as an evaluation of the potential environmental impact that the prophylactic use of macrolides may have on the development of antimicrobial resistance. Acknowledgments Figure 2. Relative risks (RRs) in 3 time intervals after treatment with macrolides versus penicillins, as determined by Cox models with time-dependent covariates for patients stratified according to previous cardiovascular disease versus no previous cardiovascular disease (top), age 50–59 years versus 60–69 years (center), and women versus men (bottom).

We gratefully acknowledge the assistance of Sygesikringen, Aarhus County, and the Department of Research and Development, Aarhus County.

References

tetracyclines, but they did not find any association between acute myocardial infarction and macrolide therapy. However, Meier et al. [15] assessed the risk over a period of 3 years, and an effect of short duration might therefore have been overlooked. In addition, they controlled only for age, sex, general practice attended, and calendar time, although the regression analysis was adjusted for smoking status and body mass index. Both may explain the difference from the results of our study. In another meticulous case-control study by Jackson et al. [16], no relationship was found between first-time myocardial infarction and use of erythromycin, tetracycline, or doxycycline, even when assessment was restricted to use of the antibiotics

1. Kuulasmaa K, Tunstall-Pedoe H, Dobson A, et al. Estimation of contribution of changes in classic risk factors to trends in coronary-event rates across the WHO MONICA Project populations. Lancet 2000; 355:675–87. 2. Abdelmouttaleb I, Danchin N, Ilardo C, et al. C-reactive protein and coronary artery disease: additional evidence of the implication of an inflammatory process in acute coronary syndromes. Am Heart J 1999;137:346–51. 3. Anderson JL, Carlquist JF, Muhlestein JB, Horne BD, Elmer SP. Evaluation of C-reactive protein, an inflammatory marker, and infectious serology as risk factors for coronary artery disease and myocardial infarction. J Am Coll Cardiol 1998; 32:35–41. 4. Arbustini E, Morbini P, Bello BD, Prati F, Specchia G. From plaque biology to clinical setting. Am Heart J 1999; 138:55–60. 5. Moazed TC, Campbell LA, Rosenfeld ME, Grayston JT, Kuo CC. Chlamydia pneumoniae infection accelerates the progression of atherosclerosis in apolipoprotein E–deficient mice. J Infect Dis 1999; 180:238–41.

1630

Østergaard et al.

6. Campbell LA, Kuo CC, Grayston JT. Chlamydia pneumoniae and cardiovascular disease. Emerg Infect Dis 1998; 4:571–9. 7. Grayston JT, Kuo CC, Campbell LA, Benditt EP. Chlamydia pneumoniae, strain TWAR and atherosclerosis. Eur Heart J 1993; 14(suppl K):66–71. 8. Saikku P. Chlamydia pneumoniae in atherosclerosis. J Intern Med 2000; 247: 391–6. 9. Cook PJ, Honeybourne D, Lip GY, Beevers DG, Wise R, Davies P. Chlamydia pneumoniae antibody titers are significantly associated with acute stroke and transient cerebral ischemia: the West Birmingham Stroke Project. Stroke 1998; 29:404–10. 10. Lindholt JS, Juul S, Vammen S, Lind I, Fasting H, Henneberg EW. Immunoglobulin A antibodies against Chlamydia pneumoniae are associated with expansion of abdominal aortic aneurysm. Br J Surg 1999; 86:634–8. 11. Gupta S, Leatham EW, Carrington D, Mendall MA, Kaski JC, Camm AJ. Elevated Chlamydia pneumoniae antibodies, cardiovascular events, and azithromycin in male survivors of myocardial infarction. Circulation 1997;96: 404–7. 12. Gurfinkel E, Bozovich G, Daroca A, Beck E, Mautner B. Randomised trial of roxithromycin in non–Q-wave coronary syndromes: ROXIS Pilot Study. ROXIS Study Group. Lancet 1997; 350:404–7. 13. Gurfinkel E, Bozovich G, Beck E, Testa E, Livellara B, Mautner B. Treatment with the antibiotic roxithromycin in patients with acute non–Q-wave coronary syndromes. The final report of the ROXIS Study. Eur Heart J 1999;20: 121–7.

JID 2001;183 (1 June)

14. Anderson JL, Muhlestein JB, Carlquist J, et al. Randomized secondary prevention trial of azithromycin in patients with coronary artery disease and serological evidence for Chlamydia pneumoniae infection: the Azithromycin in Coronary Artery Disease: Elimination of Myocardial Infection with Chlamydia (ACADEMIC) study. Circulation 1999; 99:1540–7. 15. Meier CR, Derby LE, Jick SS, Vasilakis C, Jick H. Antibiotics and risk of subsequent first-time acute myocardial infarction. JAMA 1999;281:427–31. 16. Jackson LA, Smith NL, Heckbert SR, Grayston JT, Siscovick DS, Psaty BM. Lack of association between first myocardial infarction and past use of erythromycin, tetracycline, or doxycycline. Emerg Infect Dis 1999;5:281–4. 17. Jousilahti P, Vartiainen E, Tuomilehto J, Puska P. Symptoms of chronic bronchitis and the risk of coronary disease. Lancet 1996; 348:567–72. 18. Nielsen HW, Tuchsen F, Jensen MV. Validity of the diagnosis essential hypertension in the National Patient Registry. Ugeskr Laeger 1996;158:163–7. 19. Madsen M, Balling H, Eriksen LS. The validity of the diagnosis of acute myocardial infarction in 2 registries: the Heart Registry compared to the National Patient Registry. Ugeskr Laeger 1990; 152:308–14. 20. Marchioli R, di Pasquale A, Marfisi RM, Tognoni G. Chronic infections and coronary heart disease. The GISSI-Prevenzione Investigators. Lancet 1997; 350:1028–9. 21. Labro MT. Anti-inflammatory activity of macrolides: a new therapeutic potential? J Antimicrob Chemother 1998; 41(suppl B):37–46.