Papers

Chlamydia pneumoniae IgG titres and coronary heart disease: prospective study and meta-analysisCommentary: Adjustment for potential confounders may have been taken too far

BMJ 2000; 321 doi: https://doi.org/10.1136/bmj.321.7255.208 (Published 22 July 2000) Cite this as: BMJ 2000;321:208

Abstract

Objective: To examine the association between coronary heart disease and serum markers of chronic Chlamydia pneumoniae infection.

Design: “Nested” case-control analysis in a prospective cohort study and an updated meta-analysis of previous relevant studies.

Setting: General practices in 18 towns in Britain.

Participants: Of the 5661 men aged 40–59 who provided blood samples during 1978-80, 496 men who died from coronary heart disease or had non-fatal myocardial infarction and 989 men who had not developed coronary heart disease by 1996 were included.

Main outcome measures: IgG serum antibodies to C pneumoniae in baseline samples; details of fatal and non-fatal coronary heart disease from medical records and death certificates.

Results: 200 (40%) of the 496 men with coronary heart disease were in the top third of C pneumoniae titres compared with 329 (33%) of the 989 controls. The corresponding odds ratio for coronary heart disease was 1.66 (95% confidence interval 1.25 to 2.21), which fell to 1.22 (0.82 to 1.82) after adjustment for smoking and indicators of socioeconomic status. No strong associations were observed between C pneumoniae IgG titres and blood lipid concentrations, blood pressure, or plasma homocysteine concentration. In aggregate, the present study and 14 other prospective studies of C pneumoniae IgG titres included 3169 cases, yielding a combined odds ratio of 1.15 (0.97 to 1.36), with no significant heterogeneity among the separate studies (χ2=10.5, df=14; P>0.1).

Conclusion: This study, together with a meta-analysis of previous prospective studies, reliably excludes the existence of any strong association between C pneumoniae IgG titres and incident coronary heart disease. Further studies are required, however, to confirm or refute any modest association that may exist, particularly at younger ages.

Chlamydia pneumoniae IgG titres and coronary heart disease: prospective study and meta-analysis

  1. John Danesh, clinical research fellowa,
  2. Peter Whincup, professorb,
  3. Mary Walker, senior lecturerc,
  4. Lucy Lennon, research assistantc,
  5. Andrew Thomson, computer programmerc,
  6. Paul Appleby, statisticiand,
  7. Yuk-ki Wong, research registrare,
  8. Martine Bernardes-Silva, research scientiste,
  9. Michael Ward, professore
  1. a Clinical Trial Service Unit and Epidemiological Studies Unit, University of Oxford, Oxford OX2 6HE
  2. b Department of Public Health Sciences, St George's Hospital Medical School, London SW17 ORE
  3. c Department of Population Sciences and Primary Care, Royal Free and University College London Medical School, London NW3 2PF
  4. d Imperial Cancer Research Fund Cancer Epidemiology Unit, Oxford OX2 6HE
  5. e Departments of Cardiology and Molecular Microbiology, University of Southampton, Southampton SO16 6YD
  6. University of Wales College of Medicine, Cardiff CF4 4XN
  1. Correspondence to: J Danesh
  • Accepted 6 June 2000

Abstract

Objective: To examine the association between coronary heart disease and serum markers of chronic Chlamydia pneumoniae infection.

Design: “Nested” case-control analysis in a prospective cohort study and an updated meta-analysis of previous relevant studies.

Setting: General practices in 18 towns in Britain.

Participants: Of the 5661 men aged 40–59 who provided blood samples during 1978-80, 496 men who died from coronary heart disease or had non-fatal myocardial infarction and 989 men who had not developed coronary heart disease by 1996 were included.

Main outcome measures: IgG serum antibodies to C pneumoniae in baseline samples; details of fatal and non-fatal coronary heart disease from medical records and death certificates.

Results: 200 (40%) of the 496 men with coronary heart disease were in the top third of C pneumoniae titres compared with 329 (33%) of the 989 controls. The corresponding odds ratio for coronary heart disease was 1.66 (95% confidence interval 1.25 to 2.21), which fell to 1.22 (0.82 to 1.82) after adjustment for smoking and indicators of socioeconomic status. No strong associations were observed between C pneumoniae IgG titres and blood lipid concentrations, blood pressure, or plasma homocysteine concentration. In aggregate, the present study and 14 other prospective studies of C pneumoniae IgG titres included 3169 cases, yielding a combined odds ratio of 1.15 (0.97 to 1.36), with no significant heterogeneity among the separate studies (χ2=10.5, df=14; P>0.1).

Conclusion: This study, together with a meta-analysis of previous prospective studies, reliably excludes the existence of any strong association between C pneumoniae IgG titres and incident coronary heart disease. Further studies are required, however, to confirm or refute any modest association that may exist, particularly at younger ages.

Editorial by Koenig

Introduction

A study published in 1988 proposed that Chlamydia pneumoniae infection was an avoidable cause of coronary heart disease.1 Since then, systematic reviews have identified several dozen additional studies of C pneumoniae markers and vascular disease.24 Although some reports have suggested twofold or larger odds ratios for coronary heart disease in people with markers of chronic C pneumoniae infection, these studies have generally been small, retrospective, or liable to biases.24 We report a study of 496 cases of coronary heart disease and 989 controls “nested” in a prospective cohort of British men monitored for 16 years. We also conducted an updated meta-analysis of other prospective studies to place our results in context.

Participants and methods

Cases and controls

During 1978-80, 7735 men aged 40–59 (response rate 78%) were randomly selected from general practice registers in each of 24 British towns and entered in the British Regional Heart Study.5 Nurses administered epidemiological questionnaires, made physical measurements, and recorded an electrocardiogram. Non-fasting venous blood samples were collected in 5661 men in 18 of the towns and stored at −20°C for subsequent analysis. Further questionnaires were posted after five years (98% response among survivors) and 12 years of follow up (90% response among survivors) that asked about car ownership and childhood social circumstances (father's social class and childhood household amenities) respectively. All men have been monitored since entry for death from all causes and for cardiovascular morbidity, with a loss to follow up of less than 1%.5 Cases in our study were men who had fatal coronary events or non-fatal myocardial infarction between the beginning of follow up and December 1995 and who had a stored serum sample available for analysis. Fatal cases of coronary heart disease were ascertained through NHS central registers on the basis of a death certificate with International Classification of Disease (ICD-9) codes 410-414. Non-fatal myocardial infarction was based on reports from general practitioners, supplemented by evidence from general practice records, meeting World Health Organization criteria.5 Of 507 potential cases (223 deaths from coronary heart disease and 284 non-fatal myocardial infarctions), 496 had C pneumoniae measurements available. A total of 1026 controls, who were “frequency matched” to cases on town of residence and age in five year bands, were randomly selected from among men who had survived to the end of the study period without a myocardial infarction; 989 of these controls had C pneumoniae measurements available.

Laboratory methods

Laboratory workers unaware of the disease status of the participants analysed blood samples for C pneumoniae using whole organism antigen and time resolved fluorimetry.6 The assay showed good agreement with microimmunofluorescence in a validation study of 480 people (intra-assay and interassay coefficients of variation were 4% and 8%). Serum lipid concentration, albumin concentration, leucocyte count, and packed cell volume were measured with standard assays, and C reactive protein and serum amyloid A concentrations were determined by sensitive enzyme immunoassays.5

Statistical methods and systematic review

We compared case and control groups using unmatched stratified logistic regression fitted by unconditional maximum likelihood (Stata Corporation, College Station, Texas, USA). Adjusted analyses included the following explanatory variables: age; cigarette smoking habit (never, former, current); daily cigarette consumption; non-fasting blood concentrations of total cholesterol, high density lipoprotein cholesterol, and triglyceride; markers of current social class (registrar general's 1980 classification with a separate category for armed forces); housing tenure (owner, private rent, council rent); marital status; current car ownership; father's occupation (manual, non-manual); and childhood social circumstances (father's occupation, family car ownership, bathroom in house, hot water tap in house, bedroom sharing). We prespecified analysis of C pneumoniae IgG titres by thirds of the values in controls—that is, the top third was defined as seropositive and the bottom third as seronegative. Previous systematic reviews suggested the need for adjustments for smoking and indicators of socioeconomic status in adulthood and childhood to help reduce any residual confounding in studies of coronary heart disease and persistent infective agents, and some previous studies of C pneumoniae infection and coronary heart disease have reported adjustments for indicators of social class both in adult life and in childhood (see Discussion).25 We therefore prespecified that odds ratios would be reported both with and without such adjustments. For analyses of C pneumoniae IgG titres with a variety of known and suspected risk factors, emphasis was given to differences greater than 2.6 standard deviations (P≈0.01) to make some allowance for multiple comparisons.

Methods to identify studies for an updated meta-analysis of prospective studies of coronary heart disease and C pneumoniae IgG titres or IgA titres published before May 2000 have been described. 2 3 Cases were compared with controls only within the same studies to avoid potential biases (see BMJ's website for full details).

Results

As would be expected, we found highly significant differences between cases and controls with respect to various known vascular risk factors such as smoking, obesity, blood pressure, and blood lipid concentration (table 1). C pneumoniae IgG titres were significantly associated with age and leucocyte count (table 2). Among cases, titres were also associated with smoking status, although this was attenuated by adjustment for indicators of socioeconomic status (data not shown). No significant associations were observed between C pneumoniae IgG titres and various indicators of socioeconomic status and values of blood lipids, blood pressure, plasma homocysteine, C reactive protein, serum amyloid A protein, albumin, and packed cell volume.

Table 1.

Baseline characteristics of men with coronary heart disease and of controls matched for age, sex, and town. Values are mean (SD) unless stated otherwise

View this table:
Table 2.

Comparisons of levels of risk factors and other characteristics in controls by thirds of C pneumoniae IgG titres. Values are means (SD) unless stated otherwise

View this table:
Table 3.

Odds of coronary heart disease in men who were IgG seropositive for C pneumoniae relative to those who were seronegative

View this table:

Discussion

Previous retrospective serological studies have suggested that chronic C pneumoniae infection is an important cause of coronary heart disease in the general population,1 but this hypothesis has not been adequately tested in larger prospective studies. In comparison with retrospective studies, prospective studies should reduce selection biases, minimise any influence of disease itself on the factor being investigated, and generally include better adjustment for potential confounding factors—for example, only about half of the retrospective studies of C pneumoniae and coronary heart disease published before 1998 reported adjustment for cigarette smoking. 2 3 Our prospective, community based study, with 16 years of mean follow up, included more coronary heart disease cases than all but one previous study.7 We found an odds ratio for coronary heart disease of 1.59 (1.17 to 2.16) in men with high baseline C pneumoniae IgG titres after adjusting for smoking and markers of adult social class and an odds ratio of 1.22 (0.82 to 1.82) after additional adjustment for markers of childhood social class. The partially adjusted and fully adjusted odds ratios were statistically compatible with each other (because of relatively wide, overlapping confidence intervals) and were also compatible with either a moderately positive association or no association at all. We therefore conducted a systematic review of previous relevant studies of C pneumoniae and coronary heart disease to assess further any association.

Meta-analysis

Including the present study, we identified 15 prospective studies of C pneumoniae IgG titres and coronary heart disease up to May 2000.720 The studies included a total of 3169 cases of non-fatal myocardial infarction or death from coronary heart disease; the weighted mean age at baseline was 56 years with a weighted mean follow up of 10 years. All adjusted for smoking and some other classic risk factors, but only seven (including our study) reported adjustment for markers of adult socioeconomic status712 and only two for markers of childhood social class.9 Ten of the studies used microimmunofluorescence assays (seven studies used ≥1:64 as a cut-off titre for seropositivity, 8 9 11 12 15 18 20 one study used ≥1:128,13 one used ≥1:32,17 and one did not specify the cut off16), and five used other methods (two used enzyme linked immunoassays, 10 19 two used time resolved fluorimetry,7 and one did not specify the exact method14). Despite these differences, there was no significant heterogeneity among the 15 studies (χ2=10.5, df=14; P>0.1), and a combined analysis yielded an odds ratio of 1.15 (95% confidence interval 0.97 to1.36) for coronary heart disease (figure).

Subsidiary analyses yielded a combined odds ratio of 1.15 (0.84 to 1.57) in the eight studies (822 cases) that did not adjust for indicators of adult socioeconomic status1320 and an odds ratio of 1.16 (0.95 to 1.41) in the 13 studies (2395 cases) that did not adjust for indicators of childhood socioeconomic status 7 8 1020 Moreover, if the partially adjusted odds ratio for our study was used in the meta-analysis of all 15 studies (instead of the fully adjusted odds ratio), the combined odds ratio was still 1.19 (0.99 to 1.41). Similar results were obtained in the 10 studies (1521 cases) that used microimmunofluorescence assays (combined odds ratio 1.11 (0.87 to 1.42)), 8 9 1113 1520 and in the nine studies (1816 cases) that reported risk in relation to C pneumoniae IgA titres (combined odds ratio of 1.13 (0.90 to 1.41)). 7 9 10 13 15 1719 21 Again, there was no significant heterogeneity among studies in any of these subsidiary analyses.

Figure1

Prospective studies of Chlamydia pneumoniae IgG titres and coronary heart disease. Black squares indicate the odds ratio in each study, with the square size proportional to the number of cases and horizontal lines representing 99% confidence intervals. The combined odds ratio and its 95% confidence interval are indicated by a diamond. Degree of adjustment for possible confounders is denoted as +++ for age, sex, smoking, and some other classic vascular risk factors, and ++++ for these plus markers of socioeconomic status (or studies done in socially homogeneous groups). Some studies did not provide separate results for non-fatal myocardial infarction or death from coronary heart disease (instead reporting coronary heart disease results combined with those for stroke 12 14 or angina 19 20). These studies, however, comprised only about 10% of the total cases in this meta-analysis, and most of the cases in these studies had major coronary events

Our meta-analysis therefore reliably excludes any strong association between C pneumoniae IgG titres (or IgA titres) and coronary heart disease. Existing data are, however, insufficient to assess reliably any odds ratios weaker than about 1.5. Moreover, published studies have not corrected for possible underestimation due to fluctuations in serum antibody titres within individuals over time (as C pneumoniae seropositivity may disappear and recur2). Further studies are therefore required to confirm or refute any more modest association that may exist, particularly at younger ages, when associations may be stronger than at older ages.22

Implications for randomised trials of antibiotic treatments

Three randomised placebo controlled trials of antichlamydial treatments have reported on coronary heart disease events, 20 2325 each including a few hundred patients with a history of coronary heart disease. All three trials put patients on brief courses of oral macrolides or macrolide derivatives (antibiotics with antichlamydial and, perhaps, anti-inflammatory effects). The first recorded only eight coronary events and yielded a non-significant result (although the investigators reported a fourfold reduction in coronary heart disease on the basis of an inappropriate non-randomised comparison).20 The second published trial recorded 22 events and also gave non-significant results after six months of follow up.23 Hence, in retrospect, this study's earlier claim of a fourfold reduction in coronary heart disease at one month was largely or wholly due to chance or selective reporting of an interim analysis.24 The third trial (also reported in an interim analysis) recorded 16 events and had non-significant results.25

Several trials are now in progress with larger sample sizes, lengthier antibiotic treatment periods, and more prolonged follow up (table 4). But even these trials may be able to provide only limited information about any effects of antichlamydial treatment in coronary heart disease. Our meta-analysis of prospective seroepidemiological studies creates considerable doubt about the existence of any independent association between persistent C pneumoniae infection and coronary heart disease. The combined odds ratio was only 1.15 (0.97 to 1.36), which is much weaker than the combined weighted odds ratio for atherosclerosis of 20 (15 to 32) obtained from pathology based studies that have assessed human arterial specimens for endovascular markers of C pneumoniae (DNA, antigens, elementary bodies, or viable organisms).24

Table 4.

Some larger randomised trials currently in progress of antichlamydial strategies for prevention of coronary heart disease

View this table:

What might account for this 20-fold discrepancy? Pathology based studies have been retrospective (thereby creating uncertainty about whether local C pneumoniae infection is a cause or consequence of atheroma), whereas the prospective serological studies assessed evidence of infection several years before the diagnosis of coronary heart disease. Most pathology based studies have also been prone to selection biases and lacked any adjustment for possible confounders such as age, sex, and smoking,24 but this could not plausibly explain much of the 20-fold difference. It is also unclear to what extent the discrepancy can be accounted for by the different definitions of vascular disease (atheroma versus major coronary events) and the different markers of infection (endovascular markers such as DNA and antigens versus circulating antibody titres) used in these different sets of studies.

Such epidemiological uncertainties have implications for the numbers needed in clinical trials. If the 20-fold odds ratio reported in the pathology based studies mainly reflected a causal effect that was largely reversible (rather than some artefact of confounding or reverse association), then antichlamydial treatments might be expected to reduce coronary event rates substantially. To confirm or refute such large effects should require trials only of similar size to the three previously reported trials (although, depending on the speed at which the risk reversed, follow up might need to be much longer). If, however, the prospective serological studies provide a more reliable guide to the likely strength of any association between C pneumoniae and coronary heart disease, the trials would need to be much larger than those previously conducted or currently in progress. Even if there is a 10% excess risk due to C pneumoniae that is fully reversible by antibiotics, none of the existing trials would be large enough to confirm or refute its existence (the largest current trials cannot detect reductions in coronary events that are less than 25%; table 4).

Conclusion

Since 1997, the number of cases of coronary heart disease in prospective studies of C pneumoniae IgG titres has increased 10-fold to over 3000 cases, with our study being one of the largest. Unlike previous retrospective studies, more reliable prospective data indicate that C pneumoniae IgG titres are not strongly associated with coronary heart disease.

What is already known on this topic

Persistent infection with Chlamydia pneumoniae has been suggested to be an avoidable cause of coronary heart disease

Most previous studies on the topic have been small and prone to biases

What this study adds

Baseline C pneumoniae IgG concentrations were not strongly associated with major coronary events or with classic or suspected risk factors

Updated meta-analysis of relevant prospective studies gave a combined odds ratio for heart disease of only about 1.1, which was not significant

Substantial uncertainty exists about any independent association between Chlamydia pneumoniae infection and heart disease

Acknowledgments

We thank H Refsum and P Ueland for the homocysteine assays; J R Gallimore and M B Pepys for the C reactive protein and serum amyloid A protein assays; J Atherton and C Hawkey for H pylori assays; and J John for valuable help. Professor G Shaper established the British Regional Heart Study.

Contributors: The study was designed and conducted by the British Regional Heart Study collaborative group. All the named authors contributed to the analysis and interpretation of these data and to writing the paper. JD and PW are the guarantors.

Footnotes

  • Funding The British Regional Heart Study is a British Heart Foundation research group and also receives support from the Department of Health. JD was supported by a Merton College fellowship and a Frohlich award.

  • Competing interests None declared.

References

  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.

Commentary: Adjustment for potential confounders may have been taken too far

  1. Robert West, reader in epidemiology
  1. a Clinical Trial Service Unit and Epidemiological Studies Unit, University of Oxford, Oxford OX2 6HE
  2. b Department of Public Health Sciences, St George's Hospital Medical School, London SW17 ORE
  3. c Department of Population Sciences and Primary Care, Royal Free and University College London Medical School, London NW3 2PF
  4. d Imperial Cancer Research Fund Cancer Epidemiology Unit, Oxford OX2 6HE
  5. e Departments of Cardiology and Molecular Microbiology, University of Southampton, Southampton SO16 6YD
  6. University of Wales College of Medicine, Cardiff CF4 4XN

    Danesh and colleagues' study, based on the 24 British towns study, shows a weaker association between markers for chronic or past infection with Chlamydia pneumoniae and incident ischaemic heart disease than some earlier reports. The odds ratio comparing highest with lowest tertiles of IgG titres was 1.7 and highly significant before adjustment. It remained highly significant (1.6) after adjustment for age, town, smoking, and social class but reduced to 1.2 and became non-significant after adjustment for childhood social class. The authors include this last estimate in their statistical overview of 15 prospective studies.

    These findings provide an opportunity to review the rationale underlying adjustment in analysis. The principles of adjustment (or standardisation) are essentially the same as those of matching when selecting controls in case control studies: to match for factors or characteristics known or previously shown to be causally associated with the disease.1 With the wisdom of hindsight, when chains of causality have been shown experimentally, it is not difficult to recognise and distinguish causal confounders, associated markers, and unrelated factors or characteristics, but identification is harder during analysis and interpretation of observational studies. It is possible to overmatch when selecting controls and to overadjust in analysis.

    A previous example, adjustment for birth weight in perinatal mortality, illustrates the point. We recommend adjustment for birth weight when comparing the performance of health authorities and hospitals (and doctors) so that like is compared with like and to compensate for the higher expected perinatal mortality in a hospital that admits more than its share of low birthweight babies.2 However, in a study of the aetiology of perinatal mortality it would be inappropriate to adjust for birth weight, since low birth weight (light for dates or premature delivery) could be an indicator of the same disease process as the perinatal mortality itself.3 Adjusting for birth weight in an aetiological study of perinatal disease could be “throwing the baby out with the bathwater.”

    The analysis by Danesh and colleagues of the 24 towns study adjusts for both social class and childhood social class. Social class is a powerful epidemiological tool, and this United Kingdom measure is the envy of epidemiologists in many other countries. It is strongly associated with many diseases and is a useful proxy for exposure (occupational or environmental hazards, poverty, malnourishment, etc) and possibly for constitutional characteristics, but it is not a measure of exposure itself. Thus adjustment for social class, when we know little of the mechanism by which social class effects the disease under study, may be overadjustment.

    In this study it is adjustment for childhood social class that removes the significance of the association (the “dose response,” one of Bradford Hill's criteria for causality, although weakened, remains (see table 3)). In the 24 towns study childhood social class was estimated from a composite of measures, including bathroom in house. Like the adult classification, childhood social class is a potential proxy for many possible exposures, including nutrition and hygiene, or for various constitutional characteristics. It might be argued that poor hygiene and nutrition in childhood (low childhood social class) allows early infection by C pneumoniae to become chronic, as indicated by raised IgG titres. It is perhaps appropriate to recall that, although tuberculosis was infectious among poorer people in urban slums, doctors and nurses in tuberculosis sanatoriums remained remarkably free of the disease.4 Associations between early antecedents, including childhood social class, and adult ischaemic heart disease are recognised,5 but there is poor understanding of the mechanisms by which childhood social class influences age at which ischaemic heart disease is manifest.

    Until mechanisms are better understood, it seems wise to report associations without, as well as with, adjustment for childhood social class. Indeed, it could be argued that these associations should be summarised without, as well as with, adjustment for adult social class—if C pneumoniae might be one mechanism by which social class affects age of onset of clinical heart disease. These reservations notwithstanding, the statistical overview does suggest that any association between markers of chronic infection by this organism and fatal or non-fatal myocardial infarction is not strong.

    Footnotes

    • Embedded Image Further details of the meta-analysis are available on the BMJ's website

    References

    1. 1.
    2. 2.
    3. 3.
    4. 4.
    5. 5.
    View Abstract