Infertility due to obstruction of the fallopian tubes is one of the main severe and lasting consequences of infection with the bacterial sexually transmitted infections gonorrhoea and chlamydia. The disutility and costs associated with infertility and its treatment provide a substantial contribution to the assumed costs of chlamydia in the health economic analyses justifying chlamydia control programmes.1 However, infertility is difficult to define and measure.2 A lower bound on infertility is provided by medical diagnosis, which requires both that patients seek help and that diagnosis is available. More cases are added if we include all those seeking care for infertility or who believe they are infertile.2 Such measures might represent those for whom infertility is a problem, but they will greatly underestimate the true rates of infertility. An alternative is to look at birth histories and periods without any pregnancies, but such methods only readily apply to populations where contraception is not practised.3 A final method is to estimate the risks of infertility associated with different causes and calculate expected rates.4 This approach is generally adopted in health economic analyses5 where the proportion of infections leading to disease and the proportion of these that result in infertility is estimated. If these analyses are to be valid we need reliable estimates of the risk of infertility that follows from chlamydia infection.

With developments in chlamydia control, including the use of more sensitive nucleic amplification tests, and screening of those asymptomatic, the effectiveness and benefits associated with the programmes have been challenged.6 In justifying intensive programmes it would be useful to have reliable estimates of the risks of salpingitis, infertility and ectopic pregnancy in those who have and have not had chlamydia. To assess the extent to which good estimates exist, Wallace and colleagues (see page 171) carried out a systematic review to see how well the excess risk of infertility associated with chlamydia could be estimated.7 They found very little evidence that met their inclusion criteria, which raises two important questions. First, should we believe that chlamydia is an important cause of infertility and, second, how well can we estimate the risk so we can generate improved cost-effectiveness analyses. The key to these questions is the clearly defined exclusion criteria used in the published review. Retrospective case control studies were excluded because these provide odds ratios rather than risk ratios and studies using chlamydia antibody tests were excluded because of their poor sensitivity. However, these methods are the very ones that, along with plausible biological mechanisms, convincingly showed the relationship between chlamydia and infertility.8 Additionally, in the absence of prospective cohort studies, such retrospective studies may still provide us with useful results.

The first point to note is that the odds ratio for exposure in diseased versus non-diseased equals the odds ratio for disease in the exposed versus the unexposed, and that this in turn approximates the risk ratio when the prevalence of disease is low. This relationship relies on rates of disease being low and is independent of the prevalence of exposure. To rehearse the justification for these statements consider a study of a single disease and a single exposure. If we denote disease in the exposed as De and in the unexposed Du, and not diseased in the exposed as Ne and unexposed Nu, then the exposure odds ratio in those with disease compared to those without disease is (De/Du)/(Ne/Nu). This can be rearranged to (De/Ne)/(Du/Nu): the odds ratio of disease in the exposed versus the unexposed. This is similar to the risk ratio, which additionally includes those with the disease De and Du in the denominators: (De/(De+Ne))/(Du/(Du+Nu)). It is clear from this that when the De and Du are small they contribute little to the denominators and consequently the odds ratio and the risk ratio are similar. The error in calculating the risk ratio from the odds ratio can be estimated if we know the prevalence of disease. Thus, if infertility is low then retrospective case control studies will provide a reasonable estimate of the risk associated with chlamydia. Unfortunately, many of the pivotal studies showing a significant association between infertility and chlamydia antibodies came from populations where infertility was extremely widespread and where errors in the estimate of risk ratios will be great.8

In prospective studies, nucleic acid amplification tests provide us with a reliable measure of chlamydia infection. However, in retrospective case control studies, where infertility defines a case, we require a measure of a history of chlamydia infection, not of current infection. Tests for a specific immune response provide such a history. The problem is sensitivity of the tests—a poorly performing test will lead to misclassification bias, and in the case of low sensitivity many of those that should be in the exposed category will be believed unexposed. Such misclassification bias tends to the null model—that is, a relationship will be harder to detect when there really is one. Furthermore, if we know the sensitivity of the test through comparison with a gold standard then we can estimate the actual risk based on that observed using the flawed test.

One concern in the analysis of case control studies is whether confounding variables have been appropriately measured and controlled for. This is especially the case given sexually transmitted infections such as gonorrhoea, syphilis and HIV leading to reduced pregnancy rates. Such a concern also applies for prospective cohort studies. Another area that requires attention is the definition of chlamydia infection associated with a particular risk of infertility. With more sensitive tests, less severe infection associated with fewer complications and sequalea may be detected.

See linked article, page 171

From the review of Wallace and colleagues,7 from experiences in screening programmes9 and because programmes are based on cost-effectiveness analyses,1 5 it is clear that more work is required before we can fully justify and appropriately design chlamydia control programmes. However, in this work the case control methods and antibody test that have served us well in the past should not be ignored and perhaps studies where infertility is rare should be a priority.