• Chlamydia trachomatis
• pelvic inflammatory disease
• randomised control trial
• salpingitis
• mass screening

Andersen et al1 report findings from 9 years' follow-up in one county in Denmark of 30 000 men and women aged 21–23 years, of whom 9000 (4000 women and 5000 men) were randomly selected in 1997 to receive a mailed invitation to provide a self-obtained specimen for chlamydia screening. That study has exploited the rare opportunity offered by Danish health registers to link longitudinal health records for a cohort of young people, which included men (a rarity). Using an intention-to-treat analysis, the authors conclude that a single postal invite for chlamydia testing, with a repeat offer to those found positive, did not reduce the long-term risk of reproductive complications.1 What do these data add to the evidence base concerning the effectiveness of chlamydia screening? What lessons should inform further studies?

Previous randomised controlled trials (RCTs) of a single chlamydia screen, with 1-year follow-up for the incidence of pelvic inflammatory disease (PID), have noted the limitation of evaluating the impact of a single round of chlamydia screening, because the risk of chlamydia infection is ongoing,2 as should the intervention be.3 The study by Andersen et al1 with its long follow-up for ectopic pregnancy and infertility increases the probability of the acquisition of chlamydia infections after (or before) the intervention and accentuates this bias to a null finding for these outcomes.

That the authors found no statistically significant reduction in any reproductive complications must be considered in the light of the limited uptake of screening in either arm—which compromises the study's power. Only a quarter of the intervention arm participants actually returned a specimen for testing; approximately one tenth of female subjects (intervention and control arm) received a usual care screening test during the same period. In the previous RCTs of chlamydia screening Scholes et al4 tested approximately 64% of the intervention arm population; Ostergard et al5 tested all those in the denominator of the experimental arm, as did Oakeshott et al.2 The study set out to detect a lowering in PID by approximately one-third, which is approximately the maximum reduction in PID possible via chlamydia control, given that the frequency of previous chlamydia exposure among PID cases has been reported as approximately 30%.6 Assuming the rate of PID among non-participants to be the same as among the control arm, a 30% reduction in the PID rate among participants (to 0.51×0.70=0.36) would have resulted in an observed PID rate among the intervention arm of 0.47 ((0.51×0.71)+(0.36×0.29)) and a hazard ratio (HR) of 1.09 (0.51/0.47), ie, a slightly smaller effect than was observed and which was not significant (HR 1.12, 95% CI 0.70 to 1.79). With the limited coverage in the study, therefore, the efficacy of a single screen would have to be far greater than can be reasonably expected for a significant benefit to have been seen. As the authors acknowledge, the observed incidence of PID in the cohort was lower than assumed in the study size determination, which contributed to this likelihood of a null finding. Others have also had this problem with power for PID outcomes.2 For the other, rarer, complications significant reductions could have been observed only if effect sizes were also quite unreasonably high. A further bias to a null finding is likely to have been introduced by measurement error due to non-specificity in measures of reproductive complications, which for PID at least this study seems even more prone to than others, given the identification of cases by outpatient prescriptions for doxycycline.

This is the first time that the benefits of chlamydia screening in men have been studied in an RCT. Unfortunately, however, a significant finding was not possible given the low uptake. Assuming the rate of epididymitis among non-participants to be the same as among the control arm, the observed (non-significant) HR of 1.25 would have resulted from a reduction of over 95% in epididymitis among participants.

Another familiar problem is apparent in the subanalysis that looked only at those in the intervention arm by whether they had accepted testing or not (ie, non-randomised analysis), and found no difference in the frequency of outcomes by whether individuals participated in screening or not; such as analysis is valid only if the two subpopulations had comparable risks of infection and the authors themselves indicate that non-participants had a lower risk of infection.7 Therefore, that rates of complications were comparable may in fact reflect a benefit for those screened.

The study demonstrates that a single offering of screening by mail, in addition to usual care, did not appreciably reduce either PID or any of the associated long-term complications within the population studied. The small proportion of the at-risk population accepting this offer is sufficient reason for the intervention to have ‘failed’. Noting, in addition, the frequency of routine care tests in both arms, the long follow-up and the non-specificity of some measurements, this study does not answer the question of whether chlamydia screening reduces reproductive complications.

Three fundamental questions regarding chlamydia screening need to be answered. How much benefit does chlamydia screening provide to those being tested, in terms of reducing PID and other complications? How much does screening benefit the population by reducing the prevalence of infection? and how do we implement chlamydia screening to deliver in the best way whatever benefits screening has to offer?

In relation to the latter question, Andersen et al1 have demonstrated the inadequacy of the chlamydia screening approach used in this study. Given the limited coverage achieved by this implementation, it should come as no surprise that their intervention had no discernable benefit. It is proving extremely challenging to conduct methodologically sound clinical trials that address questions about the benefits of screening, as the experiences of Andersen et al1 and others demonstrate. One thing is sure—the two big problems of power and bias should be anticipated and accommodated. Incorporating a method for achieving long-term follow-up does not overcome these problems. Perhaps we will get more enlightening results from innovative approaches to analyses of observational data from situations in which the implementation of screening has been relatively successful. We still have a lot to learn.