Abstract
Sir: Robert Van Howe published in the December 2007 issue a meta-analysis of data on genital ulcerative disease (GUD) and sexually transmitted urethritis in relation to male circumcision status. 1 We have reservations about this paper, in particular regarding the citing of source data, but as well, poorly defined search terms and inclusion criteria, failure to include appropriate studies and absence of adjustment for confounding factors.
Previous statistical analyses by Van Howe relating to circumcision have been criticized. 2–4 In his latest meta-analysis, there is application of incorrect methodology and the data presented differ from the original published data that appear in the literature he cites (Table 1). The differences are in many cases significant and difficult to explain.
Comparison of data from Van Howe's paper that shows association of various sexually transmitted infections with lack of circumcision with the actual data in four of the studies he cites
*First author and (in brackets) reference number in Van Howe's paper
†Shown is the odds ratio (OR) or adjusted odds ratio (AOR) (with 95% confidence intervals in brackets), exactly as stated in the original study, namely association with lack of circumcision (Cook; Parker) or with circumcision (Laumann; Dave). NSU = non-specific urethritis
‡Van Howe's paper shows all OR as for association with lack of circumcision
§Association with circumcision, as was stated in original paper
**Laumann et al. do not state an OR for chlamydia, nor do they provide sufficient data to calculate an OR, yet Van Howe provides one
Van Howe's identification of relevant papers appears insufficient. For example, four studies addressing the association of male circumcision and chancroid identified in an earlier meta-analysis 5 (references 38 and 40–42 therein) were excluded.
Another error is that Van Howe appears to present crude measures of association between circumcision and sexually transmitted infection (STI), even though adjusted figures are available for many studies, and are more appropriate because they partially control for confounding, such as by religion or sexual behaviour. For example, Diseker et al. (Van Howe's ref. 151) reported adjusted ORs of 1.3 and 1.6 for the association of gonorrhoea and lack of circumcision in cross-sectional and cohort analyses, respectively, but Van Howe cites the crude ORs of 1.09 and 1.24.
The study population used in each paper also needs careful consideration. Around half of the studies included in Van Howe's Table 1 took place among STI clinic attenders. 1 STI populations have inherent selection biases, 5 as Van Howe recognizes. He does not, however, note that the comparison group are likely to be men with another STI, which might itself be associated with circumcision status. This would result in an under-estimation of any true protective effect of circumcision. Similarly, Van Howe's adjustment of raw data from two studies (his refs 32 and 361) would have introduced selection bias for the same reason. For example, in Wilson's study (Van Howe's ref 36), lack of circumcision was associated with increased risk for every STI that was included. Van Howe rejected Wilson's original control group and instead chose to compare men with a particular STI with all men in the STI population.
Van Howe uses the keyword ‘circumcision’, which will fail to identify those papers that examined circumcision as one of the many potential risk factors for STI, but that did not include ‘circumcision’ as a keyword, or in the title or abstract. Further, this search strategy could introduce ascertainment bias, since the keyword ‘circumcision’ is more likely to be used for studies that find associations.
Finally, Van Howe's definition of GUD is not fully clear. Did the analyses of GUD include studies with specific aetiologies of GUD (e.g. chancroid, syphilis and genital herpes)? If so, why are the studies of chancroid not included with GUD? We are left to conclude that by ‘GUD’ he actually means GUD of unknown aetiology, which is a relatively weak definition.
To illustrate the effect of Van Howe's errors, we consider non-specific urethritis (NSU). Table 2 shows all 10 NSU studies used by Van Howe and lists the source data alongside his cited data. We then provide the outcome of performing a meta-analysis on the correct source data (Table 2). Our analysis thereby shows that NSU is not associated with circumcision status in these data, contrary to Van Howe's assertion.
Non-specific urethritis: all errors and a correct meta-analysis showing no association with lack of circumcision
*First author and (in brackets) reference number in Van Howe's paper
†No OR or CI provided in the original study; figures presented above under ‘Original OR’ are calculated from published frequency data
‡Aynaud et al. reported an OR of 2.3 (1.08–5.11) for all bacterial urethral infections. The figure presented above includes all bacterial urethral infections except Neisseria gonorrhoea. The term ‘non-specific urethritis’ is slightly problematic; many of Van Howe's sources refer to ‘non-gonococcal urethritis’, which may include urethritis caused by chlamydia, for example. In Aynaud et al., several specific diagnoses are in fact given, including chlamydia, but for compatibility with sources reporting ‘non-gonococcal urethritis’ we assume here that ‘non-specific’ has the same meaning as ‘non-gonococcal’, and this was Van Howe's intent, since he states ‘…addressed non-specific or non-gonococcal urethritis (NGU)…’
§‘Original OR' is inverse of figures in original study
**‘Original OR' was calculated using the control group in the original paper
††Each row represents an analysis of the 10 papers presented above. Where no corrections are specified, we have used Van Howe's OR (A) for all the 10 papers. Where corrections to figures are identified, we have substituted the original OR (B) for these papers
In conclusion, we find Van Howe's meta-analysis to be based on incorrect data and to be flawed methodologically. We show as an example that at least one of his results is incorrect as a result.