Re: Risk of venous thromboembolism from use of oral contraceptives containing different progestogens and oestrogen doses: Danish cohort study, 2001-9
(Continued from part 1)
Validity of data
It is important to realise, that determining a reliable risk estimate of a certain pill demands valid exposure as well as valid end point data. Every time an exposure is misclassified or an end point is wrong, we will underestimate the real risk of VTE in current users of OCs.
Concerning the exposure data, it is difficult to imagine a more precise data source than a prescription registry. Not only does a prescription registry give precise information about the date a woman receives package of pills, but it also provides valid information about which type of pill is prescribed. As compared with retrospective studies, in which women are asked about use of OCs months or years back in time, or prospective studies, in which a certain exposure at a certain time is not up-dated, a prescription registry is by far the most reliable exposure data available.
The small uncertainty as to exactly which day a woman commences her use as compared with the date she buys the OC is minor as compared to misclassifications made from other data sources. And more important: Any such misclassification would underestimate both the risk of VTE with current use and the rate ratio between OCs with DRSP versus OCs with LNG.
Concerning the outcome data (VTE), the primary data source was discharge diagnoses from hospital wards. The fact that Denmark collects discharge diagnoses in a National registry does not make these diagnoses less reliable than diagnoses of VTE in other studies.
Moreover, we validated all 4,246 events by cross linking them with subsequent anticoagulation therapy, and restricted our analyses to these anticoagulation-confirmed events. This validation (not surprisingly) elevated the rate ratio estimates between users of OCs with DRSP versus LNG from 1.64 in our 2009 publication to 2.1 (1.6-2.8) in the new analysis (2).
Next, in a random sample of 200 women with a discharge diagnosis of VTE, through chart review we found a positive predictive value of confirmed VTE according to the registry data to be 99%. But again it is important to realise, that even when no record of anticoagulation therapy is present in the prescription registry, this is not the same as a false VTE diagnosis.
First, about 10% of women get the anticoagulation therapy for free from the departments, and they are therefore not included in the prescription registry. In addition, 5-10 per cent are diagnosed based upon clinical symptomes of VTE despite lack of confirmation in the paraclinical investigations, possibly because the clot was too small, or alternatively, that the clot might have dissolved spontaneously. In both cases no treatment may have been warranted, but the woman was found and told to have probably had a VTE. If we add the 10% ward treated women and those with clinical symptoms but without treatment, we approach the 88% achieved in our 2002 study, which was based upon information from departments and from questionnaires, and in which the women themselves confirmed their diagnosis.
Also the fact that some real events were not included in the group of confirmed events will not change the rate ratio estimates, and only marginally affect the confidence intervals. The fact that the proportion of confirmed events among users of OCs with DRSP (73.7%) and in users of OCs with LNG (73.2% and 74.2% for combined and cyclic products, respectively) was similar, strongly contradict a differential referral or a differential diagnosis of VTE among these two groups of OC users. And the number of confirmed and non-confirmed events was the same in the EMA report as in the BMJ publication. But JD & SS compared figures from the period 2001-2005 in the EMA report with figures from another period 2001-2009 in the BMJ publication.
In conclusion our study had well defined validation criteria for VTE, applied to all groups of users of OCs. The rate ratio estimate for confirmed events; 2.1 (1.7-2.7) was slightly higher than for the non confirmed events; 1.8 (1.2-2.6), indicating that the more valid the diagnoses are, the larger rate ratio estimates of VTE we find between users of OCs with DRSP versus LNG-users.
And last but not least. If anything, the inclusion of uncertain events will tend to decrease the risk estimates as demonstrated by the rate ratios described above, and cannot provide a basis to discount positive findings in our study.
Considering how carefully JD & SS have evaluated the smallest methodological details in our studies, it is surprising that they never made any reflection as to which direction their many proposals of bias would move the risk estimates. In fact, any of the alleged biases suggested by JD & SS if anything all tend to underestimate the rate ratios between users of OCs with DRSP versus OCs with LNG.
On the authors behalf
Continues in part 3
2. Lidegaard O, Nielsen LH, Skovlund CV, Skjeldestad FE, Lokkegaard E. Risk of venous thromboembolism from use of oral contraceptives containing different progestogens and oestrogen doses: Danish cohort study 2001-9. BMJ 2011; 343: d6423.
Competing interests: The primary investigator received no salary for his work with this study, the EMA report or the manuscript. OLi has within the last three years received honorariums for speeches on pharmacoepidemiological issues, including fees from Bayer Pharma Denmark and Novo Nordisk, and will be an expert witness for plaintiffs in a legal US case in 2011-2; FES received compensation for his work in the steering committee of the European Medicines Agency report. The other authors had nothing to declare.