Intensity and Duration of Oral Anticoagulation in Antiphospholipid Antibody Associated Thrombosis

To the Editor

We were delighted to see the report by the McMaster group and Universities of Cincinnati and Toronto¹, which by using decision analysis showed the same conclusion that we came to²: that patients with antiphospholipid syndrome presenting with thrombosis have a better outcome if longterm warfarin (target international normalized ratio (INR) 3.0-4.0) is used. We also agree that the use of aspirin does not offer clear additional benefit.

We were, however, disappointed that our work and that of others was incorrectly quoted. For example (p. 491), the authors state that "Others have recommended more vigorous anticoagulation, targeting an INR between 3.0 and 4.0" and cite 2 articles by Ginsberg, et al³ and Schulman, et al⁴ that do not support their argument. Similarly (p.499) they state that "In some studies with close monitoring of the anticoagulant therapy, no patient taking warfarin (INR 2-3) had recurrent thrombotic events" and cite our paper², which does not support their view.

Despite these shortcomings we wholeheartedly concur with the conclusions of the authors. We agree that the lack of prospective studies in the management of antiphospholipid antibody associated thrombosis is a major drawback. We feel, however, that recruitment into randomized prospective trials comparing target INR of 2.0-3.0 and 3.0-4.0 will be difficult in view of resistance of both clinicians and patients due to the data suggesting an INR of 3.0-4.0 is preferable.

BEVERLY J. HUNT, MD, FRCP, FRCPath; MUNTHER A. KHAMASHTA, MD, MRCP, PhD, Lupus Unit, St. Thomas' Hospital, London SE1 7EH, UK.

Silicone Gel Breast Implants

To the Editor:

In my recent letter¹ I challenged Brown, et al's finding of an association between extracapsular ruptured breast implants and fibromyalgia (FM)². I argued that they inappropriately merged patients with intact implants and those with intracapsular ruptures and used this group as the comparison for patients with extracapsular rupture. This consequently led to a spurious finding. From an epidemiologic and clinical perspective, I find Drs. Brown and Pennello's justification¹ of their analytic approach illogical, lacking merit, and requiring further thoughtful comment.

They justified their pooling by stating: "if extracapsular silicone was the sole cause of FM, then ruptures without extracapsular silicone could be combined with intact implants particularly since there was no association with rupture (i.e., intracapsular and extracapsular) when compared to all others (i.e., intact and indeterminate)" — italics mine. The authors' justification seems at least partially grounded in a hypothesis that extracapsular silicone is "the sole cause of FM," but their data, when analyzed properly, do not support the hypothesis, and thus should be rejected (but are not) by the investigators. Also, discovering that FM has "no association with rupture when compared to all others" clearly fails, as explained below, to justify their decision when all the data across implant rupture status groups are examined.

The relationship between FM and implant status shows that the protection of intracapsular rupture against FM is stronger (odds ratio = 0.50; the reciprocal is an odds ratio of 2.0) than the increased risk with extracapsular rupture (odds ratio = 1.88) when each rupture type is individually compared against intact implants¹. (Note that neither of these odds ratios are statistically significant.) These numbers suggest that those with extracapsular rupture, not those with intracapsular rupture, are more like those with intact devices. If the authors are intent on combining the intact group with one of the rupture groups, the data suggest the more appropriate combination would be intact implants and extracapsular ruptures. Of course, doing this leads to a different study conclusion — that intracapsular ruptures protect against FM! Combining the data this way shows a statistically significant protection against FM by intracapsular rupture (odds ratio = 0.37, 95% confidence interval 0.17-0.77). Clearly, the appropriate findings to report should be based on an analysis that does not collapse any rupture status categories, because the FM results are so different by rupture status. I believe this analytic and reporting approach is what most clinicians and epidemiologists would consider correct.

If I understand the authors, they found a statistically significant difference (p = 0.003) among intact, intracapsular, and extracapsular devices — presumably for FM¹. This finding further supports that pooling rupture groups is incorrect. This situation highlights a well known epidemiologic problem of when to "lump" or "split" categories. Generally, epidemiologic principles dictate against lumping categories when evidence suggests they are substantially different. For example, the risk of a disease (e.g., as measured by an odds ratio) for males and for females generally should not be pooled into one risk measure when substantial gender differences exist — such combining hides gender differences. The authors tell us they found statistically significant differences in the association of FM across the three implant status categories. Consequently, this is precisely the type of dataset where collapsing should not be done because it gives misleading findings.

The authors correctly state that I found in their data a non-statistically significant odds ratio of 1.88 for FM and extracapsular rupture. They use this finding to continue to justify their conclusions by stating: "it [being the 1.88] is still highly suggestive of an association between FM and extracapsular rupture." What troubles me is their selective use of information. I also state¹ in the next sentence after I report the 1.88 odds ratio, that I found an odds ratio of 0.50 between FM and intracapsular rupture. Why did they not use this finding to say that it is highly suggestive that intracapsular rupture protects against FM? How can they embrace and highlight the one association yet completely ignore the other? Without a doubt, this second association undermines their study conclusions. From a biological and clinical perspective, how can intracapsular rupture on the one hand protect a woman against FM, and an extracapsular rupture, on the other, increase a woman's risk, and all this found in the same dataset? The researchers claim their findings indicate and suggest an association. With all due respect to my FDA colleagues — How? To me, an impartial and balanced view of their data suggests otherwise.

In light of what the authors' data show with straightforward analysis, I'm left, unfortunately, wondering about their objectivity. Did they have preconceived biases about ruptured implants and FM that led to the way they analyzed and presented their data? Did the authors have strong preconceived notions (even unconsciously) that needed confirmation? Where is the empirical method of allowing the results of proper data analysis to modify incorrect ideas/hypotheses? If this concern is unwarranted, then I'm left asking myself, did the authors overlook (inadvertently or not) basic epidemiologic analysis principles and practices? Whatever the reason, in the end, the rational conclusion from examining the investigators' information across all implant status categories is that no consistent pattern between FM and breast implant rupture exists in this study.

STEVEN J. BOWLIN, DO, MPH, PhD, Dow Corning Corporation, Midland, Michigan, USA.

Dr. Bowlin is an employee of Dow Corning Corporation.