Reader Comments

Post a new comment on this article

Statistical Flaws Mar analysis

Posted by Hanabi on 21 Jan 2010 at 07:42 GMT

There are serious statistical flaws in this paper which are not evidence based. This is all too common in these types of computer simulations where the subject is complex, there are very many uncontrollable and unknown variables and too many simplifying assumptions have been made. To their credit the researchers mention two of them, unknown changes in risky behavior due to perceived immunity and the lack of inclusion of the cost of adverse events. The authors also fail to consider the results of the successful HIV vaccine trial, and the complete lack of evidence that the results of the African trials have any bearing on the epidemic in the developed world.

First, their own data shows a lack of correlation between circumcision and HIV rate among Black and Hispanic heterosexual men. The lifetime risk to Black men was 6.23% yet 73% of Black men were circumcised, while the lifetime risk to Hispanics was only 2.88% with a circumcision rate of 42%. The correlation coefficient is 3% indicating no causal relationship between circumcision and HIV risk. If the lifetime reduction in HIV risk were as high as 60%, we would expect at least some correlation between circumcision and lifetime HIV risk.

A similar analysis conducted on data from English speaking countries shows that circumcision in English speaking countries is strongly correlated to increased risk of HIV infection (R2= 93%). Data from the United States, Canada, Great Britain and Australia show that circumcision increases a man’s lifetime risk by 0.75%. Unlike data used in previous publications comparing the United States to Brazil and India, two developing countries, this data is relevant to developed countries with large medical infrastructures. While a 93% correlation does not indicate a necessary causality, it is a VERY strong indicator of a common causal connection, such as increased risky behavior among circumcised men. The authors themselves give evidence for such changes citing an increase in sexual encounters in the Auvert RCT (the fact that this made no change in the HIV rate in the intervention indicates a non-causal relationship between circumcision and HIV infection rate in the three studies). Until this data can be adequately explained any call for infant circumcision is not only premature but incredibly dangerous.

Second, the cost analysis did not, and in fact could not, include the costs of adverse events because, unlike vaccines, the FDA and CDC do not monitor them for cosmetic surgery. If the authors are to be taken seriously, they must include this in the cost. Surgical mishap is not common but is very expensive. For example, a recent jury award of $2.5 million was given to the parents of a boy who had the glans of his penis amputated during a circumcision. This did not happen in some 3rd world backwater but in Atlanta GA (on the doorstep of the CDC) in a high quality medical facility. Every year in the US it has been estimated that 10 or so such mishaps occur and two children die from botched circumcisions. The CDC, and therefore the authors, have no idea of the true costs because CDC does not monitor this. Further, in two hospital studies, circumcision was found to increase the risk of Methicillin Resistant Staphylococcus Aureus (MRSA) by 5 to 12 times during the neonatal period. The cost and risk of treating infection, let alone MRSA, have not been figured into this analysis. There have now been confirmed cases of Vancomycin resistant strains of Staphylococcus Aureus making the costs and risks of surgery potentially even higher.

The authors failed to include the fact that as patents expire, the cost of HIV treatment will decrease, while the cost of circumcisions, currently underestimated, will only grow. This is a serious exclusion in the most sensitive data element in their analysis.

The authors wildly overestimate the effectiveness of circumcision at 60%. While the African studies did show this, it was only after excluding early HIV conversion that such high numbers were obtained. This is a flaw in the understanding of the study results which actually showed a 44% reduction in an Intention To Treat analysis. It is widely accepted that using such a Modified Intension To Treat analysis overestimates the effectiveness of the intervention. Further, the study was terminated early further artificially increasing the estimation of effectiveness, an effect that even the authors of some of the papers have acknowledged. Finally, the studies could not be blinded, the men knew if they had been circumcised. The well documented Hawthorn Effect tends to skew the results toward a more efficacious intervention. When these data are included in the analysis it is likely that the real effectiveness is below 40% making circumcision not cost effective even at the least expensive assumption used for the cost of circumcision.

There was no analysis of alternatives. For whatever reason, neither the CDC nor the authors have ever compiled an analysis of alternatives to ascertain the most effective least expensive intervention in the fight against HIV. In other words, they do not know if what they are recommending has the greatest likelihood of success at least cost. Almost assuredly adult circumcision, targeted toward men engaged in risky behavior would be more cost effective. Since the average age of infection was established at 34, there is no reason to rush to neonatal circumcision programs. There are 16 years between the age of majority in the US (18) and the average age of infection. During this time an adult can make his own decision regarding his lifestyle and the costs and benefits. If the authors wish to make such interventions available to the men who need it most without a one size fits all solution, adult circumcision can be paid for by Medicaid or insurance. Since only 6 men in 10,000 people are at risk, 100 adult circumcisions would be less expensive than about 5,000 infant circumcision (assuming adults are 10X more expensive) with better effectiveness because of the targeted approach. Unfortunately no analysis was done or presented so the true magnitude of the savings is speculation at this time.

The authors completely failed to consider the results from the Thailand vaccine trial. This trail showed a 31% reduction among 16,402 healthy Thailanders. Since the vaccine is effective against HIV in all people, not just heterosexual men, it is far more cost effective than circumcision could ever be. My analysis shows that a 30% effective vaccine would be 6 times more effective than a 50% effective circumcision because of the more general protection afforded. The cost of the vaccine would be less, Gardasil for example costs about $300 for the 3 shot regimen. Yet, the NIH and the FDA have come out and said that the vaccine does not meet the standard of effectiveness to be rolled out. One wonders how the authors of this paper reconcile this when the vaccine, even at 30% effectiveness, would be many times more effective and less expensive than mass circumcision campaigns.

The authors assume no change in behavior in circumcised men. This flies in the face of several studies that have conclusively shown that perceived protection from antiretroviral therapy has changed behavior in gay men. Since HIV affects so few heterosexual men 0.06% (6 men in 10,000 people) of the general population (CDC data from 0.6% HIVrate X 74% men * 14% heterosexual – 2009 published data) a small change in behavior would completely overwhelm any effect in this tiny fraction of men.

The authors fail to consider any potential for increased risk to women. The Wawer study showed a nearly statistically significant increase in risk to women (82% probable increased risk to women by Fisher Exact). The sample was too small to establish significance (much like the Celebrex studies were too small to establish significance for harm). Yet, even a small increase in risk to women from circumcised men would again overwhelm any benefit obtained since 83% of HIV cases in women are from high risk heterosexual contact. This analysis would be in keeping with the type of speculative analysis performed for gay men which is also non-statistically significant.

Finally, the authors fail to perform any trend analysis on circumcision in the US. Since the rate of circumcision has been falling since 1990 there should be an increase in the percentage of intact men becoming sexually active. If the author’s assumptions about the effectiveness of circumcision are correct, there should be an increase in the HIV rate among HIV positive men. Yet, the latest finding from the CDC show a drop in the HIV rate among heterosexual men down from 16% (the number used by the authors) in 2008 (estimate from 2006 data) to 14% (estimate from 2007 data). Clearly, this is a trend consistent with the data from English speaking countries and in direct contradiction to the conclusions of this overly simple analysis.

In the fight against HIV/AIDS there can be no compromise for clear analysis and no room for interventions that are not KNOWN to be effective. Anything else would be unethical and simply dangerous.

Scot Anderson

No competing interests declared.

RE: Statistical Flaws Mar analysis

Hanabi replied to Hanabi on 22 Jan 2010 at 04:03 GMT


There was a typo in my response to this article. The regression analysis shows a 0.0075% lifetime reduction in the general population for every 1% fewer men circumcised. The correlation is still 93%. Translating this result shows that if the US would increase male circumcision by 13% it would likely incur an 0.1% increase in HIV rate for the general population. That is the current HIV rate in the US would go from 0.6% to 0.7%.

I forgot to mention in the analysis of alternatives that the three randomized controlled studies from Africa shows data for efficacy for targeted adult circumcision. That is what the studies did, they asked for adult volunteers who where at risk for HIV. A generalized study of infant circumcision has not been made so regression analysis as I have shown, is the best indicator. Further, this shows a likely reduction of lifetime efficacy reducing the number to well below 40%, the minimum the analysis considers.
Finally, there was no discussion of ethics. In general it is unethical to perform surgery or any other medical intervention unless there is immediate risk or deformity. Infants are not at risk from sexually acquired HIV and it is wild speculation that they might be in 34 years. This makes the intervention, no matter how cost effective, unethical in the developed world. This is doubly true considering that there is time for the man to decide for himself with his own biases, beliefs and lifestyle.

S. Anderson

No competing interests declared.