We thank Steve Hammett, Jim Moore and Gary Glissman for their interest in our paper and for their comments. They all express the opinion, to a varying degree, that the content should be detrimental for patients with KS and 47,XYY syndrome. We understand this point, but we disagree, and we can iterate that as physicians caring for hundreds of patients with sex chromosome disorders through many years, being involved in research and the generation of information material for patients and relatives, that we apparently see the plight of our patients differently. We firmly believe that clinicians caring for these patients will be better suited to do so with these data at hand, and likewise that other health care professionals will benefit from comprehensive and holistic knowledge of the syndrome. Before submission of our paper we had a general discussion of the results of our study within the group of authors and with other colleagues and everybody agreed that it would be unethical not to publish the data! We have seen a similar situation with patients with schizophrenia which also have a high rate of criminality. There is a large body of evidence to support this high rate of criminality, and this knowledge has led to the development of tools to estimate risk of criminality, as well as an increased focus on improved medicinal treatment of those with an estimated high risk of future criminality 1. In other words, the knowledge of an increased rate of criminality has been used in a constructive way to improve patient care. We and others have through the very recent years expanded the clinical phenotype of these syndromes, and one of our reviewers, Nicole Tartaglia, has also within the last year contributed with valuable information concerning the social deficits present in many persons with KS and 47,XYY 2, 3 in clinical studies, while we have presented compelling data from epidemiological studies showing that the socio-economic situation of males with these syndromes is inferior in comparison with the background population 4, 5. We are confident that all these studies with different approaches will help better define and characterize these two syndromes. This will in due course lead to better treatment and hopefully also to earlier diagnosis. Suppression of data such as we have presented, we believe, will be detrimental for these two groups of patients in the long run. Stigmatization of groups of patients, a concern voiced by our commenters, is of course always a concern, and is also a concern for us. However, we do not think and believe that our paper will lead to stigmatization of patients. Steve Hammett mentions that a link between gay communities and sexual offenses to children have been propelled by the religious right (in the USA, we presume), and states that he does not hope that our data will lead to a similar link. The gay community, we believe, was never harmed by openness, rather it seems that openness has been and is the way to increased understanding and tolerance of gay people. In the same way, we see our study as a step towards greater understanding and treatment of people with KS and 47,XYY. We see our data as the foundation for future and better studies of how to alleviate some of the inherent social, psychiatric and psychological problems related to the syndromes. We also believe that openness, presentation of factual concerns based on sound data and the free dissemination of data as the foundation for scientific progress. Steve Hammett and Jim Moore have concerns about our approach of not matching for socio-economic status, as was also voiced by our reviewers. This kind of thinking is based on misconceptions about which kinds of answers epidemiology can offer. Our approach in this epidemiological study is methodologically sound and we can only reiterate that matching is not necessarily the right thing to do when the factors (socio-economy parameters) that one matches on are pivotal in understanding the phenomenon of interest, in this case - "criminality". There is a considerable body of research on matching in case-control studies, showing that it is dangerous and sometimes outright wrong to match on factors that, like in the current setting, may be causally involved in the chain of events leading to increased criminality. This can lead to overmatching 6 - "matching on factors that are affected by the study exposure or disease (i.e being KS or control) is almost never warranted and is potentially capable of biasing study results beyond any hope of repair. It is therefore crucial to understand the nature of such overmatching and why it needs to be avoided". And since we have previously shown that having KS or 47,XYY is related to poorer socio-economic outcome 4, 5, one cannot match on such factors since we do not know the precise causal relationship between these factors, the diseases (i.e. KS or 47,XYY) and criminality. Such an approach may even lead to "an irreparable form of selection bias" 6. Thus, there is nothing "contentious" or "problematic" about our data. They are generated in a society with a much lower crime rate than in the USA (70 murders per year per 5.6 million people) and with a much higher detection rate of crimes committed. Steve Hammett also voices concern in relation to the fact that only 25% of the expected number of males with Klinefelter syndrome is present in our analysis. Although we do discuss this issue at length, being the ones that originally presented the issue some years ago 7, we would like to point to the fact that we actually present a sensitivity analysis as Supplementary data. Here, we present two different scenarios. One, where we assume that the rate of criminality among the non-diagnosed would be half that observed among the cases, and another analysis where we assume that the rate of criminality would be similar to that of the controls. On the background of these analyses we show that it is highly likely that the crime rate would remain significantly increased among an entirely unbiased population of both KS and 47,XYY with complete diagnosis of all cases. However, it is clear that only diagnosis of all cases with KS and 47,XYY would fully elucidate this important question. Gary Glissman rightly points to neurobiology as an area where we need new information in relation to KS and 47,XYY. We can only agree with this notion, and we believe that ongoing studies will help in deciphering the somewhat scattered information currently at hand in relation to changes observed on MRI of the brain, neurocognitive changes and genetics. Clearly, this is a very interesting area which poses hope for better understanding, treatment and coupling with endocrinology. Finally, we would like to add a comment on the nature of a study as ours. We have utilized existing registers and merged these in order to better delineate the course of rare syndromes 4, 5, 7-14. Epidemiology is a powerful tool in this context. Epidemiology offers interesting insight, poses new questions, and can present challenging dilemmas. But rarely epidemiological studies can offer solutions, new treatment strategies or prove the superiority of a new drug. Here, we need clinical studies, preferably randomized placebo-controlled clinical trials. A range of studies like ours should therefore serve as part of the background for future clinical studies. Such clinical studies, we are ourselves, engaged in.

Reference List

(1) Topiwala A, Fazel S. The pharmacological management of violence in schizophrenia: a structured review. Expert Rev Neurother 2011 January;11(1):53-63.

(2) Cordeiro L, Tartaglia N, Roeltgen D, Ross J. Social deficits in male children and adolescents with sex chromosome aneuploidy: a comparison of XXY, XYY, and XXYY syndromes. Res Dev Disabil 2012 July;33(4):1254-63.

(3) Ross JL, Roeltgen DP, Kushner H et al. Behavioral and social phenotypes in boys with 47,XYY syndrome or 47,XXY Klinefelter syndrome. Pediatrics 2012 April;129(4):769-78.

(4) Bojesen A, Stochholm K, Juul S, Gravholt CH. Socioeconomic trajectories affect mortality in Klinefelter syndrome. J Clin Endocrinol Metab 2011 May 11;96(7):2098-104.

(5) Stochholm K, Juul S, Gravholt CH. Socio-economic factors affect mortality in 47,XYY syndrome - a comparison with the background population and Klinefelter syndrome. Am J Med Genet 2012, in press.

(6) Rothman KJ, Greenland S, Lash TL. Design strategies to improve study accuracy. In: Rothman KJ, Greenland S, Lash TL, editors. Modern Epidemiology. 3rd ed. Philadelphia: Lippincott Williams & Wilkins; 2008. p. 168-82.

(7) Bojesen A, Juul S, Gravholt CH. Prenatal and postnatal prevalence of Klinefelter syndrome: a national registry study. J Clin Endocrinol Metab 2003 February;88(2):622-6.

(8) Bojesen A, Juul S, Birkebaek N, Gravholt CH. Increased mortality in Klinefelter syndrome. J Clin Endocrinol Metab 2004 August 1;89(8):3830- 4.

(9) Bojesen A, Juul S, Birkebaek NH, Gravholt CH. Morbidity in Klinefelter syndrome: a Danish register study based on hospital discharge diagnoses. J Clin Endocrinol Metab 2006 April;91(4):1254-60.

(10) Stochholm K, Bojesen A, Jensen AS, Juul S, Gravholt CH. Criminality in men with Klinefelter's syndrome and XYY syndrome: a cohort study. BMJ Open 2012 February 22;2(1):e000650.

(11) Stochholm K, Juul S, Juel K, Naeraa R.W., Gravholt CH. Prevalence, incidence, diagnostic delay, and mortality in Turner syndrome. J Clin Endocrinol Metab 2006 July 18;91:3897-902.

(12) Stochholm K, Juul S, Gravholt CH. Mortality and incidence in women with 47,XXX and variants. Am J Med Genet A 2010 February;152A(2):367 -72.

(13) Stochholm K, Juul S, Gravholt CH. Diagnosis and mortality in 47,XYY persons: a registry study. Orphanet J Rare Dis 2010 May 29;5:15.

(14) Stochholm K, Hjerrild B, Mortensen KH, Juul S, Frydenberg M, Gravholt CH. Socio-economic parameters and mortality in Turner syndrome. Eur J Endocrinol 2012;166(6):1013-9.

Conflict of Interest:

None declared

Christiaan Monden, lecturer (1), Jeroen Smits, associate professor (2)

1. University of Oxford, 2. Radboud University Nijmegen

Vaupel, Zhang and Van Raalte (VZ&V) have made an interesting contribution to the study of variation in length of life (or life disparity as they call it) on the basis of life table data [1]. A fascinating aspect of this literature is that the inequality measures that are used - such as the Gini coefficient, Standard Deviation, Coefficient of Variation, Interquartile Range, and also the e+ used by VZ&V - are highly correlated, not only amongst themselves, but also with life expectancy[2-4]. In fact, the correlation between life expectancy (LE) and life inequality (LI) is so high (often over -0.9[4]), that one might expect an increase in life expectancy to be almost always associated with lower inequality.

VZ&V also observe this in their data: the country with the highest e(0) in a given year often had the lowest inequality (e+) in that particular year. From this finding, VZ&V seem to conclude that countries that are leaders in life expectancy are also the most equal ones. However, they base this conclusion on a comparison of differences between countries within a specific year, which in our vision is not the most informative approach in this context.

This can be seen in Figure 1 -- an adapted version of a figure presented in our 2009 paper[4] --, in which e(0) is plotted against LI (measured by the GINI coefficient) for 4,690 life tables. The data are for males in 194 countries and include besides all the life tables used by VZ&V many additional life tables we collected from other sources[5]. The figures are for all ages (instead of 15+ as in our 2009 paper) to make them better comparable with VZ&V. We show outcomes of tables with an e(0) of over 40, because that range includes all life expectancy leaders of VZ&V.

Figure 1 can be viewed here - http://dx.doi.org/10.6084/m9.figshare.95456

Figure 1 contains three kinds of points. The light-colored background points show the association between LI and LE for all 4,690 country-year combinations. The black and green points reveal the positions of the life expectancy leaders from VZ&Vs paper (Table S4 of VZ&V; 169 leaders for each year from 1840-2008). The black points represent the 59 leaders who were true record breakers; they reached an e(0) level that had not been observed before (e.g. Sweden in 1898). The green points represent follow- up leaders; they were VZ&Vs best performers in a particular year, but at a level of e(0) that had already been reached in the past (e.g. Norway 1899- 1901).

Figure 1 makes clear that the countries that reached a certain e(0) first are not the most equal countries at that level of e(0). On the contrary, in almost all years the life expectancy leaders are found in the middle or upper part of the inequality distribution. In other words, compared to countries that reached a certain level of life expectancy later, the life expectancy leaders seem to perform only average (or worse) in terms of inequality.

One might observe that at a high level of life expectancy (say an e(0) of over 75) in Figure 1, the life expectancy leaders have relatively low inequality. Note however that the picture at those levels is not yet complete, as the mass of other countries has not yet arrived. Based on what we see at the lower e(0) levels in Figure 1, it seems likely that when the lagging countries reach e(0) of over 75, many of them will do so at a lower level of inequality compared to VZ&Vs leaders.

Reducing premature mortality generally leads to higher life expectancy and lower inequality in life spans. However, Figure 1 makes clear that reaching a high life expectancy earlier than other countries does not result in the lowest possible inequality at that life expectancy level. An interesting question that thus remains is why many countries that reach a certain level of life expectancy later can do so with lower inequality.

References

1. Vaupel JW, Zhang Z, Van Raalte A. Life expectancy and disparity: an international comparison of life table data. BMJ Open 2011;1:e000128 doi:10.1136/bmjopen-2011-000128.

2. Wilmoth JR, Horiuchi S. Rectangularization revisited: variability of age at death within human populations. Demography 1999;36:475-95.

3. Shkolnikov V, Andreev E, Begun AZ. Gini coefficient as a life table function. Computation from discrete data, decomposition of differences and empirical examples. Demogr Res 2003;8:305-58.

4. Smits J, Monden C. Length of life inequality around the globe. Soc SciMed 2009;68:1114-1123.

5. The data used for Table 1 are available at www.lengthoflife.org.

Conflict of Interest:

None declared

George T. Grossberg, MD, St. Louis University School of Medicine, St Louis, Missouri, USA

Yvonne Wirth, MD, PhD, Wirth Consulting, Stuttgart, Germany

Suzanne Hendrix, PhD, Pentara Corporation, Salt Lake City, Utah, USA

Michael Tocco, PhD, Forest Research Institute, Jersey City, New Jersey, USA

Stephen M. Graham, PhD, Forest Research Institute, Jersey City, New Jersey, USA

Dear Dr Groves:

After careful review of the article titled "Memantine and cholinesterase inhibitor combination therapy for Alzheimer's disease: a systematic review" (Farrimond et al, BMJ Open, 2012)[1], we have identified significant inaccuracies that we believe should be addressed.

Several important errors and inconsistencies occur in Figure 3, some of which are also repeated in Table 2 (Analysis 3, Function). First, the figure shows a numerical advantage for placebo over memantine extended- release (ER) in trial MD-50, which is incorrect: in that trial [2], as well as in trial MD-02 [3] and in the subset of patients from MD-12 with moderate AD [4], the memantine group demonstrated a smaller mean decline on the measure of function than the placebo group. The corrected Figure 3, which properly summarizes the data from these 3 trials, along with the authors' original version, can be found at the following site:

http://dx.doi.org/10.6084/m9.figshare.95823

As shown in the figure, for the outcome of function, Farrimond et al calculated a standard mean difference (SMD) [95%CI] of -0.04 [-0.21, 0.13] (P = 0.65), whereas, according to our calculations, the correct overall SMD [95%CI] is -0.11 [-0.21, 0.00] (P = 0.05). Therefore, there is a large discrepancy between their analysis and ours in terms of both the observed effect (-0.04 vs -0.11) and the accompanying P-value (0.65 vs 0.05).

In addition, our analysis of Cognition (Table 2, Analysis 1b), performed by applying the methodology outlined by the authors to the same published data that they used, resulted in an SMD [95%CI] of -0.23 [-0.47, 0.00] (P= 0.05, significantly in favor of memantine), rather than their published SMD of -0.16 [-0.54, 0.23] (P = 0.43). These corrected values reveal that the TA127 analysis would have indeed demonstrated significant cognitive benefits in favor of memantine/donepezil combination therapy had the data been pooled, contrary to the authors' assertion.

We also bring to your attention several statements pertaining to published literature or previously reported data. In the introduction, the authors refer to the Patel and Grossberg manuscript [5] as a "company- sponsored" review, which in fact was produced independently of industry support, as stated in the acknowledgments ("no sources of funding were used to assist in the preparation of this review"). Moreover, Table 1, which correctly cites the Forest Clinical Trial Registry website as the reference for the MD-50 study, incorrectly states that "the baseline characteristics of patients in this unpublished study are not given." These data are in fact reported in the summary clinical study report posted on the website, with the exception of the mean NPI score (Placebo + ChEI = 16.8; Memantine + ChEI = 17.2). In reference to the memantine ER formulation (28 mg) used in the MEM-MD-50 study, the authors state in the discussion section that "the dose of 28 mg memantine in this preparation was designed to be equivalent to 20 mg daily of the currently marketed preparation," which is incorrect: the new memantine formulation was designed to provide a 40% increase in the maximum daily dose with a much slower absorption rate (median Tmax of 12 h, compared with 3-7 h observed with the immediate-release formulation)[6].

The authors also speculate that "the trend for an adverse effect on ADL may account for the fact that [the MD-50 trial] data have not been published in peer review literature." This interpretation may have been influenced by their misperception that placebo treatment was superior to memantine for ADL in that trial (as exemplified by the error in Figure 3). In fact, the manuscript describing this trial has been submitted for publication. As reported previously [2], the trial demonstrated significant benefits in favor of memantine ER on both primary efficacy parameters (SIB and CIBIC-Plus) and two additional parameters (NPI and Verbal Fluency Test), as well as a numerical advantage in favor of memantine on the ADCS-ADL19.

Furthermore, the authors state that "the fact that clinical data have not been released from the 12-month trial, Lu1011 [2], is disturbing"; however, the trial manuscript, which includes the clinical data, was published --and indexed on PubMed and Embase-- in January, 2012 [7], two months before the manuscript by Farrimond et al was accepted for publication.

Finally, the authors state that "since the impact on clinical global impression depends on exactly which studies are included, and there is no benefit on function, the clinical relevance of combination therapy is not robustly demonstrated." We believe that this statement is unfounded, for the following reasons: (1) As demonstrated above (see the corrected Figure 3), there is a significant benefit on function when all three studies are included in the meta-analysis. (2) In our opinion, the authors' observation that study selection determines the overall effect on the clinical global impression measure (CIBIC-Plus) contradicts the purpose of conducting a meta-analysis. No relevant data set should be excluded from a meta-analysis without a compelling reason - for example, significant heterogeneity. Indeed, the authors' own analysis of global clinical impression for all 3 datasets (Figure 1) demonstrates a significant overall effect of memantine (p<0.001) without evidence of significant heterogeneity, suggesting no reasons for excluding any of the trials. Incidentally, if one data set were to be excluded, the most reasonable choice would be the moderate subset from the MD-12 trial, due to its higher baseline mean MMSE score (15, compared with a mean MMSE score of 10 for both MD-02 and MD-50). In that case, pooling of MD-02 and MD-50 data would yield a significant overall treatment effect for cognition (SIB; p<0.001), function (ADCS-ADL19; p=0.03), behavior (NPI; p<0.001), and clinical global impression (CIBIC-plus; p<0.001).

In short, we believe that the conclusions in this manuscript are not supported by the data, which, when pooled across all 3 trials, demonstrate significant benefits of adding memantine to ChEIs on cognitive, functional, behavioral, and global clinical outcome measures.

We strongly urge the authors to reassess and correct the analysis and presentation of data in this manuscript (particularly Figure 3), and to revise the discussion and conclusions so that they are consistent with the corrected data. We respectfully suggest that an appropriate response would be to withdraw the original publication and replace it with a corrected version in BMJ Open.

Sincerely,

George T. Grossberg, MD; Co-author of manuscripts from trials MD-02 [3], MD-12 [8], and MD-50 (submitted) and the Patel and Grossberg 2011 review [5]

Yvonne Wirth, MD PhD; Statistical consultant and co-author of meta-analysis [4] that was the source for MD-12 data for patients with moderate AD

Suzanne Hendrix, PhD; Statistical consultant

Michael Tocco, PhD; Medical Affairs, Forest Research Institute

Stephen M. Graham, PhD; Co-author of manuscripts from trials MD-02 [3] and MD-50 (submitted)

References

[1] Farrimond LE, Roberts E, McShane R. Memantine and cholinesterase inhibitor combination therapy for alzheimer's disease: A systematic review. BMJ Open 2012;2(3).

[2] Grossberg G, Manes F, Allegri R, Miguel L, Robledo L, Gloger S, et al. A multinational, randomized, double-blind, placebo-controlled, parallel-group trial of memantine extended-release capsule (28 mg, once daily) in patients with moderate to severe alzheimer's disease [abstract]. Alzheimers Dement 2008;4(4)(suppl 2):T793.

[3] Tariot PN, Farlow MR, Grossberg GT, Graham SM, McDonald S, Gergel I. Memantine treatment in patients with moderate to severe alzheimer disease already receiving donepezil: A randomized controlled trial. JAMA 2004;291(3):317-24.

[4] Winblad B, Jones RW, Wirth Y, Stoffler A, Mobius HJ. Memantine in moderate to severe alzheimer's disease: A meta-analysis of randomised clinical trials. Dement Geriatr Cogn Disord 2007;24(1):20-7.

[5] Patel L, Grossberg GT. Combination therapy for alzheimer's disease. Drugs Aging 2011;28(7):539-46.

[6] Periclou A, Hu Y. Extended-release (er) memantine capsule (28 mg, once daily): A multiple-dose, open-label study evaluating steady-state pharmacokinetics in healthy volunteers [abstract]. Alzheimers Dement 2008;4(4 (suppl 2)):T792.

[7] Wilkinson D, Fox NC, Barkhof F, Phul R, Lemming O, Scheltens P. Memantine and brain atrophy in alzheimer's disease: A 1-year randomized controlled trial. J Alzheimers Dis 2012;29(2):459-69.

[8] Porsteinsson AP, Grossberg GT, Mintzer J, Olin JT. Memantine treatment in patients with mild to moderate alzheimer's disease already receiving a cholinesterase inhibitor: A randomized, double-blind, placebo- controlled trial. Curr Alzheimer Res 2008;5(1):83-9.

Conflict of Interest:

Dr. George T. Grossberg has received consulting fees from Avanir Pharmaceuticals, Baxter Bioscience, Forest Laboratories, Eli Lilly, Lundbeck, Novartis Pharmaceuticals, and Otsuka Pharmaceutical; his department also receives research funds from Abbott Laboratories, Baxter, Forest, Janssen Pharmaceuticals, Novartis, and Pfizer Pharmaceuticals, as well as data safety monitoring committee fees from Abbott, Merck, and Schering-Plough. Dr. Yvonne Wirth is employed by Wirth Consulting, a statistical consultant of Merz, Inc. Dr. Suzanne Hendrix is employed by Pentara Corporation, a statistical consultant of Forest Research Institute, Inc. Drs. Michael Tocco and Stephen Graham are employed by Forest Research Institute. Medical writing and editorial assistance was provided by Drs. Vojislav Pejovic and Michael L. Miller from Prescott Medical Communications Group, Chicago, IL

This paper by Dahlen et al has reignited the public debate about obstetric intervention rates in Australia. Strangely - given the size of the dataset available to the authors - adverse perinatal outcomes were not examined in the study. However in the discussion the authors assert "these (higher) rates do not appear to be parallel to or be associated with a better infant outcome" and go on to cite a small single centre RCT of case loading midwifery in support of their assertion. The authors failed to cite or discuss a large population based study - again from Australia - that demonstrated a higher prevalence of adverse perinatal outcomes in public hospitals than in private hospitals between 2001 and 2004. After adjusting for multiple risk factors the study (Robson SJ et al, Med J Australia 2009; 190(9): 474 -7)found that the adjusted odds ratios (AORs) for a higher level of resuscitation and perinatal death were 2.37 (95% CI 2.17 - 2.59)and 2.02 (95% CI 1.78 - 2.29)respectively. I find it disappointing that Dahlen et al could not cite or discuss the Robson study given its contemporaneous nature and degree of relevance to their own study. An unkind interpretation would be that the authors have selectively used statistics in order to influence public debate in Australia.

Conflict of Interest:

None declared