- Email: [email protected]
Is it refractory idiopathic thrombocytopenic purpura?
THE LANCET
cardiac (non-arrhythmic) death, since the monitor shows a picture compatible with electromechanical dissociation. If no monitor strip had been available, the definition would have been arrhythmic death, diluting the power of the study to assess the effect of amiodarone on arrhythmic death, but not subject to bias with the rigorous procedures that were in place. In any case, the emphasis on a single patient anecdote by contrast with randomised controlled data on 1202 patients in CAMIAT and 1486 patients in EMIAT is surprising to say the least. *John A Cairns, Stuar t J Connolly, Robin Rober ts, Michael Gent, on behalf of the CAMIAT Investigators Faculty of Medicine, University of British Columbia, Vancouver, BC, Canada V6T 1Z 3
1
2
3
Hinkle LE Jr, Thaler HT. Clinical classification of cardiac deaths. Circulation 1982; 65: 457–64. Gottlieb SS. Dead is dead: artificial definitions are no substitute. Lancet 1997; 349: 662–63. Cairns JA, Connolly SJ, Roberts RS, Gent M. Canadian Amiodarone Myocardial Infarction Arrhythmia Trial (CAMIAT): rationale and protocol. Am J Cardiol 1993; 72: 87F–94F.
Commentator’s reply SIR—Despite the justifications and explanations of some of your correspondents, we should remember that CAMIAT1 and EMIAT2 were negative studies with respect to total mortality and that the isolated endpoint of sudden death is not clinically relevant. The findings do not exclude the importance of arrhythmias as a cause of morality, and I agree with Malik and Camm that it is appropriate to address the problem of arrhythmic death by selecting high-risk patients and evaluating the effects of an intervention. Perhaps they are even correct in assuming that patients with a high risk of arrhythmic (rather than general) mortality need to be studied (if such a group can be defined). However, until and unless total mortality is decreased in a defined population, a conclusion that the clinician should be “encouraged to consider amiodarone” does not seem warranted. Many of your correspondents continue to disregard the fact that sudden death is not equivalent to arrhythmic death, even if obvious cases of ischaemia or progressive heart failure are excluded. Deaths from myocardial infarctions, pulmonary emboli, and a host of other causes will be sudden, and may be a large proportion of these deaths. Conversely, a patient with ischaemia or heart failure could have a preventable arrhythmic death despite having other cardiac symptoms. Norris’ use of a study in which death was
1770
arbitrarily classified to prove that arbitrary classifications are beneficial is tautologous and not proof of this point, Although the proportion of arrhythmic deaths may be different in patients with heart failure than in those post myocardial infarction, the interaction between these two diseases is considerable. Indeed, Norris clearly realises the close connection between these diseases by proposing studies in patients with poor left ventricular function. I agree that the positive interaction of amiodarone and blockade is intriguing and deserves further investigation, especially in patients with heart failure. Of course, these studies must be powered to detect an improvement in total mortality. We should appreciate Cleland and Erhardt’s attempts at making sense of classification of death. The use of classification to try to understand physiology and the reasons behind an effect and to generate hypotheses may well be useful. As I am sure these workers would agree, however, the use of classification of sudden death as a justification for the use of a drug which does not affect mortality, is quite another matter. Julian and colleagues attempt to justify their study design by analogy to early studies of aspirin and lipidlowering agents which used causespecific benefits to lead to the larger more conclusive studies. However, the notion of amiodarone use post myocardial infarction is well past the early stage of investigation, with a metaanalysis (cited in the CAMIAT article) already suggesting mortality benefit.3 At this stage, large expensive studies should not be underpowered and designed to look at clinically unimportant and potentially misleading endpoints. Cairns and colleagues are correct in saying that “the most definitive evidence of overall drug benefit is provided by a statistically significant reduction of all-cause mortality”. We did not need another small study to be part of a metaanalysis. Stephen S Gottlieb Depar tment of Medicine, University of Mar yland School of Medicine, Baltimore, MD 21201, USA
1
2
3
Cairns JA, Connolly SJ, Roberts R, Gent M. Randomized trial of outcomes after myocardial infarction in patients with frequent to repetitive ventricular premature depolarizations. CAMIAT. Lancet 1997; 349: 675–82. Julian DG, Camm AJ, Frangin G, et al. Randomized trial of effect of amiodarone on mortality in patients with left ventricular dysfunction after recent myocardial infarction: EMIAT. Lancet 1997; 349: 667–74. Teo KK,Yusaf S, Furbrug CD. Effects of prophylactic antiarrhythmic drug therapy in acute myocardial infarction: an overview of results from randomised controlled clinical trials. JAMA 1993; 270: 1589–95
Is it refractory idiopathic thrombocytopenic purpura? SIR—Cahill and Newland (April 12, p 1066)1 diagnose Epstein’s syndrome, but without reporting the electronmicroscopic finding of the kidney biopsy sample, the family history, and genetic study of their patient. As they mention, crescentic glomerulonephritis is not a feature of Alport or Epstein’s syndrome. The main finding in these syndromes is the electronmicroscopic finding of kidney biopsy sample showing variable thickening-thinning, lamellation, and basket-weaving of the glomerular basement membrane.2 In addition, the deafness in Alport and Epstein’s syndrome is characterised by progressive loss of sensorineural hearing3 and not by recurrent otitis media, as seems to be the case from this report. Alport syndrome is transmitted in X-linked dominant or autosomal recessive pattern with the defects in the gene COL4A5 in the long arm of the X chromosome or COL4A3 or COL4A4 on chromosome 2.2 Therefore, the family history is of crucial importance. Has a genetic investigation been conducted in the patient’s family? Ocular changes, icthyosis, atopic dermatitis, and leiomyomatosis can also occur in Alport and Epstein’s syndromes;3 does the patient have any of these features? Pairach Pintavorn Southern Jersey Medical Center Inc, Pleasantville, NJ 08201, USA
1
2
3
Cahill MR, Newland AC. Is it refractory idiopathic thrombocytopenic purpura? Lancet 1997; 349: 1066. Churg J, Bernstein J, Glassock RJ. Alport syndrome. In: Churg J, Bernstein J, Glassock RJ, eds. Renal disease: classification and atlas of glomerular disease. Igaku-Shoin Medical Publishers, 1995: 409–11. Kashtan CE. Alport syndrome. In: Massry SG, Glassock RJ, eds. Textbook of nephrology. Baltimore: Williams and Wilkins, 1995: 869–72.
Authors’ reply SIR—In our patient with a macrothrombocytopathy, deafness, and glomerulonephritis, Epstein’s syndrome is the most likely diagnosis. There was no family history, and full blood counts on the parents, sibling, and other close relatives were normal. However, both an autosomal recessive condition and a new mutation are consistent with this information, and although genetic studies might be persuasive, working clinical diagnoses without genetic confirmation are commonly relied on to guide treatment. The patient has severe progressive sensorineural deafness and is dependent on the use of a combination of bilateral hearing aids and lip reading.
Vol 349 • June 14, 1997
THE LANCET
Recurrent otitis media (which he also had) was a common childhood disorder and is mentioned because the insertion of grommets (and preceding full blood count) led to the diagnosis of thrombocytopenia. Although our patient’s kidney biopsy sample showed crescentic glomerulonephritis, Epstein’s syndrome is rare, with few reported cases worldwide. But variation between cases is a possibility. The elctronmicroscopic characteristics of the platelets in this patient agree with previously reported abnormalities. Our patient did not have ichthyosis, atopic dermatitis, leiomyomatosis, or ocular abnormalities. The comments of Pintavorn in no way detract from the important message of our papers: do not assume a diagnosis of refractory idiopathic thrombocytopenic purpura until other causes of thrombocytopenia have been excluded. *Mar y Cahill, Adrian Newland Depar tment of Haematology, Royal London Hospital, London E1 1BB, UK
Association of insulin gene VNTR polymorphism with polycystic ovary syndrome SIR—We have several concerns about the evidence for linkage and association of the insulin gene VNTR polymorphism with polycystic ovary syndrome (PCOS) that Waterworth and colleagues report (April 5, p 986).1 In a case-control study, they report an odds ratio of 8·2 for the association of the III/III genotype with anovulatory PCOS. For this association the correct two-sided p-value by Fisher’s exact test is 0·011, not 0·005 as they report. Even this probability should be interpreted with caution, since it is based not on testing a hypothesis specified in advance, but on post hoc restriction of the dataset to anovulatory cases and combining of the other two genotypes into a single category. The family-based association study would be an appropriate design in which to test the specific hypothesis raised by the case-control study: that the III/III genotype is associated with anovulatory PCOS. Assuming random mating, the genotypes constructed from pairs of untransmitted parental alleles should be in Hardy-Weinberg equilibrium, and representative of the population in which the cases arose. Analysis of table 3 shows, however, that the control genotype frequencies (15, 24, and 0 for genotypes I/I, I/III, and III/III, respectively) are not in HardyWeinberg equilibrium (2-test, p=0·006). This departure from
Vol 349 • June 14, 1997
expected frequencies could result from chance, systematic error in genotyping, or non-random mating. Whatever the reason for this strange result, the assumptions on which the family-based association study design depends2 are clearly violated, and any evidence of association in this dataset should be disregarded. With such doubtful evidence for an association, we may look more critically at the evidence for linkage of the INS gene region with PCOS. Of the five markers used in the linkage analysis, two—INS VNTR and HUMTH01—are only 0·01 cM apart and are in strong linkage disequilibrium.3 The Genehunter program4 used to test for linkage assumes that all markers used are in linkage equilibrium with each other. If this assumption is violated, probability distributions calculated by the program would be incorrect. The effect on detection of linkage is hard to predict, but disequilibrium between adjacent markers would mean that haplotypes identical by state would be shared within pedigrees more often than expected. This in turn could lead to overestimation of identity-by-descent sharing and false-positive evidence for linkage. To examine this possibility, Waterworth and colleagues could repeat the Genehunter test for linkage after excluding any markers that are in linkage disequilibrium with each other. Until these uncertainties are resolved, their proposal to designate the INS locus as PCOS1 is premature. *Paul McKeigue, Sarah Wild Depar tment of Epidemiology and Population Sciences, Epidemiology Unit, London School of Hygiene and Tropical Medicine, London WC1E 7HT, UK
1
2
3
4
Waterworth DM, Bennett ST, Gharani N, et al. Linkage and association of insulin gene VNTR regulatory polymorphism with polycystic ovary syndrome. Lancet 1997; 349: 986–90. Falk CT, Rubinstein P. Haplotype relative risks: an easy reliable way to construct a proper control sample for risk calculations. Ann Hum Genet 1987; 51: 227–33. Bennett ST, Lucassen AM, Gough SC, et al. Susceptibility to human type 1 diabetes at IDDM2 is determined by tandem repeat variation at the insulin gene minisatellite locus. Nat Genet 1995; 9: 284–92. Kruglyak L, Daly MJ, Reeve-Daly MP, Lander ES. Parametric and nonparametric linkage analysis: a unified multipoint approach. Am J Hum Genet 1996; 58: 1347–63.
Authors’ reply SIR—We have repeated the Genehunter analysis for linkage of 11p15·5 to PCOS. Irrespective of whether HUMTH01, the INS VNTR, or both loci are excluded from the analysis, we obtain the same support for linkage at D11S922 (single point non-parametric linkage value [NPL]=2·77, p=0·005;
multipoint [excluding HUMTH01] NPL=3·2, p=0·002). Previous studies have reported associations of the INS VNTR class III/III genotype with type 2 diabetes and related disorders.1,2 Women with PCOS have increased risk of developing type 2 diabetes, and anovulatory PCOS is associated with hyperinsulinaemia. Therefore, our hypothesis was that the INS VNTR class III allele frequency would be higher in anovulatory PCOS patients than in controls. This prediction was confirmed in the casecontrol data set: the exact p-value obtained is 0·0059 (Statxact, Cytel Corporation, Cambridge, MA). The value cited by McKeigue and Wild (p=0·011) is conservative. We use the more conventional mid-point value. Analysis of the entire 3⫻3 table (our table 2) shows striking departure from independent assortment (p=0·008 for genotypes, Kruskal-Wallis test). The source of this departure lies predominantly in the comparison of normal versus anovulatory PCOS data sets (p=0·0027). We did a family-based association analysis with an independent family data set. The constructed AFBAC control genotypes are not in Hardy-Weinberg equilibrium, as pointed out by McKeigue and Wild. We believe this to be a chance event. Nevertheless, the trend is positive and consistent with the independent case-control analysis. Furthermore, we have analysed these families for evidence of linkage of PCOS to the insulin VNTR in the presence of association with the transmission disequilibrium test, in which allele transmission from heterozygous parents to affected offspring is measured and is not dependent on Hardy-Weinberg equilibrium. 3 Class III alleles were transmitted more often than class I VNTR alleles from class I/III heterozygous parents to offspring with anovulatory PCOS (27 transmissions, 17 non-transmissions; 61% transmission; 2-test, p=0·132). Because parent-of-origin effects have been observed at this locus in type 1 diabetes4 and there is evidence for gametic imprinting of this region of 11p15·5,5 we evaluated allele transmission from mothers and fathers, separately. Maternal transmissions did not depart from that expected for random transmission (12 transmissions, 12 nontransmissions; 50%). However, class III alleles were transmitted significantly more often than class I alleles from I/III fathers (15 transmissions, five nontransmissions; 75%; p=0·025). This analysis was extended to include transmissions of VNTR class III alleles in the 17 PCOS pedigrees used for the linkage study. In the combined data, class III alleles had a 47% transmission from mothers (14 transmissions, 16
1771
cardiac (non-arrhythmic) death, since the monitor shows a picture compatible with electromechanical dissociation. If no monitor strip had been available, the definition would have been arrhythmic death, diluting the power of the study to assess the effect of amiodarone on arrhythmic death, but not subject to bias with the rigorous procedures that were in place. In any case, the emphasis on a single patient anecdote by contrast with randomised controlled data on 1202 patients in CAMIAT and 1486 patients in EMIAT is surprising to say the least. *John A Cairns, Stuar t J Connolly, Robin Rober ts, Michael Gent, on behalf of the CAMIAT Investigators Faculty of Medicine, University of British Columbia, Vancouver, BC, Canada V6T 1Z 3
1
2
3
Hinkle LE Jr, Thaler HT. Clinical classification of cardiac deaths. Circulation 1982; 65: 457–64. Gottlieb SS. Dead is dead: artificial definitions are no substitute. Lancet 1997; 349: 662–63. Cairns JA, Connolly SJ, Roberts RS, Gent M. Canadian Amiodarone Myocardial Infarction Arrhythmia Trial (CAMIAT): rationale and protocol. Am J Cardiol 1993; 72: 87F–94F.
Commentator’s reply SIR—Despite the justifications and explanations of some of your correspondents, we should remember that CAMIAT1 and EMIAT2 were negative studies with respect to total mortality and that the isolated endpoint of sudden death is not clinically relevant. The findings do not exclude the importance of arrhythmias as a cause of morality, and I agree with Malik and Camm that it is appropriate to address the problem of arrhythmic death by selecting high-risk patients and evaluating the effects of an intervention. Perhaps they are even correct in assuming that patients with a high risk of arrhythmic (rather than general) mortality need to be studied (if such a group can be defined). However, until and unless total mortality is decreased in a defined population, a conclusion that the clinician should be “encouraged to consider amiodarone” does not seem warranted. Many of your correspondents continue to disregard the fact that sudden death is not equivalent to arrhythmic death, even if obvious cases of ischaemia or progressive heart failure are excluded. Deaths from myocardial infarctions, pulmonary emboli, and a host of other causes will be sudden, and may be a large proportion of these deaths. Conversely, a patient with ischaemia or heart failure could have a preventable arrhythmic death despite having other cardiac symptoms. Norris’ use of a study in which death was
1770
arbitrarily classified to prove that arbitrary classifications are beneficial is tautologous and not proof of this point, Although the proportion of arrhythmic deaths may be different in patients with heart failure than in those post myocardial infarction, the interaction between these two diseases is considerable. Indeed, Norris clearly realises the close connection between these diseases by proposing studies in patients with poor left ventricular function. I agree that the positive interaction of amiodarone and blockade is intriguing and deserves further investigation, especially in patients with heart failure. Of course, these studies must be powered to detect an improvement in total mortality. We should appreciate Cleland and Erhardt’s attempts at making sense of classification of death. The use of classification to try to understand physiology and the reasons behind an effect and to generate hypotheses may well be useful. As I am sure these workers would agree, however, the use of classification of sudden death as a justification for the use of a drug which does not affect mortality, is quite another matter. Julian and colleagues attempt to justify their study design by analogy to early studies of aspirin and lipidlowering agents which used causespecific benefits to lead to the larger more conclusive studies. However, the notion of amiodarone use post myocardial infarction is well past the early stage of investigation, with a metaanalysis (cited in the CAMIAT article) already suggesting mortality benefit.3 At this stage, large expensive studies should not be underpowered and designed to look at clinically unimportant and potentially misleading endpoints. Cairns and colleagues are correct in saying that “the most definitive evidence of overall drug benefit is provided by a statistically significant reduction of all-cause mortality”. We did not need another small study to be part of a metaanalysis. Stephen S Gottlieb Depar tment of Medicine, University of Mar yland School of Medicine, Baltimore, MD 21201, USA
1
2
3
Cairns JA, Connolly SJ, Roberts R, Gent M. Randomized trial of outcomes after myocardial infarction in patients with frequent to repetitive ventricular premature depolarizations. CAMIAT. Lancet 1997; 349: 675–82. Julian DG, Camm AJ, Frangin G, et al. Randomized trial of effect of amiodarone on mortality in patients with left ventricular dysfunction after recent myocardial infarction: EMIAT. Lancet 1997; 349: 667–74. Teo KK,Yusaf S, Furbrug CD. Effects of prophylactic antiarrhythmic drug therapy in acute myocardial infarction: an overview of results from randomised controlled clinical trials. JAMA 1993; 270: 1589–95
Is it refractory idiopathic thrombocytopenic purpura? SIR—Cahill and Newland (April 12, p 1066)1 diagnose Epstein’s syndrome, but without reporting the electronmicroscopic finding of the kidney biopsy sample, the family history, and genetic study of their patient. As they mention, crescentic glomerulonephritis is not a feature of Alport or Epstein’s syndrome. The main finding in these syndromes is the electronmicroscopic finding of kidney biopsy sample showing variable thickening-thinning, lamellation, and basket-weaving of the glomerular basement membrane.2 In addition, the deafness in Alport and Epstein’s syndrome is characterised by progressive loss of sensorineural hearing3 and not by recurrent otitis media, as seems to be the case from this report. Alport syndrome is transmitted in X-linked dominant or autosomal recessive pattern with the defects in the gene COL4A5 in the long arm of the X chromosome or COL4A3 or COL4A4 on chromosome 2.2 Therefore, the family history is of crucial importance. Has a genetic investigation been conducted in the patient’s family? Ocular changes, icthyosis, atopic dermatitis, and leiomyomatosis can also occur in Alport and Epstein’s syndromes;3 does the patient have any of these features? Pairach Pintavorn Southern Jersey Medical Center Inc, Pleasantville, NJ 08201, USA
1
2
3
Cahill MR, Newland AC. Is it refractory idiopathic thrombocytopenic purpura? Lancet 1997; 349: 1066. Churg J, Bernstein J, Glassock RJ. Alport syndrome. In: Churg J, Bernstein J, Glassock RJ, eds. Renal disease: classification and atlas of glomerular disease. Igaku-Shoin Medical Publishers, 1995: 409–11. Kashtan CE. Alport syndrome. In: Massry SG, Glassock RJ, eds. Textbook of nephrology. Baltimore: Williams and Wilkins, 1995: 869–72.
Authors’ reply SIR—In our patient with a macrothrombocytopathy, deafness, and glomerulonephritis, Epstein’s syndrome is the most likely diagnosis. There was no family history, and full blood counts on the parents, sibling, and other close relatives were normal. However, both an autosomal recessive condition and a new mutation are consistent with this information, and although genetic studies might be persuasive, working clinical diagnoses without genetic confirmation are commonly relied on to guide treatment. The patient has severe progressive sensorineural deafness and is dependent on the use of a combination of bilateral hearing aids and lip reading.
Vol 349 • June 14, 1997
THE LANCET
Recurrent otitis media (which he also had) was a common childhood disorder and is mentioned because the insertion of grommets (and preceding full blood count) led to the diagnosis of thrombocytopenia. Although our patient’s kidney biopsy sample showed crescentic glomerulonephritis, Epstein’s syndrome is rare, with few reported cases worldwide. But variation between cases is a possibility. The elctronmicroscopic characteristics of the platelets in this patient agree with previously reported abnormalities. Our patient did not have ichthyosis, atopic dermatitis, leiomyomatosis, or ocular abnormalities. The comments of Pintavorn in no way detract from the important message of our papers: do not assume a diagnosis of refractory idiopathic thrombocytopenic purpura until other causes of thrombocytopenia have been excluded. *Mar y Cahill, Adrian Newland Depar tment of Haematology, Royal London Hospital, London E1 1BB, UK
Association of insulin gene VNTR polymorphism with polycystic ovary syndrome SIR—We have several concerns about the evidence for linkage and association of the insulin gene VNTR polymorphism with polycystic ovary syndrome (PCOS) that Waterworth and colleagues report (April 5, p 986).1 In a case-control study, they report an odds ratio of 8·2 for the association of the III/III genotype with anovulatory PCOS. For this association the correct two-sided p-value by Fisher’s exact test is 0·011, not 0·005 as they report. Even this probability should be interpreted with caution, since it is based not on testing a hypothesis specified in advance, but on post hoc restriction of the dataset to anovulatory cases and combining of the other two genotypes into a single category. The family-based association study would be an appropriate design in which to test the specific hypothesis raised by the case-control study: that the III/III genotype is associated with anovulatory PCOS. Assuming random mating, the genotypes constructed from pairs of untransmitted parental alleles should be in Hardy-Weinberg equilibrium, and representative of the population in which the cases arose. Analysis of table 3 shows, however, that the control genotype frequencies (15, 24, and 0 for genotypes I/I, I/III, and III/III, respectively) are not in HardyWeinberg equilibrium (2-test, p=0·006). This departure from
Vol 349 • June 14, 1997
expected frequencies could result from chance, systematic error in genotyping, or non-random mating. Whatever the reason for this strange result, the assumptions on which the family-based association study design depends2 are clearly violated, and any evidence of association in this dataset should be disregarded. With such doubtful evidence for an association, we may look more critically at the evidence for linkage of the INS gene region with PCOS. Of the five markers used in the linkage analysis, two—INS VNTR and HUMTH01—are only 0·01 cM apart and are in strong linkage disequilibrium.3 The Genehunter program4 used to test for linkage assumes that all markers used are in linkage equilibrium with each other. If this assumption is violated, probability distributions calculated by the program would be incorrect. The effect on detection of linkage is hard to predict, but disequilibrium between adjacent markers would mean that haplotypes identical by state would be shared within pedigrees more often than expected. This in turn could lead to overestimation of identity-by-descent sharing and false-positive evidence for linkage. To examine this possibility, Waterworth and colleagues could repeat the Genehunter test for linkage after excluding any markers that are in linkage disequilibrium with each other. Until these uncertainties are resolved, their proposal to designate the INS locus as PCOS1 is premature. *Paul McKeigue, Sarah Wild Depar tment of Epidemiology and Population Sciences, Epidemiology Unit, London School of Hygiene and Tropical Medicine, London WC1E 7HT, UK
1
2
3
4
Waterworth DM, Bennett ST, Gharani N, et al. Linkage and association of insulin gene VNTR regulatory polymorphism with polycystic ovary syndrome. Lancet 1997; 349: 986–90. Falk CT, Rubinstein P. Haplotype relative risks: an easy reliable way to construct a proper control sample for risk calculations. Ann Hum Genet 1987; 51: 227–33. Bennett ST, Lucassen AM, Gough SC, et al. Susceptibility to human type 1 diabetes at IDDM2 is determined by tandem repeat variation at the insulin gene minisatellite locus. Nat Genet 1995; 9: 284–92. Kruglyak L, Daly MJ, Reeve-Daly MP, Lander ES. Parametric and nonparametric linkage analysis: a unified multipoint approach. Am J Hum Genet 1996; 58: 1347–63.
Authors’ reply SIR—We have repeated the Genehunter analysis for linkage of 11p15·5 to PCOS. Irrespective of whether HUMTH01, the INS VNTR, or both loci are excluded from the analysis, we obtain the same support for linkage at D11S922 (single point non-parametric linkage value [NPL]=2·77, p=0·005;
multipoint [excluding HUMTH01] NPL=3·2, p=0·002). Previous studies have reported associations of the INS VNTR class III/III genotype with type 2 diabetes and related disorders.1,2 Women with PCOS have increased risk of developing type 2 diabetes, and anovulatory PCOS is associated with hyperinsulinaemia. Therefore, our hypothesis was that the INS VNTR class III allele frequency would be higher in anovulatory PCOS patients than in controls. This prediction was confirmed in the casecontrol data set: the exact p-value obtained is 0·0059 (Statxact, Cytel Corporation, Cambridge, MA). The value cited by McKeigue and Wild (p=0·011) is conservative. We use the more conventional mid-point value. Analysis of the entire 3⫻3 table (our table 2) shows striking departure from independent assortment (p=0·008 for genotypes, Kruskal-Wallis test). The source of this departure lies predominantly in the comparison of normal versus anovulatory PCOS data sets (p=0·0027). We did a family-based association analysis with an independent family data set. The constructed AFBAC control genotypes are not in Hardy-Weinberg equilibrium, as pointed out by McKeigue and Wild. We believe this to be a chance event. Nevertheless, the trend is positive and consistent with the independent case-control analysis. Furthermore, we have analysed these families for evidence of linkage of PCOS to the insulin VNTR in the presence of association with the transmission disequilibrium test, in which allele transmission from heterozygous parents to affected offspring is measured and is not dependent on Hardy-Weinberg equilibrium. 3 Class III alleles were transmitted more often than class I VNTR alleles from class I/III heterozygous parents to offspring with anovulatory PCOS (27 transmissions, 17 non-transmissions; 61% transmission; 2-test, p=0·132). Because parent-of-origin effects have been observed at this locus in type 1 diabetes4 and there is evidence for gametic imprinting of this region of 11p15·5,5 we evaluated allele transmission from mothers and fathers, separately. Maternal transmissions did not depart from that expected for random transmission (12 transmissions, 12 nontransmissions; 50%). However, class III alleles were transmitted significantly more often than class I alleles from I/III fathers (15 transmissions, five nontransmissions; 75%; p=0·025). This analysis was extended to include transmissions of VNTR class III alleles in the 17 PCOS pedigrees used for the linkage study. In the combined data, class III alleles had a 47% transmission from mothers (14 transmissions, 16
1771