Ann. Hum. Genet., Lond. (1978), 41, 277 Printed in Great Britain
277
Parental age and birth order in the aetiology of some sex chromosome aneuploidies BY A. D. CAROTHERS, S. COLLYER, R. DE MEY
AND
A. FRACKIEWICZ
M.R.C. Clinical and Population Cytogenetics Unit, Western General Hospital, Edinburgh, EH4 3XU INTRODUCTION
The possibility of an association between maternal age at birth and the incidence of certain sex chromosome aneuploidies is suggested by analogy with the well-established case of Down’s syndrome (Penrose & Smith, 1966). Lenz et al. (1959), Ferguson-Smith et al. (1964), Court Brown, Law & Smith (1969), Borgaonkar & Mules (1970) and Tumba (1974) have all reported an increased maternal age at birth for certain types of sex chromosome abnormality as compared with control groups. A common feature of these studies, as of the present one, is that propositi were ascertained, not through random samples of the general population, but through selected categories such as those attending endocrine, sub-fertility or psychiatric clinics, inmates of hospitals for the mentally sub-normal and so on. From such data a straight comparison of the mean maternal age at birth of the propositi with that of the general population will not establish a causal relationship with the incidence of the abnormality since the samples may be biased in respect of parental age. Surveys of the consecutive newborn are closer to a random sample and several have recently been published (Lubs & Ruddle, 1970; Priedrich & Nielsen, 1973; Bell & Corey, 1974: Jacobs et al. 1974; Hamerton et al. 1975). However, the number of abnormal individuals revealed in this way is rather small and even if the data from all these surveys are combined the evidence on maternal age is inconclusive for even the most frequent abnormalities. We have attempted to avoid these problems by analysing data on a comparatively large number of propositi from selected categories, but with careful attention to possible sources of bias. Our specific objectives were: to demonstrate the presence or absence of a maternal age effect for certain sex chromosome aneuploidies; to show that, if present, the effect is not a result of sampling bias and is not a secondary one arising from the correlation between maternal age and paternal age or between maternal age and birth order; finally, to estimate the size of the effect and the probable limits between which it lies. The particular abnormalities considered are those on which we have the most data, namely, the three aneuploidies with karyotypes 47,XXY, 4 7 , X X X and 47,XYY. For brevity these will be referred to simply as X X Y , XXX and X Y Y. (Some of our data were included in the study by Court Brown et al. (1969) referred to above, and our results are therefore not entirely independent.)
THE DATA
Data for this study were taken from the Registry of Abnormal Karyotypes at this Unit. The Registry and the method of determining karyotypes from cultured blood have been fully described by Court Brown et al. (1964). Propositi were included only if they had a non-mosaic I9
HGE
41
278
A. D. CAROTHERSAND
OTHERS
X X Y , X,YX, or X Y Y karyotype and were ascertained before the end of 1975 in one of the following categories : ( a ) patients a t endocrine and/or subfertility clinics, (b) patients identified a t hospitals for the mentally subnormal or mentally ill, (c) individuals referred from general practices or hospital wards, ( d ) inmates of penal institutions. The method has been t o compare parental ages and birth order of the propositi with those of their liveborn full-sibs, who provide a control for many genetic and environmental factors. Propositi were included only if their own years of birth were known and also those of both their parents and of all their full-sibs born alive before the date of ascertainment. Most dates of birth were verified from birth certificates, and in only a few cases was it necessary to rely on the memories of the propositus or of a near relative. I n 11 cases ( < 1 % of the total), the year of birth of a liveborn full-sib was estimated as halfway between the preceding and following sibs. Where the mother was of child bearing age, defined as less than 45 years, a t the date of ascertainment, an attempt was made to discover her subsequent conceptual history up to the end of 1975, and all further liveborn full-sibs were included in the data. This proved possible for all but 9 X X Y ' s and 4 X Y Y ' s , the main reason for failure being that the mother had emigrated from Scotland. I n fact, for reasons given in the following section, the failure t o trace subsequent liveborn sibs has a negligible effect on the results of the analysis. The numbers of propositi satisfying all criteria are given in Tables 1-3. Birth order was defined as one greater than the number of known previous pregnancies of the mother, with stillbirths and abortions counted and twins counted as two pregnancies. I-9'ince very few of the sibs of propositi have been karyotyped, all were assumed t o be normal. The proportion of abnormals among them is probably too small to affect the conclusions, but this is an unverified assumption. STATISTICAL METHODOLOGY
Let to, x, y, x denote respectively the year of birth, maternal age a t birth, paternal age a t birth and birth order of a particular individual and put -
w
= (PU,
x, y, 2 ) .
Let A denote ascertainment as a propositus in this study, and K the possession of the abnormal karyotype being considered. Then P(AI8) = P(KIV).
P(AIK, 8 ) .
We shall assume that
P(KI8) = 1;,f(x, Y, z ) , where r, is the absolute incidence of the abnormal karyotype among all live-births in ycar and f is some unspecified function, and that
P(AIK
to
= g,,
where g, is the probability of ascertainment of an affected individual born in year w.The first assumption allows for variation in the absolute incidence of the abnormal karyotype from year t o year, but requires that the relative incidence a t different parental ages and birth ordcrs remains the same. The second allows the probability of ascertainment of an affected indiviclual to depend on w,but not on x, y or x . With these assumptions
P(AI@)= g,ru,f(G Y, 2 ) .
Parental age in some sex chromosome aneuploidies
279
Table 1. Summary of results for XXY's
Propositi included in subset
No. in subset
All All All All
234 234 234 234
Ascertainment Ascertainment Ascertainment Ascertainment
A B C D
I22
M* 0
2
4
co 2 2
47 47 18
2
Sibship size 2, 3, 4 or 5 t Sibship size 3 6
127 85
2
Born before 1930 Born 1930 or after
99 I35
2 2
2
2
Estimated rel. risk at mat. age 40 Y S
Lower 95 yo C.L.
Upper 95 yo C.L.
3'97 3.69 3.66 439
2.62 2'47 2.46 3'05
6.02 5.51 5'46 6.3 I
3'40 2'42 5.58 6.80
1.76 1.16 0.72
6.58 5'05 1426 6432
2.96 4'63
1.49 2'74
5.88 7'84
3.76 3'72
2.04 1.87
6.92 7'42
2-18
+
The assumed model isf = exp [a. bz2]. * M defines the smoothing function. See Statistical Methodology section. t Propositi with no full sibs are excluded.
Table 2. Summary of results for XXX's
Propositi included in subset
Estimated rel. risk a t mat. age 40 Yrs
Lower 95 yo C.L.
1.71 1.98 2.06 2.36
I 1.23
aJ
438 4'33 4'52 5.06
2
3'42
1'49
7.87
2
I 1.91
1.03
137.60
20
2
I1
2
16.44 3'29
1.17 I '42
231'17 7.61
I7 26
2 2
4.60 2.80
1.27 0.94
16.73 8'33
No. in subset
43 43 43 43
All All All All Ascertainment B Ascertainment A Ascertainment C Ascertainment D Sibship size 2, 3, 4 or 5 t Sibship size 2 6 Born before 1920 Born 1920 or after
M* 0
2
4
Upper 95 yo C.L.
9.46 9'90 10.86
3\: I 0 I1 OJ
The assumed model is f = exp [ax2]. * M defines the smoothing function. See Statistical Methodolorn section. t Propositi with no full sibs are excluded.
Now consider sibships of n individuals with parameters given by Let The conditional probability, given n and V j ( j = 1,. . ., n ) , that such a sibship is ascertained through the ith sib alone is Pi- Prob (the sibship is ascertained through the ith sib and a t least one other). For rare abnormalities such as those we are considering, the second term can be 19-2
A. D. CAROTHERSAND
280
OTHERS
Table 3. Summary of results for XYY's
No. in subset
Propositi included in subset
03
0.19 0.27
1.27
2
0.3 I
0.06
1'59
2
0.29
0.09
1'00
2 2
0.3 I 0'73
0'10
I1
0'97 2'95
22 22
2 2
0.40
0.09
1.71
0.73
0-16
3'33
20
3, 4 or
Upper 9 5 % C.L.
1-18 0'94
Ascertainment B Ascertainment A Ascertainment C Ascertainment D
>6
Lower
95 % C.L.
0.17 0.17
44 44 44 44
2,
M*
0.45 0.40 0'43 0.59
All All All All
Sibship size Sibship size
Estimated rel. risk at mat. age 40 Yrs
0
2
4
3\
1'00
24
st
Born before 1940 Born 1940 or after
29
0.18
The rtssumed model isf = exp [m]. * M defines the smoothing function. See Statistical Methodology sect,ion. t Propositi with no full sibs are excluded.
ignored. For example, the combined results of the five newborn surveys referred to above suggest that none of these abnormalities has an incidence exceeding one per 1500 live births, and in fact no sibship with multiple ascertainment was obtained in the present study. Hence the probability reduces simply to Pi and, conditioned on there being exactly one propositus in the sibship, becomes I
P'l
n
2 Pj.
j=1
An unbiased estimate of q,,, is where nW = number of propositi born in year to, Nw = total number of livebirths in yenr IU in the catchment area of the study (in this case taken to be Scotland). I n fact the conditional probability is unaffected if g, is multiplied by a constant, showing that it is necessary to asstime only that the number of livebirths in the catchment area is proportional t o the number in the country as a whole. The conditional probability now becomes
&=
,f(xi,Y i , z i )
C hijf(xjr ~ j= 1
,
j zj) ,
where
(nwj/Nu,)/(nW1/~wt). Since in general n, is small and since gw might be expected to change rather slowly with time, it seems reasonable to replace nwl/N,, by a moving average of type hij =
the values of Nw being abstracted from the tables of the Registrar-General for Scotland. We have arbitrarily chosen to set M = 2, corresponding t o a 5-year moving average, but we shall show that within quite wide limits the choice of M has a negligible effect on the conclusions.
Parental age in some sex chromosome aneuploidies
281
The conditional likelihood for the whole data set is given by the product of the conditional probabilities, &, over all sibships in the study. By maximizing this we obtain maximum likelihood estimates of the parameters of any specified model for f ( x , y, z ) together with their approximate asymptotic variances and covariances. Several computer routines are avilable for doing this and the details need not concern us. Any family of models for f is suitable so long as it is both mathematically tractable and sufficiently flexible to approximate any function likely t o be encountered. The family used here is that defined by f(X,
Y, 2)
=
B exp [h(x,Y, 41,
where B is a constant and h a general polynomial. I n fact, B is redundant since it is a common factor of both the numerator and denominator of Q so that without loss of generality we may put B = 1. This indicates that the absolute risk of an abnormality cannot be evaluated by this method. However, the relative risk is estimable and is arbitrarily defined t o be 1 at parental age 30 years and birth order 7. The need to include additional parameters in any model can be assessed using likelihood-ratio tests. For example, with the models, model 1 : h
=
ax,
model 2 : h = ax+bx2 under the null hypothesis that b - 2 log
=
0 , the statistic
[maximum likelihood under model 21 maximum likelihood under model 1
is approximately distributed as x2 with 1 degree of freedom (see, for example, Mood & Graybill, 1963, chap. 12). A ‘best’ model can be defined, though not uniquely, as one satisfying the criteria : (1) That the addition of any further terms in x, y or z does not increase the likelihood significantly. (2) That the removal of any term reduces the likelihood significantly. Constraints of time and expense make it necessary t o confine the search for a ‘best’ model to fairly simple functions - in the present case u p to order 3 or 4 in x , y and x . Since h is a linear function of the model parameters, the derivation of approximate confidence limits for it from the asymptotic variance-covariance matrix is straightforward. These are converted into confidence limits for f by exponentiation. The form of the expression for Q shows why the omission from the data of sibs born after the date of ascertainment has a negligible effect on the results. The value of nwj,and hence of A,,, for these sibs will in general be zero, since nearly all propositi in this study were born before 1959, the first year of ascertainment. Potential sources of bias were investigated by examining various subsets of the data. The full data on each abnormality mere subdivided as follows: (1) By size of sibship - small (2, 3, 4 or 5) and large (6 or more). (2) By method ascertainment - A , B , C and D for the X X Y ’ s ; B (the largest category) and all others ( A , C and D ) for t,he X X X ’ s and XYY’s. (3) By year of birth of propositi - early (before 1930, 1920 and 1940 for the X X Y ’ s , X X X ’ s and X Y Y’s respectively) and late (in or after the stated years).
A. D. CAROTHERSAND
282
OTHERS
Ideally the ‘best’models for each subset would be found separately and the predictions from the models compared, but t o reduce the considerable computing time and expense we have adopted the simpler course of taking the ‘best’ model derived from the full data set and estimating its parameters for each subset. For example, if the ‘best’ model for the full data set is of the form
f = exp [ax+bx2] we obtain the maximum likelihood estimates of a and b for each subset separately. A convenient and easily summarized method of comparing predictions from different subsets is t o est>imate the relative risks a t parental age 40 years, together with 95 yo confidence limits. RESULTS
XXY The ‘best’ model was
f(x, y, z ) = exp [O.O81831(x- 30) +0.004874(.2:- 30)2].
No significant improvement in the likelihood was obtained by adding further terms in 5 , y or z whereas the ‘best’ models involving y only or x only were significantly improved by adding terms in x. I n other words, a model excluding a maternal age effect cannot fully explain the data, whereas a model including only a maternal age effect can. This does not rule out the possibility of paternal age or birth order effects, but shows that they need not be postnlated. The ‘best’ model for f, together with its 95 yo confidence limits, is shown in Fig. 1. Table 1 gives the estimated risk a t maternal age 40 for the various data subsets. It can be seen that the estimates are mutually consistent and, in all but the smallest category (method of ascertainment D),significantly greater than 1. The possibility that the observed increase in risk is confined t o a part,icular subset may therefore be excluded.
xxx The ‘best’ model involving .2: only and that involving y only both gave an equally good fit to the data. However, both gave a significantly better fit than the ‘best ’ model involving z only, and neither was significantly improved by adding terms in x . We conclude that it is not possible to explain the data fully without postulating a maternal and/or paternal age effect,, but that it is possible to do so without postulating a birth order effect. The ‘best’ models were f(x,y, z ) = exp [0*014645(x-30)2] and f(x,y, z ) = exp [O-O09564(y- 30)2]. These, together with their approximate 95% confidence limits, are shown in Figs, 2 and 3 respectively. It is interesting t o note that both models predict an increased risk a t ages below, as well as above, 30 years. Table 2 shows that the estimated risks a t age 40 obtained from the various subsets are mutually consistent and greater than 1. There is therefore no evidence that the observed increase in risk is confined t o any subset.
XYY The ‘bcst ’ model was
f(x, y, z )
=
exp [ - 0.092567(x- 30)],
indicating a small but significant inverse relationship between risk and maternal age (Fig. 4).
Parental age in some sex chromosome aneuploidies 283 It is easily verified by substitution in the expression for Q that x and y are interchangeable in this model, since for any particular sibship x-y = constant. Thus we cannot distinguish paternal from maternal age effects. However, the model gave a significantly better fit than the best one involving z alone and was not significantly improved by adding any terms in z. It is therefore not necessary to postulate any dependence on birth order. Table 3 shows that the reduced risk a t age 40 is common to all subsets of the data.
General Tables 1-3 show that the choice of smoothing function has little effect on the results. I n all cases similar estimates are obtained whether a 1-year or a 9-year moving average is used. It can also be seen that if the differences in the annual probabilities of ascertainment are ignored by puttingM = co (or equivalently hij = l), a small but consistent overestimate of the risk at later parental ages is obtained in all cases.
DISCUSSION
The results agree with previous studies in indicating the presence of an effect of maternal age, or some correlated factor, on the incidence of X X Y’s and X X X ’ s . They go further in suggesting that, of the three highly correlated factors examined, maternal age is the most likely causative factor for the XX Y’s,while birth order is the least likely for theXXX’s. They differ in indicating the presence of a small but significant inverse relationship between parental age and the incidence of X Y Y’s. The effects are apparently independent of the method of ascertainment, the size of sibship and the year of birth of the propositus. However, studies of this nature are so fraught with problems of interpretation that we must consider in more detail a number of possible objections. (1) The propositi in this study are all post-pubertal and one might argue that there could be a different maternal age distribution among newborn and post-pubertal abnormals, such as would arise if the affected offspring of older mothers were more, or less, likely to reach puberty than those of younger mothers. This seems most unlikely since none of these abnormalities is severe and there is no evidence of an appreciable pre-pubertal mortality. (2) The birth of an affected individual may influence the subsequent conceptual history of the mother. I n particular, if the mother is less likely to have further children, affected individuals would occur more frequently a t the end of a sibship and this might lead to an apparent increase in risk a t later maternal ages. Again, it seems unlikely that such an effect is present for these particular abnormalities since they are rarely diagnosed before puberty, by which time most sibships are complete. I n the case of Down’s syndrome, a much more severe abnormality, Sigler et al. (1967) were unable to find any evidence of a difference in fertility, either before or after the birth of an affected individual, between the mothers of children with the disorder and the mothers of an equal number of normal controls, matched for sex, place of birth, date of birth and maternal age at birth. (3) Since the estimates of relative risk given in Figs. 1-4 have been derived from a highly selected group of sibships, a further limitation is the possibility that the results may not apply to parents selected a t random from the general population. The following argument shows under what conditions such a generalization can be made.
I
I
2.
ic
--
2 --
4
6 --
8-
I
35
Fig. 3. X X X .
Paternal age at birth (years)
I
14--
16 --
18 --
30
- 12-%. 2 lo--
v)
- - _- - _
25
I
40
/.,
I
I
I
I II
t
,
I
I I
I
.
I
I
I
20
I
15
3 --
4--
5--
I
I
35
I
40
-----____ Maternal or paternal age (years) Fig. 4. X Y Y .
I
z
1.
5
c. 0)
g 6--
7 --
8 --
9 --
I 30
I
25
I
45
I
Figs. 1-4. Solid lines: estimated relative risk as a function of maternal or paternal age. Broken lines: approximate upper and lower 95 yocoddencelimits. In Figs. 1 and 2, the age limits between d i i c h the curves are shown were chosen so that exactly 20 maternal ages of propositi or of their sibs occurred above the upper limit, and 20 below the lower one. In Fig. 3, paberntl.1 age replaces maternal age in this criterion. I n Fig. 4, the limits were chosen so that exactly 40 paternal or matcrnal ages occurred above the upper limit a n d 40 below the lower one. Tlic: relative risk is defbed to be 1 at 30 years in all cases.
20
15
'.i
z. 2 10
I
,
3 m
0
p
?
Parental age in some sex chromosome aneuploidies
285
Suppose that the population consists of a group of parents who are capable of producing the particular abnormality under consideration and another group who are not. Obviously, our data are derived entirely from the former. Let N p ( i ) ,NR(i)denote the number of children born to mothers of age i from the whole population and from the group at risk respectively, and n(i)denote the number of abnormal children born to mothers of age i. Then the absolute risk at age i in the general population is
44 N,(i) and in the group at risk is
n(i) N,(i) * Hence the risk a t age i relative to that at age j in the general population is
and in the group at risk is
A necessary and sufficient condition for these to be equal is that
Extending this result to all ages leads to the condition that the distribution of maternal age for the group at risk be identical to that for the general population. The same holds for the paternal age distribution. Unfortunately, two factors make it impossible to check whether the condition holds in practice. The first is that a majority of the sibs included in this study were born before 1939, the year in which records of the maternal age distribution in Scotland fist became available. This difficulty might be overcome by using records from particular hospitals or from neighbouring countries, but these are obviously of limited use. The second, more serious, problem is that there is no way of identifying parents ‘at risk’ unless they produce an abnormal child. This may introduce a bias since, if there is a parental age effect, the mean parental ages a t birth of the normal sibs of abnormal propositi may differ from those of the general population. Though these arguments are included as a caution, it must be said that they are purely mathematical. There are no grounds for suspecting that a category of ‘special risk’ parents exists nor that, if it does, it has a parental age distribution sufficiently different from that of the general population to render the estimates of relative risk inapplicable to the latter. Because of these difficulties, it would be of some interest to see whether the results are consistent with those of newborn surveys. If the estimates of relative risk are assumed t o apply to the general population, they can be combined with the distribution of maternal ages a t birth in the general population to predict the mean difference in maternal age a t birth between abnormal and normal individuals. With the X X Y data from Fig. 1 and figures taken from the 1975 report of the Registrar-General for Scotland, the predicted value is between 1 and 5 years. The 56 X X Y’s reported in those newborn surveys referenced above for which control data were
286
A . D . CAROTHERSAND
OTHERS
given had a mean maternal age 0.7 years greater than that of their controls, with an approximate standard error of 0.9 years. The results are therefore compatible, though the estimates are too imprecise to provide a convincing test of consistency. Similar calculations for the X X X ’ s and X Y Y’s are uriinformativc since the corresponding estimates are even less precise. It is also worth noting that the risk curves in Figs. 1 and 2 are consistent with those currently available for Down’s syndrome (see e.g. Penrose & Smith, 19G6, chap. lo), though once again the lack of precision could conceal considerable differences. The newborn surveys suggest a rather flatter curve for the X X Y ’ s than for Down’s syndrome, since the predicted increase in maternal age in the latter case is 4 4 years for present Western populations whereas, as we have seen, the corresponding figure for X X Y ’ s from the newborn surveys is much lower than this. Clearly considerably more data are required in order to improve the estimates to the point where useful comparisons can be made between different abnormalities. As a further check on the validity of our conclusions, we have analysed by similar means the data on a group of 113 propositi with familial chromosome rearrangements, also taken from the Registry of Abnormal Karyotypes. These included translocations, inversions, variants, supernumerary markers and deletions, which were found after ascertainment of the propositus t o be present in a t least one blood relative. The methods of ascertainment were more various than for the sex-chromosome aneuploidies and t o make them more closely comparable we excluded those in which a majority of propositi were less than 2 years old a t ascertainment. The remaining criteria for inclusion were as for the sex chromosome aneuploidies. According to current wisdom, the incidence of such disorders is not expected to depend on parental age, and the results of the analysis support this. No significant improvement on the basic model
fk, Y, 4
= 1
was obtained by adding any terms in x, y or x . The implication is that the variations in risk described in this study are not artefacts of the method of sampling or analysis. We are willing to make our data available t o those who wish to analyse it further.
SUMMARY
The roles of maternal age, paternal age and birth order in the aetiology of the 47,XXY, 47,XXX and 47,XY Y aneuploidies were exanlined using the liveborn full sibs of propositi as controls. As in previous studies, the incidence of XX Y’s and X X X ’ s was found t o be increased a t high parental ages. A small but significant inverse relationship between parental age and the incidence of X Y Y’s was also found. The results were consistent between different methods of ascertainment, different sibship sizes and different years of birth of propositi. The XXY results could not be explained without a maternal age effect and could be explained with a maternal agc effect alone. The X X X and XY Y results could not bc explained without either a maternal or a paternal age effect, but could be explained without a birth order effect. Estimates of the relative risk a t different parental ages are derived and shown to apply to the general population, providcd the distribution of parental ages a t birth is the same for those parents who are at risk (and who may or may not comprise the entire population) as for those who are not. The validity of the method of analysis was confirmed by applying it t o a group of familial chromosome rearrangements which, as expected, revealed no dependence on parental age or birth order.
Parental age in some sex chromosome aneuploidies
287
We are indebted to the members of the Cytogenetics Section of this Unit, and particularly to the technical staff, for karyotype analyses, to Mr J. Davies for computing assistance and to Mr I. Lauder, Mr J. Wood and Professor C. A. B. Smith for several helpful suggestions. Some of the data presented in this study have previously been analysed by Mr J. Wood in an unpublished M.Sc. thesis at the University of Oxford in 1974.
REFERENCES
BELL,A. G. & COREY, P. N. (1974). A sex chromatin and Y-body survey of Toronto newborns. Can. J . Genet. Cytol. 16, 239-50. BORGAONKAR, D. S. & MULES,E. (1970). Comments on patients with sex chromosoxne anenploidy: dematoglyphs, parental ages, X t blood group. J. Med. Genet. 7,345-50. COURT BROWN, W. M., HARNDEN, D. G., JACOBS, P. A., MACLEAN,N. & MANTLE, D. J. (1964). Abnormalities of the sex chromosome complement in man. M.R.C. Special Report Series, no. 305. C o m t ~BROWN,W. M., LAW,P. & SMITH,P. G. (1969). Sex chromosome aneuploidy and parental age. Ann. Hum. Genet. 33, 1-14. FERGUSON-SMITH, M. A., MACK,W. S., ELLIS, P. M., DICKSON,M., SANGER, R. & RACE,R. R . (1964). Parental age and the source of the X chromosome in X X Y Klinefelter’s syndrome. Lancet i, 46. FRIEDRICH, U. & NIELSON,J. (1973). Chromosome studies in 5049 consecutive newborn children. Clin. Genet. 4, 333-43. HADRTON, J. L., CANNING, N., RAY,M. & SMITH,S. (1975). A cytogenetic survey of 14069 newborn infants. 1. Incidence of chromosome abnormalities. Clin. Genet. 8, 223-43. P. A., MELVILLE, M., RATCLIFFE, S., KEAY,A. J. & SYME, J. (1974). A cytogcnetic survey of 11,680 JACOBS, newborn infants. Ann. H u m . Genet. 37,359-76. LENZ,W., NOWAKOWSKI, H., PRADER, A. & S C H ~ R E C. N ,(1959). Die Atiologie des Klinefelter-Syndroms. Ein Beitrag zur Chromosomenpathologie beim Menschen. Schweiz. Med. Wochschr. 89, 727-31. LUBS,H. A. & RUDDLE,F. H. (1970). Chromosomal abnormalities in the human population: estimation of rates based on New Haven newborn study. Science 169, 495-7. MOOD,A. M. & GRAYBILL, F. A. (1963). Introduction to the Theory ofstatistics, 2nd ed. Kogakusha: McGrawHill. PENROSE, L. S. & SMITH,G. F. (1966). Down’s Anomaly. London: Churchill. REGISTRAR-GENERAL SCOTLAND (1975). Annual Report, part 2. Edinburgh : H.M.S.O. SIQLER,A. T., COHEN,B. H., LILIENFIELD, A. M., WESTLAKE, J. E. & HETZNECKER, W. H. (1967). Reproductive and marital experience of parents of children with Down’s syndrome (mongolism). J. Pediat. 70,608-14. TUMBA, A. (1974). L’influence de l’age parental sur la prodnction de l’anomalie X X X X Y . J . Gdndt. H u m . 22(l ) , 73-97.
277
Parental age and birth order in the aetiology of some sex chromosome aneuploidies BY A. D. CAROTHERS, S. COLLYER, R. DE MEY
AND
A. FRACKIEWICZ
M.R.C. Clinical and Population Cytogenetics Unit, Western General Hospital, Edinburgh, EH4 3XU INTRODUCTION
The possibility of an association between maternal age at birth and the incidence of certain sex chromosome aneuploidies is suggested by analogy with the well-established case of Down’s syndrome (Penrose & Smith, 1966). Lenz et al. (1959), Ferguson-Smith et al. (1964), Court Brown, Law & Smith (1969), Borgaonkar & Mules (1970) and Tumba (1974) have all reported an increased maternal age at birth for certain types of sex chromosome abnormality as compared with control groups. A common feature of these studies, as of the present one, is that propositi were ascertained, not through random samples of the general population, but through selected categories such as those attending endocrine, sub-fertility or psychiatric clinics, inmates of hospitals for the mentally sub-normal and so on. From such data a straight comparison of the mean maternal age at birth of the propositi with that of the general population will not establish a causal relationship with the incidence of the abnormality since the samples may be biased in respect of parental age. Surveys of the consecutive newborn are closer to a random sample and several have recently been published (Lubs & Ruddle, 1970; Priedrich & Nielsen, 1973; Bell & Corey, 1974: Jacobs et al. 1974; Hamerton et al. 1975). However, the number of abnormal individuals revealed in this way is rather small and even if the data from all these surveys are combined the evidence on maternal age is inconclusive for even the most frequent abnormalities. We have attempted to avoid these problems by analysing data on a comparatively large number of propositi from selected categories, but with careful attention to possible sources of bias. Our specific objectives were: to demonstrate the presence or absence of a maternal age effect for certain sex chromosome aneuploidies; to show that, if present, the effect is not a result of sampling bias and is not a secondary one arising from the correlation between maternal age and paternal age or between maternal age and birth order; finally, to estimate the size of the effect and the probable limits between which it lies. The particular abnormalities considered are those on which we have the most data, namely, the three aneuploidies with karyotypes 47,XXY, 4 7 , X X X and 47,XYY. For brevity these will be referred to simply as X X Y , XXX and X Y Y. (Some of our data were included in the study by Court Brown et al. (1969) referred to above, and our results are therefore not entirely independent.)
THE DATA
Data for this study were taken from the Registry of Abnormal Karyotypes at this Unit. The Registry and the method of determining karyotypes from cultured blood have been fully described by Court Brown et al. (1964). Propositi were included only if they had a non-mosaic I9
HGE
41
278
A. D. CAROTHERSAND
OTHERS
X X Y , X,YX, or X Y Y karyotype and were ascertained before the end of 1975 in one of the following categories : ( a ) patients a t endocrine and/or subfertility clinics, (b) patients identified a t hospitals for the mentally subnormal or mentally ill, (c) individuals referred from general practices or hospital wards, ( d ) inmates of penal institutions. The method has been t o compare parental ages and birth order of the propositi with those of their liveborn full-sibs, who provide a control for many genetic and environmental factors. Propositi were included only if their own years of birth were known and also those of both their parents and of all their full-sibs born alive before the date of ascertainment. Most dates of birth were verified from birth certificates, and in only a few cases was it necessary to rely on the memories of the propositus or of a near relative. I n 11 cases ( < 1 % of the total), the year of birth of a liveborn full-sib was estimated as halfway between the preceding and following sibs. Where the mother was of child bearing age, defined as less than 45 years, a t the date of ascertainment, an attempt was made to discover her subsequent conceptual history up to the end of 1975, and all further liveborn full-sibs were included in the data. This proved possible for all but 9 X X Y ' s and 4 X Y Y ' s , the main reason for failure being that the mother had emigrated from Scotland. I n fact, for reasons given in the following section, the failure t o trace subsequent liveborn sibs has a negligible effect on the results of the analysis. The numbers of propositi satisfying all criteria are given in Tables 1-3. Birth order was defined as one greater than the number of known previous pregnancies of the mother, with stillbirths and abortions counted and twins counted as two pregnancies. I-9'ince very few of the sibs of propositi have been karyotyped, all were assumed t o be normal. The proportion of abnormals among them is probably too small to affect the conclusions, but this is an unverified assumption. STATISTICAL METHODOLOGY
Let to, x, y, x denote respectively the year of birth, maternal age a t birth, paternal age a t birth and birth order of a particular individual and put -
w
= (PU,
x, y, 2 ) .
Let A denote ascertainment as a propositus in this study, and K the possession of the abnormal karyotype being considered. Then P(AI8) = P(KIV).
P(AIK, 8 ) .
We shall assume that
P(KI8) = 1;,f(x, Y, z ) , where r, is the absolute incidence of the abnormal karyotype among all live-births in ycar and f is some unspecified function, and that
P(AIK
to
= g,,
where g, is the probability of ascertainment of an affected individual born in year w.The first assumption allows for variation in the absolute incidence of the abnormal karyotype from year t o year, but requires that the relative incidence a t different parental ages and birth ordcrs remains the same. The second allows the probability of ascertainment of an affected indiviclual to depend on w,but not on x, y or x . With these assumptions
P(AI@)= g,ru,f(G Y, 2 ) .
Parental age in some sex chromosome aneuploidies
279
Table 1. Summary of results for XXY's
Propositi included in subset
No. in subset
All All All All
234 234 234 234
Ascertainment Ascertainment Ascertainment Ascertainment
A B C D
I22
M* 0
2
4
co 2 2
47 47 18
2
Sibship size 2, 3, 4 or 5 t Sibship size 3 6
127 85
2
Born before 1930 Born 1930 or after
99 I35
2 2
2
2
Estimated rel. risk at mat. age 40 Y S
Lower 95 yo C.L.
Upper 95 yo C.L.
3'97 3.69 3.66 439
2.62 2'47 2.46 3'05
6.02 5.51 5'46 6.3 I
3'40 2'42 5.58 6.80
1.76 1.16 0.72
6.58 5'05 1426 6432
2.96 4'63
1.49 2'74
5.88 7'84
3.76 3'72
2.04 1.87
6.92 7'42
2-18
+
The assumed model isf = exp [a. bz2]. * M defines the smoothing function. See Statistical Methodology section. t Propositi with no full sibs are excluded.
Table 2. Summary of results for XXX's
Propositi included in subset
Estimated rel. risk a t mat. age 40 Yrs
Lower 95 yo C.L.
1.71 1.98 2.06 2.36
I 1.23
aJ
438 4'33 4'52 5.06
2
3'42
1'49
7.87
2
I 1.91
1.03
137.60
20
2
I1
2
16.44 3'29
1.17 I '42
231'17 7.61
I7 26
2 2
4.60 2.80
1.27 0.94
16.73 8'33
No. in subset
43 43 43 43
All All All All Ascertainment B Ascertainment A Ascertainment C Ascertainment D Sibship size 2, 3, 4 or 5 t Sibship size 2 6 Born before 1920 Born 1920 or after
M* 0
2
4
Upper 95 yo C.L.
9.46 9'90 10.86
3\: I 0 I1 OJ
The assumed model is f = exp [ax2]. * M defines the smoothing function. See Statistical Methodolorn section. t Propositi with no full sibs are excluded.
Now consider sibships of n individuals with parameters given by Let The conditional probability, given n and V j ( j = 1,. . ., n ) , that such a sibship is ascertained through the ith sib alone is Pi- Prob (the sibship is ascertained through the ith sib and a t least one other). For rare abnormalities such as those we are considering, the second term can be 19-2
A. D. CAROTHERSAND
280
OTHERS
Table 3. Summary of results for XYY's
No. in subset
Propositi included in subset
03
0.19 0.27
1.27
2
0.3 I
0.06
1'59
2
0.29
0.09
1'00
2 2
0.3 I 0'73
0'10
I1
0'97 2'95
22 22
2 2
0.40
0.09
1.71
0.73
0-16
3'33
20
3, 4 or
Upper 9 5 % C.L.
1-18 0'94
Ascertainment B Ascertainment A Ascertainment C Ascertainment D
>6
Lower
95 % C.L.
0.17 0.17
44 44 44 44
2,
M*
0.45 0.40 0'43 0.59
All All All All
Sibship size Sibship size
Estimated rel. risk at mat. age 40 Yrs
0
2
4
3\
1'00
24
st
Born before 1940 Born 1940 or after
29
0.18
The rtssumed model isf = exp [m]. * M defines the smoothing function. See Statistical Methodology sect,ion. t Propositi with no full sibs are excluded.
ignored. For example, the combined results of the five newborn surveys referred to above suggest that none of these abnormalities has an incidence exceeding one per 1500 live births, and in fact no sibship with multiple ascertainment was obtained in the present study. Hence the probability reduces simply to Pi and, conditioned on there being exactly one propositus in the sibship, becomes I
P'l
n
2 Pj.
j=1
An unbiased estimate of q,,, is where nW = number of propositi born in year to, Nw = total number of livebirths in yenr IU in the catchment area of the study (in this case taken to be Scotland). I n fact the conditional probability is unaffected if g, is multiplied by a constant, showing that it is necessary to asstime only that the number of livebirths in the catchment area is proportional t o the number in the country as a whole. The conditional probability now becomes
&=
,f(xi,Y i , z i )
C hijf(xjr ~ j= 1
,
j zj) ,
where
(nwj/Nu,)/(nW1/~wt). Since in general n, is small and since gw might be expected to change rather slowly with time, it seems reasonable to replace nwl/N,, by a moving average of type hij =
the values of Nw being abstracted from the tables of the Registrar-General for Scotland. We have arbitrarily chosen to set M = 2, corresponding t o a 5-year moving average, but we shall show that within quite wide limits the choice of M has a negligible effect on the conclusions.
Parental age in some sex chromosome aneuploidies
281
The conditional likelihood for the whole data set is given by the product of the conditional probabilities, &, over all sibships in the study. By maximizing this we obtain maximum likelihood estimates of the parameters of any specified model for f ( x , y, z ) together with their approximate asymptotic variances and covariances. Several computer routines are avilable for doing this and the details need not concern us. Any family of models for f is suitable so long as it is both mathematically tractable and sufficiently flexible to approximate any function likely t o be encountered. The family used here is that defined by f(X,
Y, 2)
=
B exp [h(x,Y, 41,
where B is a constant and h a general polynomial. I n fact, B is redundant since it is a common factor of both the numerator and denominator of Q so that without loss of generality we may put B = 1. This indicates that the absolute risk of an abnormality cannot be evaluated by this method. However, the relative risk is estimable and is arbitrarily defined t o be 1 at parental age 30 years and birth order 7. The need to include additional parameters in any model can be assessed using likelihood-ratio tests. For example, with the models, model 1 : h
=
ax,
model 2 : h = ax+bx2 under the null hypothesis that b - 2 log
=
0 , the statistic
[maximum likelihood under model 21 maximum likelihood under model 1
is approximately distributed as x2 with 1 degree of freedom (see, for example, Mood & Graybill, 1963, chap. 12). A ‘best’ model can be defined, though not uniquely, as one satisfying the criteria : (1) That the addition of any further terms in x, y or z does not increase the likelihood significantly. (2) That the removal of any term reduces the likelihood significantly. Constraints of time and expense make it necessary t o confine the search for a ‘best’ model to fairly simple functions - in the present case u p to order 3 or 4 in x , y and x . Since h is a linear function of the model parameters, the derivation of approximate confidence limits for it from the asymptotic variance-covariance matrix is straightforward. These are converted into confidence limits for f by exponentiation. The form of the expression for Q shows why the omission from the data of sibs born after the date of ascertainment has a negligible effect on the results. The value of nwj,and hence of A,,, for these sibs will in general be zero, since nearly all propositi in this study were born before 1959, the first year of ascertainment. Potential sources of bias were investigated by examining various subsets of the data. The full data on each abnormality mere subdivided as follows: (1) By size of sibship - small (2, 3, 4 or 5) and large (6 or more). (2) By method ascertainment - A , B , C and D for the X X Y ’ s ; B (the largest category) and all others ( A , C and D ) for t,he X X X ’ s and XYY’s. (3) By year of birth of propositi - early (before 1930, 1920 and 1940 for the X X Y ’ s , X X X ’ s and X Y Y’s respectively) and late (in or after the stated years).
A. D. CAROTHERSAND
282
OTHERS
Ideally the ‘best’models for each subset would be found separately and the predictions from the models compared, but t o reduce the considerable computing time and expense we have adopted the simpler course of taking the ‘best’ model derived from the full data set and estimating its parameters for each subset. For example, if the ‘best’ model for the full data set is of the form
f = exp [ax+bx2] we obtain the maximum likelihood estimates of a and b for each subset separately. A convenient and easily summarized method of comparing predictions from different subsets is t o est>imate the relative risks a t parental age 40 years, together with 95 yo confidence limits. RESULTS
XXY The ‘best’ model was
f(x, y, z ) = exp [O.O81831(x- 30) +0.004874(.2:- 30)2].
No significant improvement in the likelihood was obtained by adding further terms in 5 , y or z whereas the ‘best’ models involving y only or x only were significantly improved by adding terms in x. I n other words, a model excluding a maternal age effect cannot fully explain the data, whereas a model including only a maternal age effect can. This does not rule out the possibility of paternal age or birth order effects, but shows that they need not be postnlated. The ‘best’ model for f, together with its 95 yo confidence limits, is shown in Fig. 1. Table 1 gives the estimated risk a t maternal age 40 for the various data subsets. It can be seen that the estimates are mutually consistent and, in all but the smallest category (method of ascertainment D),significantly greater than 1. The possibility that the observed increase in risk is confined t o a part,icular subset may therefore be excluded.
xxx The ‘best’ model involving .2: only and that involving y only both gave an equally good fit to the data. However, both gave a significantly better fit than the ‘best ’ model involving z only, and neither was significantly improved by adding terms in x . We conclude that it is not possible to explain the data fully without postulating a maternal and/or paternal age effect,, but that it is possible to do so without postulating a birth order effect. The ‘best’ models were f(x,y, z ) = exp [0*014645(x-30)2] and f(x,y, z ) = exp [O-O09564(y- 30)2]. These, together with their approximate 95% confidence limits, are shown in Figs, 2 and 3 respectively. It is interesting t o note that both models predict an increased risk a t ages below, as well as above, 30 years. Table 2 shows that the estimated risks a t age 40 obtained from the various subsets are mutually consistent and greater than 1. There is therefore no evidence that the observed increase in risk is confined t o any subset.
XYY The ‘bcst ’ model was
f(x, y, z )
=
exp [ - 0.092567(x- 30)],
indicating a small but significant inverse relationship between risk and maternal age (Fig. 4).
Parental age in some sex chromosome aneuploidies 283 It is easily verified by substitution in the expression for Q that x and y are interchangeable in this model, since for any particular sibship x-y = constant. Thus we cannot distinguish paternal from maternal age effects. However, the model gave a significantly better fit than the best one involving z alone and was not significantly improved by adding any terms in z. It is therefore not necessary to postulate any dependence on birth order. Table 3 shows that the reduced risk a t age 40 is common to all subsets of the data.
General Tables 1-3 show that the choice of smoothing function has little effect on the results. I n all cases similar estimates are obtained whether a 1-year or a 9-year moving average is used. It can also be seen that if the differences in the annual probabilities of ascertainment are ignored by puttingM = co (or equivalently hij = l), a small but consistent overestimate of the risk at later parental ages is obtained in all cases.
DISCUSSION
The results agree with previous studies in indicating the presence of an effect of maternal age, or some correlated factor, on the incidence of X X Y’s and X X X ’ s . They go further in suggesting that, of the three highly correlated factors examined, maternal age is the most likely causative factor for the XX Y’s,while birth order is the least likely for theXXX’s. They differ in indicating the presence of a small but significant inverse relationship between parental age and the incidence of X Y Y’s. The effects are apparently independent of the method of ascertainment, the size of sibship and the year of birth of the propositus. However, studies of this nature are so fraught with problems of interpretation that we must consider in more detail a number of possible objections. (1) The propositi in this study are all post-pubertal and one might argue that there could be a different maternal age distribution among newborn and post-pubertal abnormals, such as would arise if the affected offspring of older mothers were more, or less, likely to reach puberty than those of younger mothers. This seems most unlikely since none of these abnormalities is severe and there is no evidence of an appreciable pre-pubertal mortality. (2) The birth of an affected individual may influence the subsequent conceptual history of the mother. I n particular, if the mother is less likely to have further children, affected individuals would occur more frequently a t the end of a sibship and this might lead to an apparent increase in risk a t later maternal ages. Again, it seems unlikely that such an effect is present for these particular abnormalities since they are rarely diagnosed before puberty, by which time most sibships are complete. I n the case of Down’s syndrome, a much more severe abnormality, Sigler et al. (1967) were unable to find any evidence of a difference in fertility, either before or after the birth of an affected individual, between the mothers of children with the disorder and the mothers of an equal number of normal controls, matched for sex, place of birth, date of birth and maternal age at birth. (3) Since the estimates of relative risk given in Figs. 1-4 have been derived from a highly selected group of sibships, a further limitation is the possibility that the results may not apply to parents selected a t random from the general population. The following argument shows under what conditions such a generalization can be made.
I
I
2.
ic
--
2 --
4
6 --
8-
I
35
Fig. 3. X X X .
Paternal age at birth (years)
I
14--
16 --
18 --
30
- 12-%. 2 lo--
v)
- - _- - _
25
I
40
/.,
I
I
I
I II
t
,
I
I I
I
.
I
I
I
20
I
15
3 --
4--
5--
I
I
35
I
40
-----____ Maternal or paternal age (years) Fig. 4. X Y Y .
I
z
1.
5
c. 0)
g 6--
7 --
8 --
9 --
I 30
I
25
I
45
I
Figs. 1-4. Solid lines: estimated relative risk as a function of maternal or paternal age. Broken lines: approximate upper and lower 95 yocoddencelimits. In Figs. 1 and 2, the age limits between d i i c h the curves are shown were chosen so that exactly 20 maternal ages of propositi or of their sibs occurred above the upper limit, and 20 below the lower one. In Fig. 3, paberntl.1 age replaces maternal age in this criterion. I n Fig. 4, the limits were chosen so that exactly 40 paternal or matcrnal ages occurred above the upper limit a n d 40 below the lower one. Tlic: relative risk is defbed to be 1 at 30 years in all cases.
20
15
'.i
z. 2 10
I
,
3 m
0
p
?
Parental age in some sex chromosome aneuploidies
285
Suppose that the population consists of a group of parents who are capable of producing the particular abnormality under consideration and another group who are not. Obviously, our data are derived entirely from the former. Let N p ( i ) ,NR(i)denote the number of children born to mothers of age i from the whole population and from the group at risk respectively, and n(i)denote the number of abnormal children born to mothers of age i. Then the absolute risk at age i in the general population is
44 N,(i) and in the group at risk is
n(i) N,(i) * Hence the risk a t age i relative to that at age j in the general population is
and in the group at risk is
A necessary and sufficient condition for these to be equal is that
Extending this result to all ages leads to the condition that the distribution of maternal age for the group at risk be identical to that for the general population. The same holds for the paternal age distribution. Unfortunately, two factors make it impossible to check whether the condition holds in practice. The first is that a majority of the sibs included in this study were born before 1939, the year in which records of the maternal age distribution in Scotland fist became available. This difficulty might be overcome by using records from particular hospitals or from neighbouring countries, but these are obviously of limited use. The second, more serious, problem is that there is no way of identifying parents ‘at risk’ unless they produce an abnormal child. This may introduce a bias since, if there is a parental age effect, the mean parental ages a t birth of the normal sibs of abnormal propositi may differ from those of the general population. Though these arguments are included as a caution, it must be said that they are purely mathematical. There are no grounds for suspecting that a category of ‘special risk’ parents exists nor that, if it does, it has a parental age distribution sufficiently different from that of the general population to render the estimates of relative risk inapplicable to the latter. Because of these difficulties, it would be of some interest to see whether the results are consistent with those of newborn surveys. If the estimates of relative risk are assumed t o apply to the general population, they can be combined with the distribution of maternal ages a t birth in the general population to predict the mean difference in maternal age a t birth between abnormal and normal individuals. With the X X Y data from Fig. 1 and figures taken from the 1975 report of the Registrar-General for Scotland, the predicted value is between 1 and 5 years. The 56 X X Y’s reported in those newborn surveys referenced above for which control data were
286
A . D . CAROTHERSAND
OTHERS
given had a mean maternal age 0.7 years greater than that of their controls, with an approximate standard error of 0.9 years. The results are therefore compatible, though the estimates are too imprecise to provide a convincing test of consistency. Similar calculations for the X X X ’ s and X Y Y’s are uriinformativc since the corresponding estimates are even less precise. It is also worth noting that the risk curves in Figs. 1 and 2 are consistent with those currently available for Down’s syndrome (see e.g. Penrose & Smith, 19G6, chap. lo), though once again the lack of precision could conceal considerable differences. The newborn surveys suggest a rather flatter curve for the X X Y ’ s than for Down’s syndrome, since the predicted increase in maternal age in the latter case is 4 4 years for present Western populations whereas, as we have seen, the corresponding figure for X X Y ’ s from the newborn surveys is much lower than this. Clearly considerably more data are required in order to improve the estimates to the point where useful comparisons can be made between different abnormalities. As a further check on the validity of our conclusions, we have analysed by similar means the data on a group of 113 propositi with familial chromosome rearrangements, also taken from the Registry of Abnormal Karyotypes. These included translocations, inversions, variants, supernumerary markers and deletions, which were found after ascertainment of the propositus t o be present in a t least one blood relative. The methods of ascertainment were more various than for the sex-chromosome aneuploidies and t o make them more closely comparable we excluded those in which a majority of propositi were less than 2 years old a t ascertainment. The remaining criteria for inclusion were as for the sex chromosome aneuploidies. According to current wisdom, the incidence of such disorders is not expected to depend on parental age, and the results of the analysis support this. No significant improvement on the basic model
fk, Y, 4
= 1
was obtained by adding any terms in x, y or x . The implication is that the variations in risk described in this study are not artefacts of the method of sampling or analysis. We are willing to make our data available t o those who wish to analyse it further.
SUMMARY
The roles of maternal age, paternal age and birth order in the aetiology of the 47,XXY, 47,XXX and 47,XY Y aneuploidies were exanlined using the liveborn full sibs of propositi as controls. As in previous studies, the incidence of XX Y’s and X X X ’ s was found t o be increased a t high parental ages. A small but significant inverse relationship between parental age and the incidence of X Y Y’s was also found. The results were consistent between different methods of ascertainment, different sibship sizes and different years of birth of propositi. The XXY results could not be explained without a maternal age effect and could be explained with a maternal agc effect alone. The X X X and XY Y results could not bc explained without either a maternal or a paternal age effect, but could be explained without a birth order effect. Estimates of the relative risk a t different parental ages are derived and shown to apply to the general population, providcd the distribution of parental ages a t birth is the same for those parents who are at risk (and who may or may not comprise the entire population) as for those who are not. The validity of the method of analysis was confirmed by applying it t o a group of familial chromosome rearrangements which, as expected, revealed no dependence on parental age or birth order.
Parental age in some sex chromosome aneuploidies
287
We are indebted to the members of the Cytogenetics Section of this Unit, and particularly to the technical staff, for karyotype analyses, to Mr J. Davies for computing assistance and to Mr I. Lauder, Mr J. Wood and Professor C. A. B. Smith for several helpful suggestions. Some of the data presented in this study have previously been analysed by Mr J. Wood in an unpublished M.Sc. thesis at the University of Oxford in 1974.
REFERENCES
BELL,A. G. & COREY, P. N. (1974). A sex chromatin and Y-body survey of Toronto newborns. Can. J . Genet. Cytol. 16, 239-50. BORGAONKAR, D. S. & MULES,E. (1970). Comments on patients with sex chromosoxne anenploidy: dematoglyphs, parental ages, X t blood group. J. Med. Genet. 7,345-50. COURT BROWN, W. M., HARNDEN, D. G., JACOBS, P. A., MACLEAN,N. & MANTLE, D. J. (1964). Abnormalities of the sex chromosome complement in man. M.R.C. Special Report Series, no. 305. C o m t ~BROWN,W. M., LAW,P. & SMITH,P. G. (1969). Sex chromosome aneuploidy and parental age. Ann. Hum. Genet. 33, 1-14. FERGUSON-SMITH, M. A., MACK,W. S., ELLIS, P. M., DICKSON,M., SANGER, R. & RACE,R. R . (1964). Parental age and the source of the X chromosome in X X Y Klinefelter’s syndrome. Lancet i, 46. FRIEDRICH, U. & NIELSON,J. (1973). Chromosome studies in 5049 consecutive newborn children. Clin. Genet. 4, 333-43. HADRTON, J. L., CANNING, N., RAY,M. & SMITH,S. (1975). A cytogenetic survey of 14069 newborn infants. 1. Incidence of chromosome abnormalities. Clin. Genet. 8, 223-43. P. A., MELVILLE, M., RATCLIFFE, S., KEAY,A. J. & SYME, J. (1974). A cytogcnetic survey of 11,680 JACOBS, newborn infants. Ann. H u m . Genet. 37,359-76. LENZ,W., NOWAKOWSKI, H., PRADER, A. & S C H ~ R E C. N ,(1959). Die Atiologie des Klinefelter-Syndroms. Ein Beitrag zur Chromosomenpathologie beim Menschen. Schweiz. Med. Wochschr. 89, 727-31. LUBS,H. A. & RUDDLE,F. H. (1970). Chromosomal abnormalities in the human population: estimation of rates based on New Haven newborn study. Science 169, 495-7. MOOD,A. M. & GRAYBILL, F. A. (1963). Introduction to the Theory ofstatistics, 2nd ed. Kogakusha: McGrawHill. PENROSE, L. S. & SMITH,G. F. (1966). Down’s Anomaly. London: Churchill. REGISTRAR-GENERAL SCOTLAND (1975). Annual Report, part 2. Edinburgh : H.M.S.O. SIQLER,A. T., COHEN,B. H., LILIENFIELD, A. M., WESTLAKE, J. E. & HETZNECKER, W. H. (1967). Reproductive and marital experience of parents of children with Down’s syndrome (mongolism). J. Pediat. 70,608-14. TUMBA, A. (1974). L’influence de l’age parental sur la prodnction de l’anomalie X X X X Y . J . Gdndt. H u m . 22(l ) , 73-97.
