Homogeneity of the Age at Diagnosis in Sibs WithType 2 Diabetes: Implications for Sib-Pair Analysis

A. Lev-Ran, 1,2* E. Sprecher, 3 Y. Yerushalmy, 2 B. Schindel, 2 and E.S. Kisch1

1Diabetes Clinics, Maccabi Health Services, Petah-Tikva and Tel-Aviv, Israel2Department of Endocrinology, Zamenhof Clinic, General Sick Fund, Tel-Aviv, Israel3Laboratory for Obesity and Diabetes, Felsenstein Medical Research Center, Rabin Medical Center,Petah-Tikva, Israel

Homogeneity of the age at diagnosis of Type2 diabetes was studied in 1,228 sibs in 300unrelated families: 100 consecutive single-affected and 200 consecutive multiple-affected ones. There were 635 diabetic sibs.The mean and median age at diagnosis in allaffected individuals was 50 years (range, 19–75 years). The mean age at diagnosis in themultiply affected families was 49 years (me-dian 50); the between-sibs range of age atdiagnosis within multiple-affected familieswas (mean and median) 17 years (range,0–55), with 42% of these diagnosed within a5-year age span, 66% within 10 years, and90% within 13 years. When one parent haddiabetes, it was more often the mother (79%P = 0.0023). In order to examine this appar-ent tendency toward homogeneity of age atdiagnosis within families, with full regardfor and parsimonious to right-censoreddata, we employed a Cox proportional-hazards survival analysis, with family as theexplanatory variable. Deviance residualsresulting from that model were analyzed ina variance components, random effectsmodel ANOVA which indicated a significant(P ! 0.001) effect of family on age of diagno-sis, with an intraclass correlation of 0.29. Inmany families the clustering of age at diag-nosis appeared very tight, with single outli-ers, and in 20 families with the longest his-tory, diabetes was diagnosed in 68 sibswithin the span of 8 ± 7 years, whereas 38unaffected sibs remain free of diabetes 25 ±8 years later. The wide differences of the ageat diagnosis between families and its intra-familial homogeneity should be consideredin planning genetic analysis of Type 2 dia-betes. Am. J. Med. Genet. 91:91–94, 2000© 2000 Wiley-Liss, Inc.

KEY WORDS: diabetes diagnosis; diabeticsibs; family studies

INTRODUCTION

Type 2 diabetes is a genetic condition with signifi-cant environmental contribution, which encompasses aheterogeneous group of disorders. Mutations leading todiabetes have been identified only in a very small mi-nority of patients with distinctive syndromes, whichare quite unlike the disease as usually encountered. Asrecently summarized [DeFronzo, 1997], none of morethan 250 genes examined showed sufficient linkageacross different ethnic groups. For example, the geneNIDDM1 (noninsulin-dependent diabetes mellitus,now called Type 2 diabetes) located at 2q37 in MexicanAmericans [Hanis et al., 1996] was not found in Brit-ain, France, or Japan, whereas a quite different candi-date locus was reported in Finland and still anotherone in France. Even more importantly, the NIDDM1gene did not code for any likely candidate enzymes,hormones, or receptors [Hanis et al., 1996], whichmeans that even some basic regulatory mechanismsremain unknown. Beta-cell failure, the possible finalstep to clinical diabetes, is itself genetically pro-grammed [Polonsky et al., 1996].

This work was prompted by our clinical impressionthat in families with multiple affected sibs there wasintrafamily clustering, or homogeneity of age at diag-nosis, as opposed to very wide interfamily differences.The literature pays most attention to the mechanism(s)of heritability, but this particular phenomenon has notbeen adequately described.

SUBJECTS AND METHODS

The total group consisted of 1,228 sibs of 300 unre-lated families. These included 635 affected sibs in allfamilies, which consisted of 200 consecutive multiplyaffected families and 100 additional singly affectedones. All the patients had overt Type 2 diabetes, weretreated with oral drugs (glyburide and/or metformin) or

*Correspondence to: Arye Lev-Ran, P.O. Box 39919, Tel-Aviv61398, Israel. E-mail: levr@internet-zahav.net

Received 9 January 1999; Accepted 30 November 1999

American Journal of Medical Genetics 91:91–94 (2000)

© 2000 Wiley-Liss, Inc.

with insulin, and in the latter case the Type 2 (as dis-tinct from Type 1) nature of diabetes was confirmed bythe measurements of C-peptide.

It is important to stress that we studied the age ofdiagnosis of hyperglycemia, i.e., expression of diabetes,and not of the appearance of any of its basic defects.Such basic defect(s) of diabetes, whether they be insu-lin resistance, hypoinsulinemia, or hyperinsulinemiaprecede hyperglycemia by many years [DeFronzo,1997]. In addition, moderate hyperglycemia is usuallyasymptomatic, and therefore its diagnosis may be de-layed, but this can only negatively affect the parameterunder study, i.e., homogeneity of age at diagnosis.However, as in other countries with universal healthinsurance, in Israel fasting glycemia is a routinescreening assay, and most of adults undergo periodictests, so we can state that in most cases the previousnormal test for glycemia was performed no more than 2years before the first abnormal one. Since we studiedthe age at diagnosis, the chronological age differencebetween the sibs, which was in many cases 15–20years, was irrelevant. On the other hand, in the case ofnondiabetic sibs we were interested in the chrono-logical age now (or at death), since this showed thelatest age at which they were known to remain freeof disease. Although those sibs still alive may developdiabetes in the future, the older they are the less likelyit is.

Verification of the age at diagnosis was done throughthe Sick Funds (there are only four in the country, andthe membership includes the entire population). As anadditional control, in some families of Ashkenazi originour data were confirmed independently by a groupfrom another institution (Prof. B. Glaser, Hebrew Uni-versity Medical School) involved in an unrelated re-search project.

Our analysis proceeded in two stages. The first, sta-tistically less rigorous stage, consisted of a series ofsimple calculations and tabulations, and served to re-inforce initial clinical impressions in a fairly simplemanner. Differences in age at diagnosis between sibswere calculated in two ways: 1) as the absolute differ-ence of the age at diagnosis between the sib affected atthe earliest age and each other; e.g., taking the secondfamily listed in Table I, in which the sibs were affectedat the age 35, 37, 42, and 46, the differences betweenthe one affected at the earliest age and each other were:37–35 4 2, 42–35 4 7, and 46–35 4 11; 2) as thedifferences between the individual age at diagnosis andthe mean age at diagnosis of all the affected sibs in thefamily. In the case of unaffected sibs, we counted onlythose who are already older than the earliest age atonset in the family.

As the foregoing analysis does not take full accountof the data derived from unaffected sibs and is prone toType I error (a false-positive erroneous conclusion), weemployed a second, more complex survival and vari-ance-components-based analysis in order to make moreparsimonious use of the data from unaffected sibs. Wefirst performed a proportional hazards (Cox) survivalanalysis, with family as the predictor variable. Devi-ance residuals, which are martingale residuals trans-

formed to achieve a more normal-shaped distribution[Therneau et al., 1990], desirable for subsequent para-metric analysis, were derived. As these residuals areproduced for all observations, including the right-censored observations (nondiabetic individuals), theymake parsimonious use of all subject data, and can beused in analysis requiring continuous variables [Ther-neau et al., 1990]. The residuals were then employed ina random-effects model ANOVA with family as the sole(random) factor, and intraclass correlations and signifi-cance value were determined [Hopper, 1993].

Statistics were done with SAS and JMP (both bythe SAS Institute, Cary, NC). PROC PHREG ofSAS was used to perform the proportional hazardsanalysis. The random-effects (Satterthwaite) modelingcapacity of JMP was used to perform the ANOVA onthe deviance residuals and to derive variance compo-nents used in calculation of the intraclass correlationcoefficient.

RESULTS

Selected examples from the sample under investiga-tion are reported in Table I and show that the cluster-ing of the age at diagnosis was observed in multiplyaffected families of different ethnic origin. In somefamilies all diabetic sibs were closely clustered,whereas in others (families 7, 13, 15, 26, 28, 32, 34) twoclusters or one cluster with a single outlier were seen.

Our preliminary analyses indicated, in all 300 fami-lies, that the average age at diagnosis of diabetes was50 years (mean and median). In the 200 multiply af-fected families, the mean age at diagnosis in all af-fected sibs was 49 years (median, 50) and ranged from19–75 years. In the 100 singly affected families themean age at diagnosis was 51 years (median, 52) andranged from 24–75 years. However, within multiplyaffected families, the difference between youngest andoldest diagnosed sib averaged 17 years (mean and me-dian); the difference ranged from 0–55 years acrossfamilies. Further, in 42% of the multiply affected fami-lies all diabetic sibs were diagnosed within a 5-yearspan, in 66% within 10 years, and in 90% within a13-year span.

There was a pronounced maternal effect: in familieswith one diabetic parent, the mother was affected in 70cases, and the father in 29, a highly significant differ-ence (by two-tailed Fisher Exact test, P 4 0.0023).There was no significant difference in the sex ratioamong their offspring nor in the proportion of diabeticsamong them: 199/295 for diabetic mothers and 77/126for diabetic fathers (P > 0.2).

As our initial impressions of homogeneity of age ofonset were supported by the initial analysis, we pro-ceeded with the more rigorous statistical method. TheCox proportional hazards survival analysis model con-verged, and this analysis was significant overall by allthree available omnibus tests (−2 log likelihood, Scorestatistic, and Wald chi-square) at P # 0.0001. The ran-dom-effects model ANOVA, based on deviance residu-als, was significant, F(297,924) 4 2.66, P ! 0.0001.The intraclass correlation, derived from variance com-

92 Lev-Ran et al.

ponent estimates, was 0.29. This indicates significanthomogeneity of the age of diagnosis within families.

The argument that the longer the sibs live, the moreof them are likely to develop diabetes, is a valid one.However, in our group the unaffected sibs were 13.7 ±8.6 years older than the earliest age at diagnosis in thefamily, whereas in the group as a whole less than10% of sibs developed diabetes at the correspond-ing age. Therefore, it is likely that only a small minor-ity of unaffected sibs will eventually become diabetic.Especially convincing are the results in 20 families se-lected for their very long-standing disease, in whichthe last sib was diagnosed with diabetes many yearsago. In these families, diabetic sibs (n 4 68) were di-agnosed at an age which was only 8 ± 7 years olderthan the youngest case, while nondiabetic sibs (n 4 38)who were already 25 ± 8 years older, remain free ofdisease.

DISCUSSION

The study shows significant homogeneity of age atdiagnosis of Type 2 diabetes in affected sibs. This phe-nomenon has not been studied in detail and is veryrarely even mentioned. In one of the few studies ad-dressing this problem, in 80 pairs of identical twinswith one twin affected by diabetes, the second twinbecame diabetic within 5 years in 70%, and within 10years in 88% of cases [Barnett et al., 1981]. It is inter-esting to note that in our population of sibs the span ofyears between ages at diagnosis was somewhatgreater: in 42% of families it was not above 5 years andin 66% no more than 10 years. These observations sug-gest that there may be a greater role for genetic asopposed to environmental factors in the final stages ofthe syndrome leading to hyperglycemia. In otherwords, not only the syndrome itself but also the age ofits appearance are largely genetically determined.

TABLE I. Examples of Families With Three or More Sibs Afflicted With Type 2 Diabetes;Homogeneity of the Age at Diagnosis*

Ethnic originDiabetes

in parentsDiabetic sibs,

sex, age at diagnosisNondiabetic siblings,age now or at death

1. Ashkenazi none m70, m71, f75 none2. Ashkenazi mother f35, f37, f42, f46 m70, m843. Ashkenazi father f58, f60, f63 none4. Ashkenazi none f60, f67, f68 none5. Ashkenazi mother f59, f60, f60 none6. Ashkenazi mother m65, m68, m68 none7. Ashkenazi both m31, m35, m40, f50, f51, m64 none8. Ashkenazi mother m46, m50, f51, f52 none9. Ashkenazi both m39, m40, f44 m60, m7010. Ashkenazi none f60, f62, m62, f64 m78, m8211. Ashkenazi none m60, m60, f63 none12. Ashkenazi mother m40, f44, f45 f6313. Ashkenazi mother f50, m50, m55, f75 none14. Bukhara none f48, m50, f51, f52, f52 none15. Bukhara both f25, m26, m28, m35, f45 none16. Bukhara father m35, m37, f41, m45, m45, m46 f46, f56, m6417. Iraq none m69, f70, m75 none18. Iraq father f50, m55, m57 m71, m8019. Iraq mother m30, f40, f42, m42 f45, m57, f5820. Iraq none m55, m58, m66, m69 m73, m75, m8121. Iraq mother m50, m50, f51 none22. Georgia both m35, m36, m37, m37, m39 none23. Israel both f23, f25, m30, m33 none24. Israel both m35, m37, m40 none25. Yemen none m40, m44, m45 f6026. Yemen both f44, m45, m64 m58, m6027. Yemen father m46, f48, f50, m54, f59 f44, f5028. Yemen mother f30, f40, m41, m52, f54, f55, f55 none29. Bulgaria mother f50, f53, m54, f55, f63 none30. Bulgaria both m45, m46, m48, m50, m51, m53, m55 none31. Turkey mother m35, f37, f40, f42, m45 f5532. Morocco both f34, f41, m47, f58, f60 f52, f5833. Morocco none m42, m46, m49, f66 f63, f7534. Iran mother m38, f51, m52, f64 none35. Iran mother m50, f51, f51 m58, m6136. Syria mother m30, f31, m35, m40, m40 m60, f6337. Egypt both m52, m54, f58, f62, f62 m56

*m 4 male (brother), f 4 female (sister). Most European and American Jews are of Ashkenazi origin. Bulgarianand Turkish Jews are mostly of Sephardic (originally, Spanish Jewish) origin; Bukharans are a part of Persian(Iranian) Jewry; North African Jews are partly of Berberic origin (preceding the Arab conquest), partly Sephar-dic; Yemeni Jews are ethnically distinct (probably, ancient converts), and the origin of the Georgian Jews is notclear (may even go back to the Babylonian captivity).

Diagnosis Age in Type 2 Diabetic Sibs 93

This sort of epidemiological study is prone to bothType I (false-positive) and II (false-negative) error. Un-known duration of undiagnosed hyperglycemia and thechances that any sib may develop diabetes even at age90 introduce Type II error, i.e., the failure to diagnosethe existing condition. This error is lessened whenscreening for blood glucose is done regularly and whenthe observations last for many years. In any case, thistype of error can only decrease statistical significanceof the observed phenomenon, and not increase it. Ourstatistical method was designed to decrease chances ofType I error by taking a very conservative approach,namely the inclusion of right-censored cases, evenwhen it was very unlikely that any of the older personswould ever develop diabetes. Additionally, the possibil-ity of multiple clusters of age of onset within a familyserved to introduce instability of our estimates. In spiteof this, we found a highly significant (P < 0.0001) in-traclass correlation coefficient of 0.29. While this cor-relation is not strikingly large, in light of our analyticcaution this must be considered a very conservativeestimate; in light of the statistical significance, it mustbe considered a real effect.

Differences in the age at onset between familiespoint to likely different diabetic syndromes in the popu-lation. This may be seen in the examples presented inTable I. Since Jews are not ethnically homogeneous,their different communities may well have differentforms of diabetes. Both the differences in the age clus-ters between families, and the presence in some ofthem of two age clusters or one cluster with a singleoutlier, may reflect the presence of more than one dia-betic gene in the same family, which is to be expectedgiven the high frequency of the condition [O’Rahilly etal., 1988]. As previously noted, these additional intra-family clusters significantly decrease the intraclasscorrelation we derived.

In two other conditions, not directly related to Type 2diabetes, homogeneity of the age at onset was also re-ported. One is gross obesity, in which age at onset wascorrelated among family members [Price et al., 1998].Another is Type 1 diabetes in which the correlation inthe age at diagnosis was high not only between identi-cal twins (r 4 0.94) but also between nonidenticaltwins (r 4 0.59) and between sibs (r 4 0.53) [Fava etal., 1998]. However, since the right-censored caseswere not taken into account, these correlation valueswere quite likely inflated. In a totally different condi-tion, a genetic form of 4q35-facioscapulohumeral mus-cular dystrophy, age at loss of ambulation was strongly(r 4 0.70) correlated with the D4F104S1 fragment size

[Lunt et al., 1995]. Thus, there is likely to be a geneticbasis for the age at onset of many chronic diseases.

We suggest that the most homogeneous groups forsib-pair analysis are families in which all the affectedsibs develop the condition at about the same age. Oneof the most important aspects of genetic analysis isaccurate phenotyping, and the age at diagnosis is animportant characteristic likely to differ in the forms ofdisease of different genetic background. The most fruit-ful approach of genetic analysis in this case would be touse sib-pairs many years after the appearance of thedisease, when the unaffected sibs are already mucholder than the affected ones were at the time of diag-nosis. This approach may narrow the search for thegenes of chronic diseases, and help to differentiate oth-erwise clinically identical but genetically heteroge-neous conditions, and thus save both effort and re-sources. Type 2 diabetes is an example of the situationwhen “sib pair analysis may not necessarily be themethod of choice, but simply the method of necessity”[Schork and Xu, 1997].

REFERENCESBarnett AH, Eff C, Leslie RDG, Pyke DA. 1981. Diabetes in identical twins.

Diabetologia 20:87–93.DeFronzo RA. 1997. Pathogenesis of type 2 diabetes: metabolic and mo-

lecular implications for identifying diabetes genes. Diabetes Rev 5:177–269.

Fava D, Gardner S, Pyke D, Leslie RDG. 1998. Evidence that the age atdiagnosis of IDDM is genetically determined. Diabetes Care 21:925–929.

Hanis CL, Boerwinkle E, Chakraborty R, Ellsworth DL, Concannon P,Stirling B, Morrison VA, Wapelhorst B, Spielman RS, Gogolin-EwensKJ, Shephard JM, Williams SR, Risch N, Hinds D, Iwasaki N, OgataM, Omori Y, Petzold C, Rietzsch H, Schroder H-E, Schulze J, Cox NJ,Menzel S, Boriraj VV, Chen X, Lim LR, Lindner T, Mereu LE, WangY-Q, Xiang K, Yamagata K, Yang Y, Bell GI. 1996. A genome-widesearch for human non-insulin-dependent (type 2) diabetes genes re-veals a major susceptibility locus on chromosome 2. Nat Genet 13:161–166.

Lunt PW, Jardine PE, Koch MC, Maynard J, Osborn M, Williams M,Harper PS, Upadhyaya M. 1995. Correlation between fragment size atD4F104S1 and age at onset or at wheelchair use, with a possible gen-erational effect, accounts for much phenotypic variation in 4q35-facioscapulohumeral muscular dystrophy (FSHD). Hum Mol Genet 4:951–958.

O’Rahilly S, Wainscoat JS, Turner RC. 1988. Type 2 (non-insulin-dependent) diabetes mellitus. New genetics for old nightmares. Diabe-tologia 31:407–414.

Polonsky KS, Sturis J, Bell GI. 1996. Non-insulin-dependent diabetes mel-litus—a genetically programmed failure of the beta-cell to compensatefor insulin resistance. N Engl J Med 334:777–783.

Price RA, Reed DR, Lee JH. 1998. Obesity related phenotypes in familiesselected for extreme obesity and leanness. Int J Obes Relat MetabDisord 22:406–413.

Schork NJ, Xu X. 1997. Sib pairs versus pedigrees: what are the advan-tages? Diabetes Rev 5:116–122.

94 Lev-Ran et al.