Article Text

Download PDFPDF

Estimating risks of common complex diseases: familial and population risks
  1. K Hemminki1,2,
  2. A Försti1,2,
  3. J L Bermejo1
  1. 1
    Division of Molecular Genetic Epidemiology, German Cancer Research Center (DKFZ), Heidelberg, Germany
  2. 2
    Center for Family and Community Medicine, Karolinska Institute, Huddinge, Sweden
  1. Professor K Hemminki, Division of Molecular Genetic Epidemiology, German Cancer Research Center (DKFZ), Im Neuenheimer Feld 580, D-69120 Heidelberg, Germany; k.hemminki{at}

Statistics from

In a recent article, Lewis et al established a risk estimation model for individuals whose siblings were diagnosed with Crohn disease by incorporating data on CARD15 (NOD2) genotypes, smoking and family history.1 They used Crohn disease as a well selected model for a complex disease,2 which is a timely example in an era when new genetic data on human diseases is filling all media and when discussions abound about individualised genomic medicine.3 4 We would like to consider three separate issues emanating from risk estimation for complex diseases at an individual and population level: reliability of familial risk estimates, use of the low genotype relative risks (GRR) and implications of genomic data at the population level (relating to population attributable fraction, PAF, and familial risk).

An overwhelming proportion of the global literature on familial risks is based on anecdotal information on disease in relatives. There is ample literature illustrating the problem of false reporting of cancers in family members; for some reported cancers less than a half can be confirmed by medical records.5 6 The problem is even worse because the direction of false reporting is often opposite for the cases and the controls, the former overreporting and the latter underreporting, resulting in exaggerated risk estimates. These data are worrisome for cancer, but they may be even worse for many other diseases, for which the diagnostics have not been as uniform as they have been for cancer during the past half century (close to three generations). In some studies great efforts have been undertaken to confirm the diagnoses of the relatives whereby the sample size is often reduced and an uncontrolled selection is introduced. Unfortunately, the weaknesses in familial risk estimates are not generally known nor acknowledged. The sibling risk for Crohn disease, cited by Lewis et al,1 was 27.2, derived as a weighted average of four studies published between years 1989 and 1996. This is in line with sibling risks cited by other experts, such as 20–50 by Schreiber et al2 and 17–35 by the Wellcome Trust Consortium.7 The sibling risk for Crohn disease based on Swedish hospitalised cases is 7.2 (Hemminki et al, unpublished data; for methods see Hemminki et al8), which is in line with the Danish data giving a familial risk of 12.8 for relatively young hospitalised patients whose parents were also diagnosed with Crohn disease.9 Individual risk estimation requires reliable data on familial risks.

Lewis et al1 discussed the risks for Crohn disease in terms of the three independent CARD15 variants and the numbers of risk alleles likely to be inherited. The three risk alleles code for variant transcripts and they are assumed to be functionally related to the risk; the GRR for one risk allele was 2.25 and for two risk alleles it was 9.25 compared to individuals carrying no risk alleles. In the cited meta-analysis even higher GRRs were estimated. For the risk estimation purposes the example was well selected because the risks are high enough and convincingly linked to the allelic variants. Many other susceptibility genes and loci have been identified for Crohn disease. Alone in the Wellcome Trust Consortium study, nine independent association signals, including CARD15, were observed with GRRs ranging for heterozygotes from 1.09 to 1.54 and for homozygotes from 1.55 to 2.32.7 One can calculate that the individual PAFs for these variants ranged from 6.0–22% and that their joint PAF was close to 80%; the methods for calculation10 11 were essentially similar to those used by Lewis et al. In the Wellcome Trust Consortium study, the intronic single nucleotide polymorphism (SNP) rs17221417 tagging CARD15 showed a heterozygote GRR of 1.29 and a homozygote GRR of 1.92, resulting in a PAF of 15%. The nine variants explain a large proportion of Crohn disease aetiology but whether all of them are useful for individual risk estimation and clinical counselling remains doubtful because of the relatively low risks, opposite to the CARD15 example.

Lewis et al calculated that the familial risk conferred by the three functional CARD15 variants was 1.16, implying that the variants explain a very small proportion of the familial aggregation of Crohn disease, even if the sibling risk were 7.2 and not 27.2 or 50 (see above). One can do similar calculations for the Wellcome Trust Consortium data on Crohn disease and come up with a familial risk of 1.03 for SNP rs17221417 tagging CARD15. This figure is lower than 1.16 for the three functional CARD15 variants, apparently because SNP rs17221417 would not tag all the three variants residing on different haplotypes.1 This point illustrates the apparent discrepancy of the current large scale genotyping results which identify variants with large PAFs and tiny familial risks.7 12 13 If the markers used in the genotyping study do not capture the full range of the gene’s phenotypic effects, the calculated genetic parameters are not representative. In the above CARD15 example, a single marker would fail to capture the three variant alleles; thus both the PAF and the familial risk of this gene would be underestimated. However, a common situation in the genome-wide association studies using markers with high allele frequencies could be that the marker is in a complete linkage with a much rarer causative allele. Then the PAF would be identical with the marker and the causative allele but familial risks would be higher for the causative allele and increasing with its relative rarity. We assume that the low familial risks explained by common variants with high PAF are due to the fact that these common variants are markers of much rarer functional variants. Once these rare causative alleles are identified they may be useful for individual risk estimation.


Supported by Deutsche Krebshilfe, the Swedish Cancer Society, The Swedish Council for Working Life and Social Research and the EU, LSHC-CT-2004-503465 and FOOD-CT-2005-016320.



  • Competing interests: None declared.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.