scheme for complex disorders (Baron 1999). The optimal study design for a particular condition is influenced by the underlying genetic model, which is unknown in complex diseases. Consequently, the choice of ascertainment design will be determined primarily by the natural history of the condition under investigation and the available resources (both financial and personnel) rather than theoretical concerns.
Healthy or Unaffected Controls
For some analyses, it is necessary to have control samples to use for comparison with the patient samples. These control samples may include spouses and siblings of affected individuals, classmates, other members of the community, or even untransmitted genetic alleles. Regardless of the relationship of the control sample to the patient sample, one must ensure that the controls are ascertained from the same study population as the patients. Furthermore, the controls can be matched to the patients for confounding factors (any factor that might influence the association between the disease and genotype), such as age, sex, ethnicity, and geographic location. There are two approaches for matching controls to the cases. First, one can select controls such that the overall distribution of cases and controls is comparable with respect to the frequency of the confounders (e.g. for a study of autism spectrum disorders, both cases and controls have a sex ratio of 3 : 1 males to females). This is referred to as frequency or category matching. Alternatively, one or more control individuals may be selected to match each case based on the confounding characteristics (e.g. the case and the control are both African‐American females, eight years of age, and reside in Durham County, North Carolina). This approach is called individual matching. An alternative to matching is to consider these potential confounders in statistical analyses, although this may be a less statistically powerful approach. With the increasing availability of publicly accessible data sets, it has become feasible to utilizing existing controls, so long as there is careful consideration of the potential confounding factors. A landmark study by The Wellcome Trust Case Control Consortium (2007) was the first to robustly demonstrate the use of a common set of controls for identifying genetic factors associated with multiple conditions. Subsequently, it has become commonplace to utilize common, publicly available control samples.
It is important to keep in mind that improper selection of controls can lead to incorrect conclusions. For example, if cases and controls are not appropriately matched on ethnicity and the frequency of alleles for the genetic marker differs by ethnicity, an association study can be doomed. One may falsely conclude an association between a genetic marker and the condition if the “at‐risk” marker allele is more prevalent in the predominant ethnicity of the cases versus the controls (e.g. Knowler et al. 1988). Examples of population stratification and approaches for its detection and control of its effects are described more in Chapter 8.
Ascertainment Bias
In genetic studies, research subjects are selected for participation based on the presence or absence of the trait of interest. The family member who comes to the investigator’s attention (through admission to a hospital or solicitation of support groups, for example) is called the proband. Most often, the proband is an individual who exhibits the trait of interest. Ascertainment through an affected individual can lead to a bias in the distribution of the numbers of affected and unaffected family members present in the analysis. Because the ascertainment scheme necessitated that the family have at least one affected individual (proband), families that may be carrying the genetic liability of interest but, by chance, do not contain an affected family member will not be ascertained. This phenomenon is referred to as ascertainment bias and is demonstrated in Figure 3.2. Depending on the analysis, ascertainment bias may greatly influence the outcome of the analyses.
In general, ascertainment bias should not affect the ability to accept or reject linkage in linkage analysis (Chapter 6). However, it can affect the estimate of the recombination fraction between the genetic marker and the disease locus (Vieland and Hodge 1996). Ascertainment bias can also influence familial recurrence risk ratios (λR) (Guo 1998; Cordell and Olson 2000) and the estimate of the segregation probability of the disease locus in segregation analysis (Ewens and Shute 1986; Greenberg 1986; Stene 1989). Furthermore, it has been argued that in some cases, ascertainment bias may be a reasonable explanation for what appears to be genetic anticipation in some pedigrees (Penrose 1948).
Figure 3.2 Example of ascertainment bias in genetic analysis when ascertaining through an affected individual.
Clearly, ignoring the ascertainment scheme used in an analysis can lead to false conclusions. For example, in the fragile X literature, it had been reported that male offspring of premutation and full‐mutation mothers received larger CGG repeat expansions than did female offspring (Rousseau et al. 1994; Loesch et al. 1995). However, once potential sources of ascertainment bias were removed, it was determined that there was no association between sex of the offspring and the size of the mutation inherited from the mother (Ashley‐Koch et al. 1998). In this example, there were three types of ascertainment bias that were sequentially removed from the analysis. First, in clinically ascertained families, the transmission to the proband was removed. This is a standard ascertainment correction known as the Weinberg correction scheme (Weinberg 1912). Second, transmissions were excluded where not all individuals within a sibship were tested. That is, in fragile X families, nonsymptomatic females are tested more often than nonsymptomatic males because only the females are at risk to have affected offspring. This ascertainment bias increases the number of transmissions to premutation females compared with premutation males. Finally, the transmissions of the CGG repeat to the mothers of the proband were omitted. Because full‐mutation females have reduced fitness (Sherman et al. 1984), mothers of probands are more likely to be premutation carriers merely because they have reproduced. Consequently, including transmissions of the CGG repeat from carrier grandmothers to mothers of probands will also increase the number of transmissions to premutation females. Table 3.1 shows how sequentially correcting for the ascertainment bias in fragile X families changes the conclusion regarding the association between offspring sex and the size of the CGG expansion. Notice that the statistical tests (t‐test and logistic regression) become insignificant as the correction schemes are applied. Additionally, it illustrates that in some cases, standard correction schemes, such as the Weinberg method, are not sufficient to abolish the bias in the ascertainment scheme. Thus, all potential sources of bias in the data set must be carefully considered.
Table 3.1 Association between sex of offspring and risk of expansion in fragile X syndrome.
Source: From Ashley‐Koch et al. (1998); reprinted with permission.
| Proportion of offspring with full mutation (%) | P‐value | |||
|---|---|---|---|---|
| Ascertainment scheme (no. of cases) | Males | Females | t‐test | Logistic regression |
| Removal of cases associated with ascertainment (434) | 0.46 | 0.38 | 0.06 | 0.07 |
| Removal of incompletely ascertained sibships (338) | 0.48 |
0.43
|
