Caught in the trio trap? Potential selection bias inherent to association studies usings parent-offspring trios

The implementation of the classic case control study design has frequently been criticized because of the potential for spurious associations resulting from population stratification, that is, the existence of multiple population subtypes in what is assumed to be a relatively homogenous population [Pericak-Vance, 1998]. Recently, Bruun and Ewald [1999] addressed the question of whether the tendency in psychiatric-genetic research to prefer family-based (parent-offspring trios) to case control association studies introduces a nonnegligible selection bias. They hypothesized that these two approaches might create patient samples that not only differ on demographic factors but also on genetically influenced phenotypic characteristics. Such a bias would make it difficult to compare results from case control and family-based association samples. To corroborate their hypothesis, Bruun and Ewald [1999] compared patient characteristics of 61 bipolar patients from a case control sample with 19 bipolar patients derived from parent-offspring trios. They observed a lower median age for the trio group, but found no significant differences in sex distribution, suicidal behavior, or family history of affective disorder.

The lack of significant differences may be a result of the small sample size, which may have prevented detection of effects if present. Since we found the theoretical arguments raised by Bruun and Ewald [1999] very convincing, we addressed similar questions in a large German sample of 176 patients with bipolar disorder. All patients were systematically recruited within the framework of our ongoing studies on the genetics of bipolar affective disorder. Our standard recruitment and phenotype characterization strategy is outlined below.

Recruitment is performed at the University Hospital and the Mental State Hospital, both serving the city of Bonn and its vicinity. Hospital admissions are screened for bipolar disorder on a continuous basis. Subsequently, all potentially suitable patients are contacted and asked for their participation in the study. Upon the patient's consent, parents are informed about the study and asked for their participation. Whenever parents assent to participation, they will be integrated into our recruitment program. Consensus best estimates of lifetime DSM-IV diagnosis are achieved on the basis of a multidimensional phenotype characterization inventory, including a personal structured interview (SCID-I) [Spitzer et al., 1990], OPCRIT documentation [McGuffin et al., 1991], family history (Family Informant Schedule and Criteria) [Mannuzza et al., 1985], and a systematic review of medical records.

From the 176 patients, 122 (105 BP I disorder, 17 BP II disorder) were singletons and 54 (50 BP I disorder, 4 BP II disorder) had two living parents who participated in the study. Both groups were compared with respect to the following factors: sex, age at interview, age of onset (time when criteria for diagnosis were met for the first time), and a history of suicide attempts. In addition, we assessed for a family history of both affective disorders and suicidal behavior in first- and second-degree relatives. Written informed consent was obtained from all participants.

With respect to demographic data, in the singleton group female gender proved to be overrepresented (P = 0.022) and mean age at interview was higher (P < 0.0001) than in the trio group. As regards clinical characteristics, patients from the singleton group showed a higher mean age of onset (P = 0.0001) and higher frequency of suicidal behavior (P = 0.001). The importance of this finding is further strengthened by the observation that first-degree relatives of singleton patients also showed a trend for a higher rate of suicidal behavior than first-degree relatives of trio patients (P = 0.056; Table I)

Given the observation that females display parasuicide or deliberate self-harm more often than males [Morgan et al., 1975; Platt et al., 1988], it may be argued that our finding on suicidality could be influenced by the overrepresentation of females in the singleton group. Likewise, the higher mean age at interview among the singleton group compared to the trio sample, and thus longer risk period of these patients, might also bias the finding on suicidality. Therefore, we performed a stepwise logistic regression analysis with the outcome variable “suicidality” and the explanatory variables “trio or singleton,” “sex,” and “age at interview,” as well as “duration of disease” (i.e., time between age of onset and age at interview) that may mirror the risk period for suicide attempts more precisely than mere age at interview. Eventually, the logistic repsiession analysis yielded a result that strengthened our finding: the variable “trio or singleton” proved to be the only significant predictor for suicidality, with a higher risk for singleton patients (OR = 3.56; CI = 1.59–7.97; P = 0.002).

Our findings that there are significant differences with respect to important clinical characteristics (age of onset, suicidality) between index patients from a singleton sample and those from a trio sample support the hypothesis by Bruun and Ewald [1999] that recruitment strategies relying on trios solely will introduce a selection bias of susceptibility genes.

Patients with a late age of onset are more likely to be missed in trio-based recruitment designs, since living parents will be available less often. There is rising evidence for genetic factors being involved in the age of onset in both unipolar [Weissman et al., 1984; Kendler et al., 1992] and bipolar [McMahon et al., 1994; Leboyer et al., 1998; Schürhoff et al., 2000] affective disorders. Furthermore, age of onset has been suggested to serve as a delineator for distinct subtypes of bipolar disorder [Schürhoff et al., 2000; Schulze et al., 2001]. Thus, valuable information might get lost when concentrating on trios exclusively.

The most interesting finding, however, is the overrepresentation of suicide attempts among singletons. Taking into consideration that there is strong evidence for the heritability of suicidality [Tsuang, 1983; Wender et al., 1986; Roy et al., 1991], the increased frequency of suicidal behavior in both patients and their first-degree relatives could be due to an underlying genetic mechanism of this phenotypic characteristic that might rather be found in case control samples than in trios.

Recently, new approaches in the field of association mapping that control for stratification effects, such as genomic control [Devlin and Roeder, 1999; Bacanu et al., 2000] or the method proposed by Pritchard et al. [2000], have been viewed as promising tools. Thus, the implementation of case control designs can be reconsidered a valid approach for association mapping, provided the aforementioned procedures are applied.

In conclusion, our data suggest that giving up case control association analyses for the sake of within-family analyses could be at the risk of a “trio trap” selecting against several genetically determined factors.