Since this question is the easiest to answer it frequently is the first to be asked. For example, observations of disorders "running in families'' may come from clinicians
TABLE 1. Chain of Psychiatric Genetic Research
Is disorder familial?
What are the relative contributions of genes and environment? What is the mode of transmission? Where is the gene (or genes) located?
Methods Family study
Twin and adoption studies
Segregation analysis Linkage and association studies
1 Genetic and statistical terms are defined in the Appendix to this chapter.
who often treat patients from the same family. Of course, once familiality is informally established in a clinical setting, it remains to be confirmed with a rigorous research design, known as the family study method.
Selection of Probands. Ideally, a family study should use the blind case-control paradigm, a staple of epidemiology and behavioral science. The cases and controls used in genetic studies are known as probands. We usually select probands with the disorder from a source that is "enriched" with the diagnosis of interest. For example, patients in psychiatric clinics are more likely to have bipolar disorder than patients in a family practice clinic. Furthermore, patients in a bipolar specialty clinic are more likely to have bipolar disorder than patients in a general psychiatric clinic.
Selection from clinics instead of the general population is useful for two reasons. First, to achieve an adequate number of cases from the general population we would need to screen many individuals. This is costly and of dubious benefit. Second, multiple stages of ascertainment increase the probabilities of ill probands being "true cases'' and of normal probands not having the disorder under study. Individuals who seek treatment and are given a clinical diagnosis are more likely to have experienced the level of distress and disability that the diagnostic nosology requires for psychiatric illness. This combined with the fact that individuals free of illness are rarely referred means that psychiatric clinic populations have a higher base rate of all psychiatric illnesses than does the general population.
The positive predictive power of a diagnosis (the proportion of those with the disorder among all patients receiving the diagnosis) increases with increases in the base rate of the disorder being diagnosed (Meehl and Rosen, 1955). Thus, multiple stage ascertainment increases the positive predictive power by using clinic status to increase the proportion of "true cases'' in the sample that is assessed by the research protocol. Unavoidably, this method of increasing the positive predictive power will increase the false negative rate. In this context, false negatives are those who have a disorder but are (1) not referred to a clinic or (2) referred but do not receive a clinical diagnosis. Thus, the generalizability of results will be limited to the degree that these false negatives differ from the probands enrolled in our study. Treatment is, of course, the most notable factor that will differentiate these groups.
Likewise, multistage screening of controls decreases the probability of misclassi-fying someone with the disorder as a control. Of course, since screened controls are selected for absence of the disorder of interest, they cannot be considered representative of the general population. However, work in psychiatric epidemiology indicates that screened controls are very effective when the goal of a project is to delineate factors that differentiate controls from cases (Tsuang et al., 1988). Furthermore, unscreened controls frequently have rates of psychopathology and its correlates that are above the population expectation (Gibbons et al., 1990; Kruesi et al., 1990; Shtasel et al., 1991). Thus, unscreened controls are often heavily contaminated with cases, thereby obscuring the effects of the variable of interest.
We emphasize that controls should be screened only for the disorder being studied, not for other psychiatric disorders or conditions. When controls are screened for additional disorders, the results can spuriously indicate a familial relationship between the disorder used to select cases and the disorders that were screened from controls (Kendler, 1990). For example, we know that alcoholism and anxiety disorders both run in families. Consider a family study of alcoholism that screens control, but not alcoholic, probands for anxiety disorders. Since anxiety is familial, the rates of anxiety among relatives of controls will be decreased by the screening process. In contrast, the rates in relatives of alcoholics will not be decreased. Thus, anxiety disorders will be more prevalent among the relatives of alcoholics due to the choice of control group.
The selection of controls should satisfy the comparability principles required for meaningful inferences in case-control epidemiological studies (Miettinen, 1985; Wacholder, 1992 a - c). It is usually not possible to establish a primary study base with a geographically defined population. This is so because the clinics from which probands are selected may serve a broad geographic region that is difficult to delineate. This is especially true for specialty clinics at universities. In many cases the reputation of the clinic attracts patients from great distances.
The usual approach is to establish a secondary study base defined by the ascertainment source. The use of secondary study bases limits generalizability and does not produce a representative sample from a geographical population. Nevertheless, it does allow for meaningful case - control comparisons if the controls are individuals who could have been cases had they developed the disorder of interest during the time of investigation (Miettinen, 1985; Wacholder, 1992 a-c). When sampling from a clinic, this requires that if the control subjects had needed treatment for the disorder, they would have been referred to the clinics that provided the case probands. For example, in a general hospital outpatient setting, it is likely that patients who seek treatment for medical disorders in medical clinics would go to the same hospital's psychiatric clinic for the treatment of a psychiatric disorder.
Of course, instead of establishing a secondary study base, it is possible to match cases and controls on "relevant" variables. One problem here is defining what is and is not a "relevant" variable. Age, sex and socioeconomic status are usually considered, but others may be appropriate. However, matching should be used cautiously so as to avoid the "matching fallacy'' (Meehl, 1970) and "overmatching" (Miettinen, 1985; Greenland and Morgenstern, 1990). As discussed by Meehl (1970), matching on specific variables often unmatches on others. In addition to creating unusual samples, this also leads to reduced statistical efficiency and biased estimates (Wacholder et al., 1992c). These problems are most severe when the matching variable is strongly associated with the disorder under study.
An obvious example of the matching fallacy is as follows. Numerous studies find that attention deficit hyperactivity disorder (ADHD) interferes with school achievement (Faraone et al., 1993). Thus, matching controls to ADHD subjects on school achievement would create an unusually high functioning ADHD sample or an unusually low functioning control sample. It may be difficult to draw meaningful inferences from such samples. Instead of matching, we use statistical methods to examine and control for the effects of potentially confounding variables.
Following the selection of cases and controls, the study attempts to assess the diagnostic status of as many of the relatives of cases and controls as possible. The aim is to compare rates of illness in relatives of cases to rates in the relatives of controls. To estimate these rates of illness accurately, care must be taken to assess as many relatives as possible. However, psychiatric disorders affect emotions, thinking, and interpersonal relationships. Thus, nonparticipation may not be random with respect to illness status; family members who are ill may be more likely to refuse participation than those who are well. Paranoid schizophrenia provides a good example of this problem. Paranoia leads to distrusts of strangers, friends, and family. This makes it difficult for a paranoid person to agree to answer the many questions required by psychiatric interviews.
If a disorder has a genetic etiology, then relatives of ill probands should carry a greater risk for the illness than relatives of controls. In addition, the risk to relatives of probands should be correlated with their degree of relationship to the proband, or the amount of genes they share in common. First-degree relatives such as parents, siblings, and children, share 50% of their genes, on average, with the proband. They should be at greater risk for the disorder than second-degree relatives (grandparents, uncles, aunts, nephews, nieces, and half-siblings) because second-degree relatives share only 25% of their genes with the proband.
A genetic hypothesis predicts that the risk to relatives of ill probands is higher than that for relatives of controls and that the risk to relatives of probands increases as the amount of genes shared increases. In practice, however, it is rare that a family study will have the resources to diagnose second- or third-degree relatives. Most studies assess only first-degree relatives. Table 2 displays the familial pattern of risk found in the families of schizophrenic probands. These risk figures come from many of the earlier European family studies and conform to the
TABLE 2. Rates of Schizophrenia Among Relatives of Schizophrenic Patients
Type of Relative
Percent at Risk
First-degree relatives Parents Children
Both parents schizophrenic Brothers and Sisters
Neither parent schizophrenic
One parent schizophrenic Fraternal twins of opposite sex Fraternal twins of same sex Identical Twins
Second-degree relatives Uncles and Aunts Nephews and Nieces Grandchildren Half brothers or half sisters
Third-degree relatives First cousins General population
Source: Based on Slater and Cowie (1971) with the exception of twin data from Shields and Slater (1975). Adapted, with permission from Tsuang, Faraone, and Johnson (1997)
expectation that first-degree relatives are at highest risk, followed by second- and then third-degree relatives.
Family Study versus Family History. In planning a family study of psychiatric illness we must choose between two approaches for the evaluation of family members: the family history and family study methods (Faraone et al., 1999). The family history method collects diagnostic information about all family members by interviewing only one or several informants per family. This method uses a specialized instrument such as the interview for Family History Research Diagnostic Criteria (FH-RDC; (Andreasen et al.,1977)) or the Family Interview for Genetic Studies (FIGS; (NIMH Genetic Initiative, 1992)).
In contrast, the family study method determines diagnoses by interviewing all family members directly. Several excellent structured psychiatric interviews are available but only one was designed specifically for genetic studies: the Diagnostic Interview for Genetic Studies (DIGS; Faraone et al., 1996; Nurnberger et al., 1994).
An obvious advantage of the family history method is its low cost; interviewing a few family members is less costly then interviewing all family members. However, several researchers have shown that family history data underestimate true rates of many psychiatric disorders. The overall strategy of these studies has been to collect family history data on the same subjects who have also been diagnosed by the family study method. By using the family study method as the ''gold standard'' they can estimate the accuracy of the family history method.
Mendlewicz et al. (1975) examined the accuracy of the family history method in the context of a family study of mood disorders. The probands were 140 patients with either bipolar disorder or major depressive disorder. When the probands were used as informants for the family history method, the rates of mood disorders in the family were underestimated. The family history method was most accurate when the informant was the child or spouse of the person being diagnosed.
Similar results were reported by Andreasen et al. (1977). They determined the specificity and sensitivity of family history diagnoses using an adaptation of the Research Diagnostic Criteria they termed the Family History Research Diagnostic Criteria. Specificity is the probability of correctly classifying a subject as well by the family history method if that subject is assessed as well by the family study method. Sensitivity is the probability that the subject is diagnosed ill by family history if they are diagnosed ill by the family study method. In their application to mood disorders, the specificity was high but the sensitivity was low. For example, among family members classified as having a mood disorder by direct interview, only 59% were so classified by the family history method.
It is possible to improve the sensitivity of the family history method by using several informants to provide information about the subject being diagnosed. In the Andreasen et al. (1977) study, when only probands were interviewed via the FH-RDC, the lifetime prevalence of mood disorders among relative's was 11%. This rate increased to 17% when other relatives were interviewed along with the proband. However, both rates underestimated the 25% rate obtained by the family study method.
Orvaschel et al. (1982) found that the family history method was better for some types of relatives than for others. For example, in their study, the sensitivity of the family history method was lower when the subject being diagnosed was male. Sensitivity was higher if the relative being diagnosed was ill at the time of the family history interview. This latter finding makes intuitive sense; we are more likely to know of a relative's problems if, for example, they are in a psychiatric hospital at the time we are asked about their condition.
Other studies have shown that the accuracy of family history assessments varies by diagnosis. Thompson et al. (1982) found that sensitivities for major depression and alcoholism were much higher than for generalized anxiety, drug abuse, phobic disorder, and depressive personality. Moreover, diagnoses based on spouse or offspring reports were more sensitive than those based on parent or sibling reports.
In the family study of mood disorders by Gershon and Guroff (1984), the family history method was most sensitive (96%) when the informant was being asked about a proband. It was halved to 48% when the relative being diagnosed was not a proband. Notably, there was a positive linear relationship between sensitivity and the number of informants. The sensitivity was only 15% with one informant, but increased to 67% when five informants were used. Specificity decreased only a little, from 99% to 92%, when the number of informants was increased from one to five.
In a relatively large study of 609 mood-disordered probands and 2,216 first-degree relatives, Andreasen et al. (1986) confirmed the results from previous validity studies of the family history method. Relative to the family study method, rates of illness in relatives were always underestimated by the family history method using the FH-RDC. One key exception was the diagnosis of antisocial personality: The family history rate of this disorder was three times greater than the direct interview estimate. Thus, it may be that the family history method is more valid than direct interview when the disorder in question has a pejorative connotation.
In Andreasen et al.'s (1986) study the sensitivities and specificities of the family history method were consistent with previous reports. The sensitivities were low. They ranged from 31% for schizophrenia to 69% for "psychotic disorder.'' As expected, the specificities were higher. These ranged from 84% for probable depressive disorder to 100% for schizophrenia and schizoaffective disorder. The sensitivity of the family history method was best when the informant was a parent of the subject being asked about. For depression the sensitivities were 62% for parent informants, 51% for sibling informants and 37% for child informants.
Ideally, the diagnoses of subjects should use three sources of information: direct interviews with the subject, family history interviews with informants who are familiar with the subject, and medical records when available. All sources of information about a given individual, are then combined into a consensus diagnosis (Gershon and Guroff, 1984; Leckman et al., 1982). As suggested by the research reviewed previously, the direct interview and medical record usually provide more useful information than the family history assessment. In fact, two studies find that diagnoses based on direct interviews alone closely approximate best estimate diagnoses (Gershon and Guroff, 1984; Leckman et al., 1982). However, a diagnosis based only on medical records is often a suitable proxy to the best estimate diagnosis (Gershon and Guroff, 1984).
Silverman and colleagues (1986) evaluated the reliability of the family history method for dementing illnesses such as Alzheimers disease. When rating the same individual, different informants had high levels of agreement on the presence of dementia and its age at onset. The rates of dementia found in this family history study were similar to what had been found in previous family studies using direct methods of assessment. The authors concluded that multiple informants would likely increase the validity of the family history method but, because they did not directly evaluate relatives, inferences about validity were limited.
Kosten, Anton, and Rounsaville (1992) examined the validity of the family history method for five diagnoses used in a family study of opiate addiction: depression, anxiety, antisocial personality, alcoholism, and drug abuse. For diagnosing family members, the sensitivities were uniformly low, ranging from 6% to 39%. Specificities were greater than 95%, with the exception of depression, which had a specificity of 54%.
The authors also provided data about the type and number of informants. Spouses and children were better informants than parents or siblings. For alcoholism and drug abuse, females were better informants than males. Increasing the number of informants improved the accuracy of the family history method for depression, antisocial personality, and alcoholism, but not for anxiety disorders or drug abuse.
Collection of data using the family history method may be influenced by the presence of psychiatric illness in the informant. Kendler and colleagues (1991) asked discordant twins about depression, anxiety, and alcoholism in their parents. Twins are discordant if only one has the disease being studied. Compared with the unaffected twin, those with a history of major depression or generalized anxiety were more likely to report the same disorder in their parents. This effect was not observed for alcoholism. However, since direct interview data had not been collected on the parents, it is not certain if the affected twin was over reporting psychopathology or the unaffected twin was underreporting it. Nevertheless, this study underscores the need for more methodological work to clarify the utility of the family history method for specific disorders.
Table 3 outlines the advantages and disadvantages of the family history and family study methods. The choice between the two requires a trade-off between data quality and the expense of data collection. The family history method is the method of choice when there are not sufficient data to justify the expense of a family study. Thus, it is a good choice for initial pilot phases of a genetic investigation. However, after the family history method demonstrates familiality, the family study is the tool of choice for examining the details of familial transmission.
If the question at hand dictates the use of the family history method, the following recommendations should be considered: (1) use the FH-RDC, FIGS or some other semistructured method for eliciting the family history; (2) because the family history method has low sensitivity, use less stringent diagnostic criteria than
High quality data
that used for direct interview data; (3) use multiple informants for each person to be diagnosed; (4) seek out informants who have had substantial contact with the person to be diagnosed; (5) remember that the method is most valid when the person to be diagnosed is ill at the time of interviewing the informant. These "rules of thumb'' provide a rough guide for planning a family history study. The papers discussed in this chapter should be consulted for information about specific disorders.
Caveats. The family study is a practical and robust tool for psychiatric genetics. In many cases, it has provided the initial hint that a disorder might have a genetic component. However, we must be cautious in concluding that a disorder is caused by genes after we observe that it is familial. Disorders can "run in families'' for nongenetic reasons such as shared environmental adversity, viral transmission, and social learning. Also, since the culture and environment shared by family members tends to increase as the degree of relationship decreases, the pattern of risk due to environmental factors may mimic the pattern expected for genetic relationships.
Family data on tuberculosis provide a good example of such confounding. Data collected in the 1940s showed that the risk to family members related to probands who had tuberculosis increased with the degree of genetic relationship. That is, first-degree relatives had higher rates of tuberculosis than second-degree relatives. McGue et al., (1985) pointed out that the distribution of familial risk for schizophrenia and that for tuberculosis both showed that the risk to relatives of ill probands increased with the amount of genes shared with the proband. However, the patterns differed in a few subtle ways that allowed them to show that the risk for schizophrenia was transmitted genetically whereas tuberculosis was transmitted through the environment.
Our point is straightforward: The finding of familial transmission cannot be unambiguously interpreted. Although family studies are indispensable for establishing the familiality of disorders, they cannot, by themselves, establish what type of transmission. All mechanisms that could lead to a familial clustering of disease should be considered.
Was this article helpful?