Familial loading, age at onset and outcome

We set out to study the effect of familial loading for schizophrenia on the age at onset and outcome by comparing two extreme groups with each other, one with an extremely high and another with an extremely low familial loading, along with a third group representing the majority of patients with schizophrenia. The study population consisted of all patients with schizophrenia born between 1950 and 1969 with available family information (n=15 733). The study used the familial / sporadic distinction as a research strategy (Lewis et al 1987), but the definitions of both familiality and sporadicity were more stringent than in previous studies (Roy & Crowe 1994).

To estimate the familial loading of the patients we used the familial loading score designed by Pak Sham (Verdoux et al 1996), which takes account of family size and age structure. Familial loading score was based on the following assumptions: the lifetime risk of schizophrenia in a first-degree relative is 10 per cent for familial probands and 0.5 per cent for sporadic probands. The age range at risk is 15 to 50 years, during which

period the risk increases linearly from zero to the lifetime risk. The likelihood ratio of a proband being familial or sporadic, given that a relative of age x is affected, is:

The likelihood ratio if a relative of age x is unaffected is:

Thus, the likelihood ratio for affected relatives is independent of the relative’s age but for unaffected relatives it depends on age at the end of the follow-up period, which in this study was 31 December 1991 or the date of death. The likelihood ratio was calculated for each relative, and an overall likelihood ratio for whether the proband was familial or sporadic was obtained by multiplying together the individual likelihood ratios. Finally, the familial loading score was obtained by taking the logarithm of the product. The natural logarithm was used in this study, while Verdoux et al used the common logarithm. A negative score means that the proband has no affected first-degree relatives. (Verdoux et al 1996)

For each patient, a familial loading score was calculated for schizophrenia (ICD 8 295) and for all psychotic disorders (ICD 8 291-299). The ICD 291-299 categories are alcoholic psychoses, drug psychoses, transient organic psychotic disorders, other organic psychotic conditions, schizophrenic psychoses, affective psychoses, paranoid states, other nonorganic psychoses, and psychoses with origins specific to childhood (WHO 1967). The range of the familial loading scores was -1.0 to 23.8 for all psychotic disorders and -1.0 to 20.7 for schizophrenia. Of the 15 733 patients with available family information, 3427 (22%) had a positive loading score for schizophrenia, i.e., at least one first-degree relative with schizophrenia, and 5489 (35%) had a positive

20

loading score for all psychotic disorders, i.e., at least one first-degree relative with any psychotic disorder.

The patient population was divided into three groups based on their familial loading.

Patients from families with at least three first-degree relatives with schizophrenia were classified as having high familial loading for schizophrenia (loading score for schizophrenia >4.0), and patients with the lowest loading score for all psychotic disorders as having low familial loading for psychosis (loading score for psychoses

<-0.5). All the other patients were classified as having intermediate familial loading.

Hereafter, the group with high familial loading will be called the familial group, the group with low familial loading the sporadic group, and the group with intermediate familial loading the intermediate group.

The familial group consisted of 761 patients, the sporadic group of 725 patients, and the intermediate group of 14 247 patients. The proportion of males and females in each group did not differ significantly (χ²=5.48, d.f.=2, P=0.065). The mean number of siblings of patients was 5.7 in the familial group, 7.4 in the sporadic group, and 3.5 in the intermediate group. The proportion of patients born in the 1960s was smallest in the sporadic group and largest in the intermediate group (χ²=99.8, d.f.= 2, P<0.001) and because of this, year of birth was adjusted for in the analyses. The cases in the familial group came from families with an average of 3.3 affected first-degree relatives.

Outcome was measured by the annual duration of hospitalisation, risk of receiving a disability retirement pension, and mortality. We calculated the annual duration of hospitalisation by dividing the number of days spent in hospital during the follow-up period by the number of follow-up years.

The age at onset and the duration of hospitalisation were modelled with linear mixed models, where sex, category of familial loading, and an indicator variable for being the first to develop schizophrenia in the family were used as fixed explanatory variables, and family was used as a random effect to account for the correlation between siblings (Laird & Ware 1982). The variable "first to develop schizophrenia in the family" was

included to control for the possibility that patients already having an affected family member might have a shorter interval between the occurrence of psychotic symptoms and first hospitalisation.Year of birth was controlled for in the analyses. Because early onset is associated with higher risk of rehospitalisation (Eaton et al 1992), age at onset was used as an explanatory variable in the model for the duration of hospitalisation.

Interaction between sex and familial loading was included when significant.

The risk of receiving a disability retirement pension, and mortality were modelled with the Cox proportional hazard model (Cox & Oakes 1994). Follow-up was from the onset of the disease until the event in question, the censoring date of these analyses being the end of 1991. In these models age at onset, sex, and category of familial loading were used as explanatory variables.

The level of significance in the models was determined by the Wald χ²-test.