H567/03 Section A - Gottesman et al. (2010): family risk

The question and the national cohort

Gottesman, Laursen, Bertelsen and Mortensen's 2010 study asked how risk of psychiatric diagnosis in offspring varied with diagnoses in their biological parents.

Earlier studies of couples with psychiatric disorders often had small clinical samples. A nationwide register offered a way to identify more families and follow records over time. Studying offspring with two affected parents was described as a super-high-risk design, compared with one affected parent or population comparison groups. The term describes the sampling question; it does not say that every offspring will develop a disorder.

This was a national register-based cohort study in Denmark, using existing administrative data rather than manipulating parents' illness. The Danish Civil Registration System supplied unique identification numbers and links between biological parents and offspring. The Psychiatric Central Register supplied recorded treatment diagnoses. Linkage enabled the researchers to follow family relationships and diagnosis histories without interviewing millions of people.

The initial civil-register population included 3,391,018 people born in Denmark, alive in 1968 or born later, with links to both parents. Restricting the investigation to those aged ten or over before 1 January 2007 left 2,685,301 people and their parents, representing 1,278,977 unique parent couples. Thus “about 2.7 million” describes the main cohort, not the number with two affected parents.

Complete psychiatric inpatient registration dated from 1 April 1970; outpatient records were included from 1995. Denmark used ICD-8 through 1993 and ICD-10 from January 1994 in these records. These dates belong to Gottesman's dataset, unlike Rosenhan's earlier US study. Different classification periods require attention when interpreting long-term comparability.

Diagnostic groups and cumulative incidence

The researchers grouped families by recorded parental diagnoses, then examined offspring diagnoses up to follow-up.

They identified parents ever admitted to psychiatric facilities with schizophrenia, bipolar affective disorder or unipolar depressive disorder between 1 April 1970 and 1 January 2007. Parents and offspring were classified using discharge diagnoses, recording the first occurrence of each specified diagnosis. ICD-8 and ICD-10 codes defined the disorder groups. The oldest offspring reached age 52 at follow-up; not every person was observed to that age.

Comparisons included two parents with the same disorder, one parent with that disorder and the other never admitted, and neither parent ever admitted. A further group had one parent with schizophrenia and the other with bipolar disorder. An unrestricted population comparison included all parental diagnostic histories. The neither-admitted comparison is different from the unrestricted general population and must be labelled correctly.

The outcome was cumulative incidence by age 52: an estimated probability of having received a recorded diagnosis by that age, based on age-specific incidence during the observed follow-up. It accommodates the fact that offspring differ in age and observation time. It is not a raw percentage obtained by dividing the listed number of diagnosed offspring by the whole listed group. The paper explicitly says the cumulative incidence cannot be calculated from the raw sample sizes alone.

For example, 26 of the 270 offspring of two parents with schizophrenia had a recorded schizophrenia diagnosis, yet the estimated cumulative incidence by 52 was 27.3%. The raw fraction $26 \div 270 \times 100$, about 9.63%, answers a different question and must not replace the reported estimate. The age-adjusted estimate also is not guaranteed final lifetime risk: bipolar incidence was still increasing at age 52.

The groups were not entirely mutually exclusive. A person could receive different diagnoses across admissions and appear in more than one diagnostic group; an offspring could later be a parent. Therefore, adding all diagnostic groups would not necessarily count distinct people. The original tables also use diagnosis-specific denominators; preserving their labels is more accurate than forcing every result into one generic total.

The actual family-risk patterns

Both schizophrenia and bipolar results showed higher estimated offspring risk where both parents had the corresponding diagnosis.

These are the paper's cumulative-incidence estimates and sample sizes, not percentages derived directly from those sample sizes. The two-parent schizophrenia estimate rose to 39.2% when schizophrenia-related disorders were included. The two-parent bipolar estimate rose to 36.0% when unipolar depressive disorder was included. Broadening the outcome changes what “risk” means; these wider figures must not be described as risk of the narrow disorder alone.

For parents with one schizophrenia and one bipolar diagnosis, offspring risks were 15.6% for schizophrenia and 11.7% for bipolar disorder. These cross-disorder patterns raised questions about overlap in susceptibility, but small numbers limited precision. They do not demonstrate one identical gene for both disorders.

For any psychiatric diagnosis, the estimates were 67.5% for offspring of two parents with schizophrenia and 44.2% for offspring of two parents with bipolar disorder. In the unrestricted population comparison, the estimate was 14.1%, whereas the neither-parent-admitted comparison was 11.9%. “Any diagnosis” is broader than schizophrenia or bipolar disorder. Even the highest figure was below 100%, illustrating variation in outcomes.

Comparisons must retain units and denominators. The schizophrenia difference between two affected parents and one was $27.3 - 7.0 = 20.3$ percentage points. Its ratio was $27.3 \div 7.0 = 3.9$, or 3.9 times the estimated risk. A 20.3 percentage-point difference is not a 20.3% relative increase. Group comparisons do not provide an exact prediction for a particular child.

The findings support familial aggregation and are consistent with genetic contribution to schizophrenia and bipolar disorder. Biological relatives share genes and environments, so this design alone does not isolate inheritance. Although unipolar depression appears in the wider outcome, the reported narrow schizophrenia or bipolar percentages must not be borrowed as risk estimates for MDD.