First, statistical methods are now being applied to molecular genetic data to estimate heritability for complex disorders like major depression (
8–
10). Such estimates—termed SNP heritability—for major depression are, as for other psychiatric disorders, considerably lower than heritability estimates derived from twin studies (
9,
11–
13). The difference has been termed the “missing heritability problem” (
14). While SNP heritability might underestimate true heritability for many reasons, an additional explanation is that twin studies overestimate heritability (
15–
18). With Scandinavian registries, it has become possible to compare heritability estimates from twin studies with those obtained from other genetically informative relationships that are not susceptible to the methodological problems of twin studies, although they have potential limitations of their own (
19,
20).
Second, questions remain about important genetic epidemiologic features of major depression that are relevant for the design of molecular genetic studies. Do genetic effects on major depression differ in men and women? While two of the largest-scale twin studies of major depression (
6,
21) reported higher heritability in women than in men, a meta-analysis found no overall evidence for such effects (
7). While two large twin studies (
21,
22) suggested qualitative differences in genetic effects in the two sexes (i.e., risk genes for major depression not identical in men and women), two other studies (
6,
23) found no support for such effects.
Method
We collected information on individuals from Swedish population-based registries with national coverage, which were linked using each person’s unique personal identification number; to preserve confidentiality, personal identification numbers were replaced with serial numbers. We used the following sources: the Multi-Generation Register; Population and Housing Censuses; the Swedish Hospital Discharge Register (national coverage for 1987–2012 and partial coverage for 1969–1986); and the Outpatient Care Register (national coverage for 2001–2012). We also used a new Primary Care Registry, which included individual-level information on diagnoses from visits to primary health care centers in 15 of Sweden’s 21 counties: Blekinge (2009–2016), Värmland (2005–2015), Kalmar (2007–2016), Sörmland (1997–2017), Uppsala (2005–2015), Västernorrland (2008–2015), Norrbotten (2009–2016), Gävleborg (2010–2016), Halland (2007–2014), Jönköping (2008–2014), Kronoberg (2006–2016), Skåne (1998–2013), Östergötland (1997–2014), Stockholm (2003–2016), and Västergötland (2000–2013). Time periods for coverage differ because of the timing of the digitalization of records. In 2016, these 15 counties contained 87% of the Swedish population. We secured ethical approval for this study from the Regional Ethical Review Board of Lund University.
Major depression was identified in the Hospital Discharge, Outpatient (Specialist) Care, and Primary Care Registries by ICD code (ICD-8 codes 296.2, 298.0, and 300.4; ICD-9 codes 296.2, 296.4, 298.0, and 300.4; ICD-10 codes F32 and F33). The diagnosis could be registered at any time, although we required a first registration at age 15 or older. Individuals registered with schizophrenia (ICD-8 and ICD-9 code 295; ICD-10 code F20) or bipolar disorder (ICD-8 codes 296.1 and 296.3; ICD-9 codes 296.0, 296.1, and 296.4–8; ICD-10 codes F30 and F31) were censored from the sample.
For the first database, we selected from the Swedish Twin Registry all twin pairs with known zygosity and birth years between 1950 and 1990, and from the Swedish Multi-Generation Register all Swedish-born full- and half-sibling pairs born between 1950 and 1990 and within 5 years of each other. Zygosity was assigned using standard self-report items, which, when validated against biological markers, were 95%−99% accurate (
25). Using the Swedish national census and population registers, we assessed cohabitation status for full- and half-sibling pairs as the proportion of possible years they lived in the same household until the oldest turned 16. We included pairs reared together and apart cohabiting for ≥80% or ≤20% of the possible years, respectively. Because the Primary Care Registry did not have complete national coverage, we required that both individuals in a pair have resided ≥8 years in a county that registered diagnoses in the Primary Care Registry before we included diagnoses from the register in our analyses.
Structural equation modeling was used to assess the contributions of genetic and environmental factors to liability for major depression, assuming a liability threshold model (
26). In the first model, using monozygotic and dizygotic twins only, we assumed three sources of liability—additive genetic (A), shared environment (C), and unique environment (E)—and we assumed that monozygotic twins share all and dizygotic twins on average 50% of their genes identical by descent. Shared environment reflects family and community experiences that render the twins more similar for major depression, while unique environment includes environmental experiences not shared by twins, we well as error. In the second model, using full and half siblings, we employed the same modeling framework, assuming that full and half siblings share, on average 50% and 25% of genes, respectively, identical by descent. Shared environment (C) equaled 1 for pairs reared together and 0 pairs reared apart.
The second database was created by double-entering all cousin, half-sibling, full-sibling, and monozygotic twin pairs in the Swedish population in which both individuals in the pair were born in Sweden between 1970 and 1990, both were still residing in Sweden at age 15, the pair had no more than a 10-year age difference, and at least one of the pair (the proband) was registered for major depression. In concordant pairs in which both were registered with major depression, each was treated as a proband. Each pair was assigned their degree of genetic resemblance (1 for monozygotic twins; 0.5 for full siblings; 0.25 for half siblings; 0.125 for cousins). We also applied the same restriction as described above, including major depression diagnoses from the Primary Care Registry for pairs only when both members of the pair had resided at least 8 years in a county that used the Primary Care Registry.
Using Cox regression models, we investigated all pairs of relatives from age 15 until year of first registration of major depression in the relative, death or emigration of the relative, or end of follow-up (December 31, 2015), whichever came first. We examined eight putative indices of genetic risk for major depression in the proband: age at first registration, number of depressive episodes, lifetime history of an anxiety disorder (defined by ICD-9 codes 300.0 and 300.2 and ICD-10 codes F40 and F41), a history of antidepressant treatment, treatment by a psychiatrist, a history of ECT, sick-leave registration, and granting of early retirement. Details on these indices are provided in Table S1 in the online supplement.
In one model, we estimated results for each putative index in the four classes of relatives as a function of their genetic relationship. This model produced an estimated hazard ratio for each relative class and a measure of interaction—the significance of the difference of the hazard ratios across types of relatives. This model used data from all relatives to optimize the accuracy of hazard ratio estimates, which was especially important in monozygotic twin pairs, where the sample size was small. We controlled for the absolute age difference between the proband and relative, sex of the proband and the relative, and year of birth of the proband. Because there could be several pairs from the same cluster of relatives (e.g., sibships), we used robust standard errors to account for their nonindependence. In all models, we checked whether the regression coefficients varied over time. However, in no case were there substantial changes to our results.
In a final analysis, each covariate was standardized with a mean of zero and a standard deviation of 1, and then summarized across each proband to a summary z-score. We categorized the z-score into 10 equally sized groups and used these categories as covariates (using the lowest-risk group as the reference) in a Cox regression analysis in which we included the interaction between the z-score group and familial resemblance. This model produced an estimated hazard ratio for each relative class and for each z-score group and a measure of an interaction between them—the significance of the difference of the hazard ratios across types of relatives.
The statistical analyses were performed using SAS, version 9.4 (
27) and the OpenMx software package (
28).
Results
The sample sizes and lifetime prevalences for major depression in twin, full-sibling, and half-sibling pairs are summarized in
Table 1. Our samples were largest for full siblings reared together, intermediate for half siblings reared together and apart, and smallest for twins and full siblings reared apart. Our twin and sibling/half-sibling analyses included 26,892 and 1,691,971 pairs, respectively. Prevalences for major depression were similar for the three groups from intact families (monozygotic and dizygotic twins and full siblings reared together), around 7% in males and 12% in females. In the three groups from non-intact families (half siblings reared together and apart and full siblings reared apart), the rates were similar across groups and were 40%−50% higher in both males and females.
Tetrachoric correlations for treated major depression ranged from 0.40 to 0.50 in monozygotic twins, 0.15 to 0.25 in full siblings, and 0.08 to 0.15 in half siblings (
Table 1). Correlations were consistently higher in female-female compared with male-male pairs, same-sex compared with opposite-sex pairs, and reared-together compared with reared-apart pairs.
Table 2 presents the model-fitting results in our two samples. Parameter estimates were similar, with overlapping confidence intervals. The heritability of major depression was estimated in the twins at 0.41 (95% CI=0.21, 0.49) in males and 0.49 (95% CI=0.31, 0.56) in females. Parallel estimates from the full/half-sibling sample were, respectively, 0.36 (95% CI=0.31, 0.38) and 0.51 (95% CI=0.51, 0.53). Estimates for the shared environment were present in three of the analyses (all but male twins) but were small (≤ 5%). While the genetic correlation between the sexes was estimated at 1.00 in the twins, this is known imprecisely (95% CI=0.83, 1.00) and is consistent with the more accurate estimate from our full- and half-sibling sample (0.89, 95% CI=0.87, 0.91).
Table 3 includes the sample sizes used for our analyses of indices of genetic liability for major depression, which included 875,010 pairs of relatives, as well as descriptive information about these indices. The mean age at first registration was around 29 years, and the mean number of episodes of major depression was 2.3. Approximately 50% of the sample had an anxiety disorder diagnosis, 77% had received prescriptions for antidepressants, 34% were granted sick leave, and 9% were granted early retirement. Sixty-four percent of the sample received their diagnoses of major depression from nonpsychiatric physicians, largely in primary care settings.
The risk for major depression in our four classes of relatives as a function of the presence or absence of the indices of liability in the proband is summarized in
Table 4. Seven of these indices (all but sick leave) behaved in the predicted manner. That is, risk for major depression increased in relatives of probands with major depression who had multiple episodes, younger age at first registration, were treated by psychiatrists, received antidepressants or ECT, had a comorbid anxiety diagnosis, or were granted early retirement. Furthermore, as predicted for a genetic index, the increased risk in relatives was strongest in monozygotic twins, intermediate in siblings, weaker in half siblings, and weakest in cousins. The difference in hazard ratios across the four groups of relatives was significant for all indices except ECT.
We combined all our risk indices (except sick leave) into a single weighted score, calculated deciles of probands on that score, and examined risk of major depression in their relatives. The results are depicted in Figure S1 and Table S2 in the online supplement. Compared with relatives of probands at lowest genetic risk, the risk for major depression in the top decile was 2.82 higher in monozygotic co-twins and 1.53 times higher in full siblings of probands.
To estimate the contribution of individual risk indices, we included all of them except sick leave in a multivariate model predicting risk of illness in our classes of relatives and indexed their effect size by the chi-square value of their individual effects on risk in monozygotic co-twins. In order, the chi-square values were 339.6 for age at first registration, 147.7 for number of episodes, 109.8 for comorbid anxiety disorder, 77.3 for referral to psychiatric care, 38.2 for treatment with antidepressants, and 15.2 for having been granted early retirement. Receipt of ECT was not significantly associated with estimated risk (χ2=0.5).