In asylums where acute cases only are received, it is easily to be understood that hereditary tendency does not reach a figure so elevated; probably such cases are more frequently caused by other conditions than insanity being present in their forefathers; on the contrary, “in an asylum where … incurables are received, the hereditary predisposition appears in a much higher proportion” (Stewart, Reference Stewart1864, 53).
A strong and direct heredity (in cases of insanity) implies three things, (1) instability of brain, (2) liability to attacks at early ages, and (3) liability to a recurrence after cure (Clouston, Reference Clouston1884, 291).
For scientific psychiatry to progress, we need to collect comparable samples across research sites. This has been especially true for molecular genetic studies where, due to the small effect size of individual variants, well-powered samples can often only be achieved by consortia. However, more than 150 years ago, two prominent British Alienists, Hugh Stewart and Thomas Clouston, observed that a strong hereditary predisposition might impact on the presentation of psychotic cases. Specifically, Stewart suggests that they will have greater chronicity while Clouston observes that they are likely to have an earlier age at onset (AAO) and higher rates of recurrence.
Operationalized diagnostic criteria were created to help achieve comparability of patient samples across research centers (Spitzer, Williams, & Skodol, Reference Spitzer, Williams and Skodol1980). However, if Stewart and Clouston are correct, genetic profiles of the psychiatrically ill might be quite variable, even when collected using DSM or ICD diagnostic criteria, if the samples differed in key clinical features such as AAO, recurrence, and chronicity. If true, comparisons across samples might be problematic and result in non-replications, despite current use of diagnostic criteria. Furthermore, pooling of cross-center results, as occurs in research consortia, could increase the heterogeneity of findings.
In a previous report, we showed, using family genetic risk scores (FGRS) calculated in extended pedigrees in Sweden, that, in accord with Stewart and Clouston's predictions, the FGRS for schizophrenia (SZ) was higher in cases with schizophrenic illness with early onset and with high numbers of recurrences (Kendler, Ohlsson, Sundquist, & Sundquist, Reference Kendler, Ohlsson, Sundquist and Sundquist2021). The same pattern was seen in bipolar disorder (BD).
In this study, we move beyond this earlier report in two ways. First, instead of looking only at the FGRS for the primary disorder (i.e. SZ FGRS in patients with SZ or the bipolar FGRS in patients with bipolar illness), we examine an FGRS risk profile of eight diverse psychiatric disorders. Second, we consider three non-psychotic disorders to see if we obtain similar patterns of findings: drug use disorder (DUD), major depression (MD), and attention-deficit hyperactivity disorder (ADHD). (Before proceeding, we should note that FGRS differs fundamentally from polygenic risk scores. The information for the former comes from registry-based disorders in extended family members while the information of the latter comes from DNA sequence variation.)
Therefore, we ask two inter-related questions with national patient cohorts of DUD, MD, and ADHD in Sweden whom we subdivide by sex, AAO, recurrence, mode of ascertainment, and treatment. First, do we see the same trends predicted by Steward and Clouston and observed in SZ and bipolar illness, for the primary FGRS to be higher in cases of early onset and multiple recurrences? Second, are the FGRS profiles for the resulting subgroups of our primary disorders sufficiently similar to assume they represent the same patient population or are they sufficiently different that future researchers need to consider this diversity in future research studies?
Methods
We collected information on individuals from Swedish population-based registers with national coverage linking each person's unique personal identification number which, to preserve confidentiality, was replaced with a serial number by Statistics Sweden (for details, see online Supplementary Appendix Table 1). Ethical approval was provided by the Regional Ethical Review Board in Lund and participant consent was not required (No. 2008/409, 2012/795, and 2016/679). Our database consisted of all individuals born in Sweden 1932–1995 of Swedish-born parents (n = 5 829 952). The database included registrations for MD, DUD, and ADHD, utilizing ICD-8, 9, and 10 codes from primary care, specialist and hospital registries, and from criminal registers (for details, see online Supplementary Appendix Table 2). We included four clinical characteristics: AAO (defined as age at first registration), level of recurrence (number of independent registrations), type of ascertainment, and type of treatment (for details, see online Supplementary Appendix Table 3). We also included FGRS for MD, anxiety disorders (AD), BD, SZ, alcohol use disorder (AUD), DUD, ADHD, and autism-spectrum disorder (ASD). The FGRS were based on a mean of 32.2 1st through 5th degree relatives of the probands. Briefly (for details, see online Supplementary Appendix Table 4), we first calculated the morbid risk for the phenotype based on age at first registration. Thereafter, we transformed the binary trait into an underlying liability distribution and calculated the mean Z-score for relatives with and without the trait. For 1st degree relatives, we also multiplied the z-score with a factor designed to control for cohabitation effects. Within each type of relative, we had the sum of the individual z-scores and the weighted number of individuals which were then further weighted by their genetic resemblance to the proband. For each proband, we summed the two components across all groups of relatives and used the quotient, which was then multiplied by a shrinkage factor to take account of the number of relatives of the proband. To ensure that the FGRS would be comparable across disorders, we standardized them using year of birth and county of residence, into a z-score with mean = 0 and s.d. = 1.
To examine FGRS profiles of individuals with MD, DUD, and ADHD, we present the mean FGRS subdivided by sex, and type of treatment. For AAO analyses, we used a linear regression model with the FGRS as outcome and AAO as a continuous variable controlling for year of birth. We present the predicted FGRS at the 10th, 30th, 50th, 70th, and 90th percentiles of the distribution for individuals at the mean birth year. For recurrence, we used a linear regression model with number of recurrences as a continuous variable and we present the predicted FGRS at the same percentiles. For DUD, the 10th, 30th, and 50th percentile in the recurrence distribution was 1, while for MD the 10th and 30th were 1. To investigate ascertainment source for DUD by register, we present the mean FGRS as well as a comparison across all groups. For analyses of ascertainment based on medical register and clinical severity by ICD10 code, we also use a linear regression model treating the type of registration as an ordinal variable. In the figures, we present the predicted FGRS at the different levels of the relevant variable. Our figures depict 144 statistical comparisons. Given that many are inter-correlated, we assumed 100 independent tests, and a p value threshold of ⩽0.0005 as a guide to significance. Statistical analyses were performed using SAS 9.4 (SAS Institute, 2012)
Results
Descriptive features
MD was the most common disorder examined with a prevalence of 11.4 and 63.1% female. Parallel results for DUD and ADHD were 3.5 and 34.2%, and 1.4 and 43.3%.
Sex
The FGRS profiles of females and males with DUD, MD, and ADHD are seen, respectively, in Figs 1a–c. Of note, FGRS is a z-transformed measure of the aggregation of disorders in extended families so that scores of 0, 0.5, and 1.0, reflect, respectively, a level of familial-genetic risk that is average for the Swedish population, 0.5 s.d. higher than the general population and 1.0 s.d. higher. To obtain a sense of the magnitude of the effect of these FGRS on disorder risk, see online Supplementary Appendix Fig. 1, which shows that the risk of DUD and ADHD in individuals with an FGRS of 2 is double that for those with an FGRS of 0 and for MD, the parallel increase is approximately 50%.
Fig. 1. (a) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of sex differences in individuals diagnoses with DUD. These FGRS are depicted on the Y-axis. F, female, M, male. The differences are obtained from a linear regression analysis and reflect the size of the observed differences, here between females and males. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the difference is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. Note that the scale of the Y-axis varies across the figures in this paper. (b) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of sex differences in individuals diagnoses with MD. These FGRS are depicted on the Y-axis. F, female, M, male. The differences are obtained from a linear regression analysis and reflect the size of the observed differences, here between females and males. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the difference is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (c) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of sex differences in individuals diagnoses with ADHD. These FGRS are depicted on the Y-axis. F, female, M, male. The differences are obtained from a linear regression analysis and reflect the size of the observed differences, here between females and males. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the difference is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance.
For cases of DUD, affected females had significantly higher FGRS than affected males for all disorders with the strongest effects for MD and AD FGRS. For cases of MD, differences in FGRS by sex were significant for four disorders all higher in males: AD, BD, AUD, and DUD. For cases of ADHD, significant differences by sex were seen for MD, AD, BD, ADHD, and ASD, all higher in females.
Age at onset
FGRS profiles for DUD, MD, and ADHD by AAO are seen in Figs 2a–c. Significant effects in cases of DUD were seen for the SZ, AUD, DUD, and ADHD FGRS, all declining with increasing AAO. Effect sizes were largest for AUD and DUD FGRS. In cases with MD, significant declines were seen in all FGRS with increasing AAO. Effect sizes were largest for MD and DUD. For ADHD, increasing AAO in cases was associated with significant declines in the PRS scores for SZ, AUD, DUD, ADHD, and ASD with the strongest impact on ADHD FGRS.
Fig. 2. (a) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of differences in age of onset of DUD, indexed by age at first registration. These FGRS are depicted on the Y-axis. The figure depicts estimates from a linear regression model at the 10th, 30th, 50th, 70th, and 90th percentile of the age at onset (AAO) distribution for individuals with DUD at their mean year of birth. As depicted in the figure, these percentiles are calculated as AAOs for DUD at ages 18, 22, 28, 40, and 57, respectively. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences across AAO. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (b) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of differences in age of onset of MD, indexed by age at first registration. These FGRS are depicted on the Y-axis. The figure depicts estimates from a linear regression model at the 10th, 30th, 50th, 70th, and 90th percentile of the age at onset (AAO) distribution for individuals with DUD at their mean year of birth. As depicted in the figure, these percentiles are calculated as AAOs for MD at ages 22, 33, 44, 55, and 69, respectively. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences across AAO. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (c) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of differences in age of onset of ADHD, indexed by age at first registration. These FGRS are depicted on the Y-axis. The figure depicts estimates from a linear regression model at the 10th, 30th, 50th, 70th, and 90th percentile of the age at onset (AAO) distribution for individuals with DUD at their mean year of birth. As depicted in the figure, these percentiles are calculated as AAOs for ADHD at ages 15, 21, 27, 35, and 48, respectively. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences across AAO. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance.
Recurrence
The FGRS profiles for DUD, MD, and ADHD for increasing numbers of episodes are seen in Figs 3a–c. For cases of DUD, significant increases are seen in all FGRS with increasing levels of recurrence with the strongest effects seen for DUD and AUD. FGRS for all disorders also increase significantly with higher levels of recurrence for MD with MD FGRS having the largest effect. For cases with ADHD, with greater numbers of episodes, all of the FGRS increased significantly except for SZ. The effect size was strongest for ADHD FGRS.
Fig. 3. (a) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism spectrum disorder (ASD) as a function of differences in the number of episodes of DUD, indexed by number of independent registrations. These FGRS are depicted on the Y-axis. The figure depicts estimates from a linear regression model at the 10th, 30th, 50th, 70th, and 90th percentile of numbers of episodes for individuals with DUD at their mean year of birth. However, for DUD, the 10th, 30th, and 50th percentile in the recurrence distribution was 1 and the 70th and 90th percentiles, were, respectively, 3 and 10. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences across the number of episodes. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (b) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of differences in the number of episodes of MD, indexed by number of independent registrations. These FGRS are depicted on the Y-axis. The figure depicts estimates from a linear regression model at the 10th, 30th, 50th, 70th, and 90th percentile of numbers of episodes for individuals with MD at their mean year of birth. However, for MD, the 10th and 30th percentile values for numbers of episodes were 1 and for the 50th, 70th, and 90th percentile equaled, respectively, 2, 3, and 6. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences across number of episodes. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (c) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of differences in the number of episodes of ADHD, indexed by number of independent registrations. These FGRS are depicted on the Y-axis. The figure depicts estimates from a linear regression model at the 10th, 30th, 50th, 70th, and 90th percentile of numbers of episodes for individuals with ADHD at their mean year of birth. As depicted in the figure, these percentiles are calculated as number of episodes for ADHD equaled, respectively, 1, 2, 3, 5, and 9. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences across number of episodes. Red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance.
Source of ascertainment
Figure 4a compared cases of DUD ascertained through criminal, medical, and pharmacy registers. AUD, DUD, and ADHD FGRS in these cases showed a consistent pattern with the highest FGRS in the criminal followed by medical and then pharmacy registers. For BD and ASD FGRS in DUD cases, the opposite result was seen. For MD, AD, and SZ FGRS, levels were highest in DUD cases ascertained in the medical registry. Largest effect sizes were seen for AUD and DUD FGRS.
Fig. 4. (a) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of the ascertainment of DUD from the three main Swedish registries: criminal, medical, and pharmacy. A hierarchy was used as follows: criminal > medical > prescription. These FGRS are depicted on the Y-axis. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Because these three registries differ qualitatively, we present p values for the three possible comparisons for each FGRS. Cr, criminal; Med, medical; Pre, prescription. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (b) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of sources of ascertainment for DUD within the medical registries, that is, from in-patient (IP) facilities, specialist care (SC) out-patient facilities, and primary-care (PC) out-patient facilities. We used a hierarchy such that registration in the IP superseded other registrations and registration in an SC clinic superseded that in a PC clinic. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences across the three sources of registration treated as an ordinal variable. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (c) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of sources of ascertainment for MD within the medical registries, that is, from in-patient facilities when a suicide attempt was registered the same day as admission (IP + SA), in-patient facilities without a suicide registration (IP), specialist care (SC) out-patient facilities, and primary-care (PC) out-patient facilities. We used the following hierarchy: IP + S > IP > SC > PC. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences across the three sources of registration treated as an ordinal variable. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (d) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of sources of ascertainment for ADHD within the medical registries, that is, from in-patient (IP) facilities, specialist care (SC) out-patient facilities, and primary-care (PC) out-patient facilities. We used a hierarchy such that registration in the IP superseded other registrations and registration in an SC clinic superseded that in a PC clinic. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences across the three sources of registration treated as an ordinal variable. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance.
In Fig. 4b, we examined only DUD cases ascertained through the medical register. We see significant differences in FGRS for all disorders as a function of the mode of ascertainment of DUD with highest levels seen for those from in-patient setting and lowest levels for those ascertained in primary care clinics.
Figure 4c depicts the FGRS profiles for MD patients ascertained by: (i) hospitalization with a suicide attempt, (ii) standard hospitalization, (iii) specialty clinics, and (iv) primary care clinic. Mean FGRS for all disorders declined with decreasing intensity of the treatment setting in MD patients with the effects largest for AUD and DUD.
In Fig. 4d, significant differences are seen for all FGRS in ADHD patients as a function of the intensity of the medical setting in which they were ascertained. The largest differences were seen for DUD, AUD, and ADHD FGRS.
Other contrasts
FGRS profiles of clinical severity for MD show significant differences for all FGRS except for SZ, with high scores correlated with greater severity (Fig. 5a). The effect sizes were largest for DUD and AUD FGRS.
Fig. 5. (a) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of severity of MD as coded in ICD-10. The slopes are obtained from a linear regression analysis and reflect the size of the observed differences between the three levels of clinical severity treated as an ordinal variable. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the differences is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (b) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of the presence or absence of a history of ECT among cases with MD. These FGRS are depicted on the Y-axis. The differences are obtained from a linear regression analysis and reflect the size of the observed differences, here between cases who did v. did not receive ECT treatment. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the difference is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance. (c) The mean standardized family genetic risk scores (FGRS) (±95% confidence intervals) for major depression (MD), anxiety disorders (AD), bipolar disorder (BD), schizophrenia (SZ), alcohol use disorder (AUD), drug use disorder (DUD), attention-deficit hyperactivity disorder (ADHD), and autism-spectrum disorder (ASD) as a function of the presence or absence of a history of stimulant treatment among cases with ADHD. These FGRS are depicted on the Y-axis. The differences are obtained from a linear regression analysis and reflect the size of the observed differences, here between cases who did v. did not receive stimulant treatment. The colors of the columns reflect the class of the disorders: red (black) – internalizing, yellow (dark grey) – psychotic, green (grey) – substance use, and blue (light grey) – developmental. The p value of the difference is obtained from the same linear regression model. Given the large number of tests performed, we set a p < 0.0005 as a threshold for statistical significance.
Differences across hospitalized MD cases who did v. did not receive Electro-convulsive Therapy (ECT) were significant for FGRS for five disorders: MD, BD, SZ, ADHD, and ASD (Fig. 5b). FGRS was higher for MD, BD, SZ, and ASD in those receiving ECT, with the opposite significant effect seen for ADHD. The effect size was highest for BD FGRS.
In ADHD cases who did v. did not receive stimulant treatment, all FGRS were significantly elevated in those with treatment except SZ (Fig. 5c). The largest effect size was seen for ADHD FGRS.
Discussion
We examined the degree to which the genetic profiles of individuals with three diverse disorders – DUD, MD, and ADHD – varied as a function of sex, AAO, recurrence, modes of ascertainment, and treatment. These profiles consisted of FGRS for eight disorders reflecting a broad array of psychiatric conditions. From our many results, we focus on nine.
First, consistent with the predictions of Stewart and Clouston for psychotic illness and our own previous findings for SZ and bipolar illness (Kendler et al., Reference Kendler, Ohlsson, Sundquist and Sundquist2021), we found for DUD, MD, and ADHD that their primary FGRS was positively associated both with young AAO and high levels of recurrence. This may be a feature of diverse psychiatric disorders as well as biomedical disorders like myocardial infarction where familial risk is also highly and inversely associated with AAO (Marenberg, Risch, Berkman, Floderus, & de Faire, Reference Marenberg, Risch, Berkman, Floderus and de Faire1994; Rissanen, Reference Rissanen1979; Zdravkovic et al., Reference Zdravkovic, Wienke, Pedersen, Marenberg, Yashin and de Faire2002).
Second, the magnitude of differences found in the FGRS profile for our three disorders as a function of clinical variables that we investigated raises concerns for a broad array of clinical research designs in psychiatry, especially psychiatric genetics. Our findings predict that important differences in results from studies might emerge if the selected samples varied across these key characteristics. To avoid unexplained cross-sample discrepancies, it may not be sufficient to match on DSM categories. Instead, our results suggest that, to obtain comparable results, especially in the area of genetics, it will be necessary to control more precisely for the clinical features of the patients being studied.
Third, we saw, for DUD, MD, and ADHD, consistent evidence that primary FGRS were positively and often substantially correlated not only with early AAO, and recurrence, but ascertainment in higher intensity clinical settings. For MD and ADHD, higher primary FGRS was associated with treatment by, respectively, ECT and stimulants. For MD, clinical severity was positively correlated with the MD FGRS. For DUD, the FGRS was highest in cases ascertained through the criminal registry. Both DUD and ADHD primary FGRS were higher in affected females than males, as predicted by a female protective effect (Taylor et al., Reference Taylor, Lichtenstein, Larsson, Anckarsäter, Greven and Ronald2016), but for MD no such differences were found.
Many of our results are congruent with prior findings for our three disorders of interest. Recurrence and early AAO best predicted risk for MD in relatives of depressed patients (Sullivan, Neale, & Kendler, Reference Sullivan, Neale and Kendler2000; Weissman et al., Reference Weissman, Wickramaratne, Merikangas, Leckman, Prusoff, Caruso and Gammon1984) and this was replicated by more recent work in the Swedish Twin Registry (Kendler, Gatz, Gardner, & Pedersen, Reference Kendler, Gatz, Gardner and Pedersen2005, 2007). The risk for DUD in relatives of affected individuals is predicted by proband AAO and number of episodes (Kendler, Ohlsson, Sundquist, & Sundquist, Reference Kendler, Ohlsson, Sundquist and Sundquist2014). Risk of ADHD in relatives of ADHD probands is predicted by proband persistence/recurrence (Faraone, Reference Faraone2004; Faraone, Biederman, & Monuteaux, Reference Faraone, Biederman and Monuteaux2000).
Fourth, important patterns were also found for non-primary FGRS (FGRS for disorders other than the one being studied). Looking at DUD, higher FGRS for internalizing disorders (MD and AD) were found for females v. males, in those with high v. low rates of recurrence, in cases ascertained through medical v. criminal registries and from high v. lower intensity medical settings. In all cases, FGRS for AUD and ADHD mirrored changes in DUD FGRS.
MD demonstrated modestly higher levels of AUD and DUD FGRS in males than females and in individuals with high v. low levels of recurrence. Larger differences in the FGRS for these substance use disorders were observed in early v. late-onset MD, in those ascertained in high v. low-intensity medical settings and in those presenting with severe v. mild depressive episodes. In all comparisons, the patterns of MD FGRS in MD patients were mimicked by that seen for AD FGRS. Of particular interest, the BD FGRS was higher in MD cases who were female, who had high levels of recurrence, were treated in high-intensity clinical settings, and were clinically severely ill.
For ADHD, higher FGRS for MD and AD were seen in affected females, in highly recurrent cases, in those ascertained in more intense clinical settings, and those receiving stimulant medication. Furthermore, in virtually all comparisons, the patterns of ADHD FGRS in ADHD patients were mimicked by those for the AUD and DUD FGRS.
We suggest that our findings demonstrated in this paper have important lessons for clinical and research psychiatry. Given that individuals have sufficient primary genetic risk factors and the associated environmental exposures to become ill, the clinical patterns of that disorder, including its AAO, rates of recurrence, and intensity of clinical treatment can be influenced, sometimes substantially so, by the genetic risk to other psychiatric disorders. Those genetic risks can be for relatively closely related disorders (e.g. ADHD for DUD cases) but can also be for disorders that are more distantly related (e.g. DUD for MD cases).
Fifth, it is useful to examine the purity of genetic signal, that is the relative magnitudes of the changes in the primary v. non-primary FGRS. Examining AAO in DUD (Fig. 2a), the FGRS for DUD, AUD, and ADHD is higher in younger onset cases, but no significant changes in the FGRS are seen for MD or AD. That is, early-onset DUD provides a ‘purer’ genetic signal for externalizing v. internalizing disorders than does late-onset DUD because the FGRS for MD changes only modestly with age, while the FGRS for DUD and AUD decline substantially with older AAO.
The range of possible findings on genetic ‘purity’ can be illustrated with three examples for MD. Higher MD FGRS is seen in more intensive treatment settings, but the rate of increase of AUD and DUD across treatment intensity is even greater. Thus, while studying hospitalized cases of MD would increase the level of genetic risk for MD, the ‘purity’ of that signal might actually decline as a higher proportion of the observed differences would be in FGRS for substance use disorders. By contrast, the declining rate of FGRS for MD with later AAO is almost exactly paralleled by the decreasing levels for AUD, DUD, and ADHD FGRS. So, studying early-onset MD would increase the genetic signal for MD, but, with respect to externalizing disorder genetic risk, that signal will not be any ‘purer’ than that obtained by studying late-onset cases. Finally, the FGRS for MD increases with increasing recurrence faster than that seen for AUD, DUD, and ADHD FGRS. So, by studying highly recurrent cases of MD, it should be possible to produce not only stronger genetic signals but ones that are potentially ‘purer’ with respect to genetic risk for externalizing syndromes.
Sixth, some of the variation seen in the FGRS across our conditions is quantitative. That is, most of the FGRS move together increasing and decreasing, albeit often at different rates. This pattern is seen in our examination of treatment intensity for DUD (Fig. 4b) where all FGRS decline with decreasing intensity of the treatment setting. Compare this to the patterns found when we compare MD with and without ECT treatment where some FGRS increase significantly (MD, SZ, and ASD), several hardly change (AD, AUD, and DUD), and one significantly decreases (ADHD). We could describe this pattern as a qualitative difference. A pattern of quantitative differences is more common in our findings, especially for differences due to AAO and levels of recurrence. But qualitative effects are also seen for registry of ascertainment for DUD and sex effects for ADHD.
Seventh, our results provide evidence, consistent with prior studies, for close genetic relationships among the eight disorders making up our FGRS profiles. The pattern of changes for MD and AD FGRS was highly correlated across our analyses (Reference Kendler, Walters, Neale, Kessler, Heath and Eavesb21Kendler et al., 1995; Kendler et al., Reference Kendler, Aggen, Knudsen, Roysamb, Neale and Reichborn-Kjennerud2011; Kendler, Gardner, Gatz, & Pedersen, Reference Kendler, Gardner, Gatz and Pedersen2006; Kendler, Neale, Kessler, Heath, & Eaves, Reference Kendler, Neale, Kessler, Heath and Eaves1992; Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium, 2018; Purves et al., Reference Purves, Coleman, Meier, Rayner, Davis, Cheesman and Eley2019) as were the patterns for AUD and DUD FGRS (Johnson et al., Reference Johnson, Demontis, Thorgeirsson, Walters, Polimanti, Hatoum and Agrawal2020; Kendler et al., Reference Kendler, Aggen, Knudsen, Roysamb, Neale and Reichborn-Kjennerud2011). Changes in BD FGRS often but not always paralleled changes in MD FGRS (Lee et al., Reference Lee, Ripke, Neale, Faraone, Purcell, Perlis and Wray2013) and a similar pattern was seen for the ADHD and DUD FGRS (Johnson et al., Reference Johnson, Demontis, Thorgeirsson, Walters, Polimanti, Hatoum and Agrawal2020).
Eighth, because we are studying only cases ascertained through treatment (and, for DUD, through the crime registry), we cannot definitively separate effects on liability and effects on ascertainment. That is, to what degree does the higher liability to AUD seen in early v. late-onset MD arise because high AUD risk contributes more to the etiology of early onset of AUD or because a high AUD liability increases the chances of ascertainment of MD in a younger v. an older individual?
Finally, our results both confirm and substantially extend those of our prior report on the FGRS patterns in cases of AUD (Kendler, Ohlsson, Sundquist, & Sundquist, In press) where we saw similar results to those obtained here. Our combined results suggest a number of consistent changes in the genetic profile of internalizing, substance use, and developmental disorders across the clinical variables we have explored. Further research will be needed to determine the degree to which these patterns are observed across yet other psychopathological domains.
Limitations
These findings should be viewed in the context of seven potential methodological concerns. First, these analyses depend on the quality of the diagnoses in the Swedish registries which has been well demonstrated for SZ and BD (Ekholm et al., Reference Ekholm, Ekholm, Adolfsson, Vares, Osby, Sedvall and Jonsson2005; Lichtenstein et al., Reference Lichtenstein, Bjork, Hultman, Scolnick, Sklar and Sullivan2006; Sellgren, Landen, Lichtenstein, Hultman, & Langstrom, Reference Sellgren, Landen, Lichtenstein, Hultman and Langstrom2011). The validity of MD diagnoses is supported by its prevalence, sex ratio, risk factors, and familial aggregation patterns (Kendler, Ohlsson, Lichtenstein, Sundquist, & Sundquist, Reference Kendler, Ohlsson, Lichtenstein, Sundquist and Sundquist2018; Sundquist, Ohlsson, Sundquist, & Kendler, Reference Sundquist, Ohlsson, Sundquist and Kendler2017). Genetic epidemiological findings for AUD and DUD in Sweden are similar to those found in other samples (Kendler et al., Reference Kendler, Sundquist, Ohlsson, Palmer, Maes, Winkleby and Sundquist2012, 2015, Kendler et al., Reference Kendler, PirouziFard, Lonn, Edwards, Maes, Lichtenstein and Sundquist2016; Kendler, Maes, Sundquist, Ohlsson, & Sundquist, Reference Kendler, Maes, Sundquist, Ohlsson and Sundquist2013).
Second, age at first registration and number of registrations are used as proxies for AAO and recurrence rates, respectively. For chronic disorders like ADHD, registrations may reflect clinical exacerbations rather than recurrences, a somewhat different reflection of course of illness.
Third, the frequency of ADHD diagnosis in Sweden and its AAO have been increasing over time, suggesting a rising interest in adult ADHD [(Giacobini, Medin, Ahnemark, Russo, & Carlqvist, Reference Giacobini, Medin, Ahnemark, Russo and Carlqvist2018) and online Supplementary Appendix Fig. 2]. Year of birth was included as a covariate in our analyses to control for such effects.
Fourth, our FGRS is an estimate of genetic risk reflecting aggregation of disease in close and distant relatives and is not equivalent to a molecular polygenic risk score. Our corrections for cohabitation are approximate but impact modestly on the FGRS. Online Supplementary Appendix Table 5 shows that our final genetic risk scores are not highly sensitive to key steps in their calculation, and simulations (see below) demonstrated that our corrections for cohabitation functioned well. We also show (online Supplementary Appendix Fig. 3) that the FGRS for DUD, MD, and ADHD are relatively stable across cohorts and geography with the exception of higher scores for ADHD in the older cohort. Including year of birth in our analyses should control for such effects.
Fifth, we sought to validate the FGRS by comparing it to a recently proposed quantitative family-history score [LT-FH (Hujoel, Gazal, Loh, Patterson, & Price, Reference Hujoel, Gazal, Loh, Patterson and Price2020)] based on parents and siblings. When we applied the LT-FH method to our Swedish sample that was restricted to parents and siblings and we eliminated our cohabitation correction in the FGRS to increase comparability between the methods, the correlations between the results of the two approaches were reassuring high: +0.94 (0.02).
Sixth, we simulated pedigrees of 1st thru 5th degree relatives with a sample size comparable to the Swedish population using a wide range of disease prevalences and proportion of liability due to genetic, shared environment, and unique environmental sources (online Supplementary Appendix Table 6). We initially analyzed, by our FGRS, results simulated using genes and unique environment. We saw the expected clear and strong association of FGRS value with heritability across a wide range of prevalences (online Supplementary Appendix Fig. 4a). Next, we simulated various proportions of shared environment and demonstrated that our cohabitation correction worked well in accounting for this environmental source of family resemblance (online Supplementary Appendix Figs 4b–d). That is, our correction for cohabitation built into the FGRS does appear to adequately control for shared environmental effects in siblings and parents, suggesting that our goal of only detecting genetic risk has likely been largely achieved.
Seventh, some of the changes in non-primary FGRS as a function of AAO, recurrence, etc. might arise from their correlation with the primary FGRS. Therefore, we repeated all the analyses presented in Figs 1–5 in online Supplementary Appendix Figs 5–9 controlling for the primary FGRS. As expected, the changes in the non-primary FGRS are generally attenuated, although in most analyses, the pattern of change is similar.
Conclusions
Individuals affected with DUD, MD, and ADHD, ascertained from the general Swedish population, have substantially different FGRS profiles when subdivided by AAO, recurrence, and modes of ascertainment and treatment. More modest differences were seen in the two sexes. Indices of clinical severity such as early AAO, high recurrence rates, detection in high-intensity clinical settings, and receipt of treatment were typically associated with high FGRS both for the primary disorder and for other disorders. FGRS often differed qualitatively as a function of mode of ascertainment. These results suggest that, with respect to genetic risk profiles, collection of comparable samples across research sites may require more than the use of similar diagnostic procedures. Other clinical features need to be considered in sample collection if replication is critical. Furthermore, we may need to reconsider sampling methods to enrich genetic risk.
It was difficult to find a disorder subgroup which demonstrated an increase only in the primary FGRS. Rather, subgroups with high primary FGRS were typically accompanied by elevations in the FGRS for other disorders. For example, while studying highly recurrent disorders enriches the patient sample for cases with high FGRS, this does not ‘come for free’. The entire genetic profile is sometimes more heterogeneous in severe than in milder cases of illness. Researchers should consider the relative advantages of ascertaining cases to increase the absolute magnitude of the primary genetic liability v. to maximize the ‘purity’ of the genetic profile.
Finally, positive genetic correlations between psychiatric disorders, traditionally assumed to reflect sharing of genetic risk factors, can arise in other ways. For example, an elevated FGRS for externalizing disorders, through raised levels of impulsivity, may increase the chances that an individual with MD would be hospitalized for treatment, especially if associated with self-harm. So, if we ascertained MD cases through hospitals with frequent admissions for suicide attempt, we would see a substantial genetic correlation between MD and externalizing disorders that would arise partly because of our mode of ascertainment.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291721005535
Acknowledgments
No disclosures. Morten Dybdahl Krebs, M.D. provided helpful statistical assistance.
Financial support
This project was supported by grant DA030005 from the National Institutes of Health, the Swedish Research Council as well as Avtal om Läkarutbildning och Forskning (ALF) funding from Region Skåne.
Conflict of interest
None.
