Design

The study presented here is part of an on-going prospective high-risk cohort study among adolescent offspring of parents with bipolar disorder in The Netherlands. In this paper we discuss the findings from the first assessment. The study design, study population and prevalence of psychopathology among the offspring have been described by Wals et al (Reference Wals, Hillegers and Reichart2001). In brief, 86 parents with bipolar disorder and their spouses and 140 offspring aged 12-21 years were examined between November 1997 and March 1999. In the offspring, 38 (27%) were diagnosed with a lifetime mood disorder according to DSM-IV criteria (American Psychiatric Association, 1994), 23 (16%) had any other lifetime DSM-IV diagnosis and 79 (56%) did not have any DSM-IV diagnosis. We report the results of an analysis of retrospectively collected data from this cohort on the relationship between lifetime life events and subsequent lifetime mood disorder.

Stressful life events

The investigator-based Bedford College Life Events and Difficulties Schedule (LEDS; Brown & Harris, Reference Brown and Harris1978, Reference Brown, Harris, Brown and Harris1989) is a semi-structured interview for assessing life events and long-term difficulties in adults. The LEDS covers ten domains: education, work, reproduction, money/possessions, housing, crime/legal, health, romantic relationships, other relationships and miscellaneous events (including deaths). It collects detailed information about the event itself, the timing of its occurrence (date) and relevant contextual information for each event. Based on the contextual information, the threat for each event is rated via standardised rating procedures. The threat score represents the severity of the event, ranging from mild (1) to severe (4). Several studies have supported the reliability (e.g. interrater) and validity (e.g. multiple informant) of the LEDS with adults exhibiting a variety of psychiatric symptoms (Brown & Harris, Reference Brown and Harris1978, Reference Brown, Harris, Brown and Harris1989; Reference Ormel, Oldehinkel and BrilmanOrmel et al, 2001).

Monck & Dobbs (Reference Monck and Dobbs1985) originally adapted the LEDS methodology for use with adolescents. They developed a teenage LEDS manual with accompanying event dictionaries based on a study of 67 British female adolescents aged 15-20 years. We modified the Dutch adult LEDS interview and manual and translated the teenage event dictionaries into Dutch: the Kiddie LEDS (K-LEDS). The K-LEDS interviews were conducted by psychologists who had received K-LEDS training prior to interviewing. Because this K-LEDS interview covered the life cycle (childhood and early and late adolescence), all events and difficulties were dated on a yearly basis. In our analysis we used only the life event data because it was possible to date them more accurately than the long-term difficulties. We calculated the percentage fall-off of severe events recalled per annum, which appeared to be 11%. The events were rated from written transcriptions of the interview by three independent raters who had not been involved in the interviews and were masked to the respondents' mental health status. A panel consisting of the three raters and two of the authors (M.H., M.W.) reached consensus on the events that raised rating problems.

Schedule for Affective Disorders and Schizophrenia for School-Age Children - Present and Lifetime Version

All children were evaluated using the Schedule for Affective Disorders and Schizophrenia for School-Age Children - Present and Lifetime Version (K-SADS-PL; Reference Kaufman, Birmaher and BrentKaufman et al, 1997). The K-SADS is an interviewer-oriented diagnostic interview designed to assess current and past DSM-IV symptoms resulting in diagnoses in children and adolescents, by interviewing the parent(s) and child separately. The timing of illness episode onset was determined in cases where the DSM-IV criteria were fully met. If parents and child disagreed on the presence of a symptom, greater weight typically was given to parents' reports of observable behaviour and children's reports of subjective experiences (Reference Kaufman, Birmaher and BrentKaufman et al, 1997). The K-SADS-PL was conducted by three of the authors (M.H., M.W. and C.R.) and by five intensively trained interviewers with graduate degrees in psychology.

Family History Research Diagnostic Criteria

Parents were interviewed using the Family History Research Diagnostic Criteria (FH-RDC) (Reference Andreasen, Endicott and SpitzerAndreasen et al, 1977), which were used to calculate a continuous familial loading score for unipolar mood disorder, bipolar disorder and substance use disorder in first (n=177) and second-degree relatives (n=932) of the children. The index of family loading for the bipolar offspring is based on the number and age of the affected first- and second-degree relatives of the adolescent. Every relative examined using the FH-RDC contributed to the index, depending on whether the person was affected and the age at which the person was affected. We arbitrarily divided the continuous familial loading variable into high (> median) and low familial loading (< median). In our analyses we used the familial loading for unipolar mood disorder because all subjects had a first-degree family member with a bipolar disorder, consequently the familial loading for bipolar disorder did not differentiate. For a more detailed description of the calculation of the family loading, see Verdoux et al (Reference Verdoux, van Os and Sham1996) and Wals et al (Reference Wals, van Os and Reichart2004).

Time-dependent life event load

To study the impact of life events on the on-set of mood disorder, a time-dependent (or time-specific) life event load variable was calculated for every year of follow-up. This variable was designed to summarise the exposure load from all adverse events experienced up to a particular point in time, while accounting for number, severity and their lasting effects. The life event load was calculated according to four models. These models reflect different hypotheses concerning the time-related decay of the effect of a life event on the risk of mood disorder. In model I we tested a purely cumulative effect of the impact of life events on the development of mood disorder. Accordingly, the life event load at a particular point in time (year y) was simply calculated as the sum of the threat scores of the life events in year Y and all preceding years. In models II, III and IV, the time-dependent life event load was subjected additionally to an exponential decay function. This reflects the hypothesis that the impact of life events principally accumulates but at the same time gradually decays as time goes by. In model II the decay function implied a 25% loss per year. In models III and IV we subjected the life event load to a yearly decay of 50% and 75%, respectively. In view of the retrospective nature of data collection, we included only severe life events (threat score 3 and 4) that had occurred after the age of 4 years. Consequently, follow-up time started at age 5 years. If more than one life event occurred in the same year, the threat scores for these events were summed.

Data analysis

The relationship between life events and the occurrence of mood disorder was studied using a statistical model relating determinants whose statuses change over time to survival-type (censored) outcome data, i.e. Cox regression with time-varying covariates (Reference CoxCox, 1972). Because the 140 children originated from 86 families, data must be considered correlated through family. Therefore, we used a ‘frailty’ model, i.e. a Cox model with a cluster variable indicating family. Time-varying influence of life events was permitted by including the time-dependent life event load as a continuous time-varying covariate in the model. In this model the dependent variable was time from age 5 years to first mood disorder or, if no mood disorder occurred, time from age 5 years to interview. The results are expressed as hazard ratios indicating the instant relative risk of mood disorder per unit life event load, thus representing the strength of the association. Hazard ratios were presented with 95% confidence intervals (95% CIs). To find out which of the four life event load (decay) functions is optimally in agreement with the observed data, we compared Akaike's information criterion (-2 × maximised log-likelihood+ 3 × number of parameters) between the four regression models (Reference Akaike, Petrov and CsakiAkaike, 1973). This index can be interpreted only in a relative sense, i.e. lower values indicate better agreement. We included familial loading and gender as fixed covariates in the regression models to examine whether they confounded the association between life event load and onset of mood disorder. Confounding was considered present if inclusion of these variables substantially (by at least 10%) changed the hazard ratios for life event load. To investigate whether the relationship between life event load and mood disorder depended on familial loading (i.e. multiplicative interaction or effect modification), we included an interaction term of familial loading variable × life event load variable as a covariate in the model, and tested its statistical significance. Interaction was explored further by presenting separately the life event hazard ratios for children with familial loading above and below the median. The analyses of interaction were performed for the model showing optimal agreement with the observed data only. The level of significance in all analyses was P < 0.05 (two-sided).

Comparison with other studies

Few high-risk studies report the influence of stressful life events as a risk factor for the development of bipolar disorder. Johnson et al (Reference Johnson, Andersson-Lundman and Aberg-Wistedt2000) concluded that patients with bipolar disorder and with high constitutional vulnerability had an earlier age of onset and needed fewer stress factors (early parental separation and life events) to become ill compared with patients with unipolar illness. The Cardiff Depression Study (Reference Farmer, Redman and HarrisFarmer et al, 2002) investigated the suggested co-familiality of depression and life events and whether there might be a common familial factor influencing vulnerability to depression and the experience of life events. Using a sib-pair design, they reported no evidence for a common factor influencing both depression and life events. Kendler & Karkowski-Shuman (Reference Kendler and Karkowski-Shuman1997) showed that, in adults, negative life events were most likely to lead to the onset of major depressive disorder in individuals inferred to have a genetic liability to depression, and also that the genetic liability to depression overlaps with the genetic liability to experience stressful life events. So, through their behaviour, people can to some extent shape and select their environments.

Our findings are in line with the work of Wainwright & Surtees (Reference Wainwright and Surtees2002), who developed sophisticated analytical approaches to study adversity-disorder relationships. Their study, like ours, showed that the simplest model involving a single time-dependent covariate was in-appropriate because it failed to capture the decay in the event effects and that an exponential decay of the adverse effects of life events over time had to be modelled. In studying the effects of negative life events on the onset of mood disorders in a high-risk group of adolescents, Silberg et al (Reference Silberg, Rutter and Neale2001) found that there was no effect of independent life events on the adolescents' depression in the absence of parental emotional disorder but there was a significant effect in its presence. In our sample all subjects had a parent with bipolar disorder. As described, high familial loading was based on the number and age of unipolar affected first- and second-degree relatives of the adolescents.

Strengths and limitations

A major limitation of this study is that all data were collected at a particular point in time. Within a cross-sectional design we employed a longitudinal approach by dating the onset of episodes of mood disorders and the occurrence of life events. This approach brings a number of limitations in its wake. First, subjects with a mood disorder could have been more inclined to remember life events than those without this condition, which would result in recall bias; therefore, we restricted our analyses to severe life events and omitted the first 5 years of life. Also, the events were rated from written transcripts of the interview by three independent raters who had not been involved in the interviews and were masked to the respondents' mental health status. To explore the possibility of recall bias we divided the subjects with a mood disorder into current cases (i.e. at the time of the interview) and past cases, and compared the life events reported in the preceding 2 years. If recall bias played an important role, it would probably influence the current cases more than the past cases. The results of this analysis, however, showed that the mean threat scores were similar (4.2 and 4.0, respectively).

Second, one could question the validity of the LEDS used retrospectively to collect the life event data. Most of the studies concerning the validity of retrospective reports collected by the LEDS were restricted to a 12-month period. There have also been studies using the LEDS to test the validity of life events reported over a 10-year period. The ‘fall-off’ in terms of the length of time from the date of the reported event or difficulty to the point of interview was checked and found to be surprisingly low for all events; 4.8% per year (Reference Neilson, Brown, Marmot, Brown and HarrisNeilson et al, 1989). In our sample, as mentioned before, we calculated the percentage fall-off of severe events recalled per annum, which appeared to be 11%. Retrospective reporting of life events using checklist inventories typically declines at a rate of 5% or more each month (Reference Funch and MarshallFunch & Marshall, 1984). This suggests that in an aetiological enquiry it might well be possible to use the LEDS to cover a whole decade.

A further limitation might be that the sample is not population based. Only patients with children aged 12-21 years who were willing to participate were included. A control group of adolescents without a parent with bipolar disorder would have given more data to study the impact of stressful life events on the onset of mood disorders. Because the LEDS interview alone takes about 3 h and the rating takes another hour, financially this was not an option. Another limitation is that the group of adolescents with a mood disorder is relatively small in our sample. Consequently, the statistical power to demonstrate an interaction between life event load and familial loading was limited.

According to our study the impact of stressful life events principally accumulates but at the same time gradually decays (25% per year) as time goes by. This suggests that the effects of stressful life events do not simply add up or rapidly extinguish but, in a gradually fading fashion carry over into the future risk of an epidose of mood disorder. What drives the decay is not known, it might result from coping strategies or the effect of neutralising life events. Although high familial loading for unipolar depression was strongly related to risk of mood disorder, familial loading did not confound the relationship between life event load and mood disorder. There was also no evidence suggesting that familial loading modified the relationship between life event load and mood disorder.

Future directions

Improvements in the specification of stress modelling procedures might facilitate the integration of ideas from competing aetiological models of the onset and subsequent course of mood disorders. There are still many aspects of the stressful life event-illness relationship that should be considered in future studies: the underlying assumption of an additive effect of multiple life events, the possible dose-response effect of adverse life events and the existence of threshold effects. Other interesting topics for further research are the influence of life events on the duration and course of the mood disorders and the effects of comorbidity, temperament and specific coping skills. Goodyer (Reference Goodyer2002) referred also to limbic-cortical neural networks in his framework for future research on this topic.

CLINICAL IMPLICATIONS

1. ▪ Stressful life events increase the lifetime risk of a mood disorder in adolescent children of parents with bipolar disorder.

2. ▪ The level of familial loading for unipolar disorders does not affect the relationship between stressful life events and the onset of mood disorders in adolescent children of parents with bipolar disorder.

3. ▪ The impact of life events principally accumulates but at the same time gradually decays as time goes by.

LIMITATIONS

1. ▪ Life event data were collected retrospectively over more than 10 years.

2. ▪ No life event or psychopathology data were available for a control group of adolescents without a parent with bipolar disorder.

3. ▪ Recall bias as a result of psychopathology might be present.