|Home | About | Journals | Submit | Contact Us | Français|
We make sense of human behavior using reasons, which produce understanding via a subjective empathy-based first-person perspective and causes, which leads to explanations utilizing objective facts about the world assessed scientifically. We evaluate the common sense hypothesis that for episodes of major depression (MD), reasons act as causes. That is, individuals who have highly understandable depressive episodes will have, on average, fewer objective scientifically validated causes than those who have un-understandable episodes. The understandability of a MD as defined by the Diagnostic and Statistical Manual, 4th Edition (DSM IV) experienced in the past year in 630 personally interviewed twins from a population-based registry was rated, with high reliability, from rich contextual information. We predicted, from these understandability ratings, via linear and logistic regression, 12 validated risk factors for MD reflecting genetic and long-term environmental liability. No significant association was observed between 11 of these indices and the understandability of the depressive episode. The only significant finding—higher cotwin risk for MD associated with greater understandability— was opposite that predicted by the reasons-as-causes hypothesis. Our results do not support the hypothesis that reasons for MD act as causes. These findings, unlikely to result from low power, may be explicable from an empirical and/or philosophical perspective. Our results are, however, consistent with ‘the trap of meaning’ hypothesis, which suggests that understanding does not equal explanation and that while reasons may be critical to help us empathize with our patients, they are unreliable indices of objective risk factors for illness.
We have two main approaches toward making sense of the behavior of our fellow humans. One approach relies on a first-person subjective perspective and provides reasons for behavior. As humans, we are capable of intuiting the reasons for others’ behavior through the process of empathy. Using the terminology adopted by Jaspers1 (p. 301) such reasons lead to understanding. If a close friend became depressed after a traumatic romantic break-up, we might feel: ‘Oh, of course. He was really in love with her.’ We feel that we understand the origin of his depression.
The second way we comprehend human behavior is by causes—that is, objective third-person scientific facts that are established by systematic observation and quantification. For example, studies have consistently found that patients with depression have a threefold increased risk of depression in their relatives.2 On this basis, we conclude that a positive family history is a contributing cause of major depression (MD). Using Jasper’s terminology, causes lead to explanation.
The question that we address in this paper is the relationship between reasons and causes for developing MD. Because understanding how reasons for MD might act as causes is so central to this paper, and might seem at first glance odd or abstract, we give two clinical vignettes to ‘ground’ this concept in clinical common sense.
Assume you are evaluating two patients each of whom presents with a depressive syndrome. Like most good clinicians, you carry in your head a list of empirically validated causes of such a presentation including medical disorders (for example, endocrine abnormalities), early environmental exposures (for example, childhood sexual abuse), genetic risk factors and potential psychological vulnerabilities (for example, high levels of neuroticism). Assume that in the early stages of your history taking, patient A tells you that his business recently burnt down with great financial loss and his beloved only child was diagnosed with a potentially fatal childhood cancer. Patient B, whom you know to be a reliable informant, tells you that everything has been going well in her social and family life, and at work, and her depression emerged ‘out of the blue.’ In the terms used in this article, patient A has good reasons to be depressed while patient B does not. That is, at the level of common sense psychology, the depressive episodes of patients A and B appear to be, respectively, quite understandable and rather un-understandable.
Many clinicians would, on hearing the story of patient A, conclude: ‘Yes, it makes sense to me why he became depressed.’ They would assume that the reasons given by this patient for his depression were causes that both ‘made sense’ to them as humans and constituted a causal explanation that could be used to guide further evaluation and treatment. Given that they felt they had a good explanation for the episode, they would not spend much further effort pursuing, by history or laboratory tests, other potential causes for this depressive episode. Furthermore, sensing that the depression arose understandably from external stressors, many clinicians might be willing to take a ‘watch and wait’ approach rather than developing a more aggressive treatment plan. By contrast, the workup would be more thorough for patient B because without good reasons for being depressed there ought to be some good causes to be found. Without a good psychological explanation for the emergence of the depression, an active pharmacological treatment approach would more likely be adopted.
The question that we address in this paper is the relationship between reasons and causes for developing MD. More specifically, we evaluate two hypotheses about this relationship: the reasons are causes and the trap of meaning hypotheses. The above clinical vignettes illustrate the implications of the ‘common-sense’ hypothesis that reasons are causes. An unambiguous prediction of this hypothesis is that, on average, individuals with many reasons for developing MD will have fewer causes. That is, in a group of depressed patients, we will observe a robust inverse relationship between the degree of understandability of their MD (which reflects the number and plausibility of their reasons for developing depression) and the presence of objective causes for MD (for example, personality, genetic loading and exposure to early environmental adversities).
By contrast, our alternative trap of meaning hypothesis assumes little to no relationship between reasons and causes. That is, while the reasons patients give us for the emergence of their disorders may help us to empathize with them and improve our emotional relatedness, they actually tell us little about the objective causes of their illness. A recent article about the trap of meaning3 forcefully argued that physicians do a disservice to their patients by assuming that understandable symptoms do not need further diagnostic evaluation or treatment. Specifically, the trap of meaning hypothesis predicts that in a group of depressed patients, little or no relationship would be found between the degree of understandability of their depressive episode and the presence of objective causes for MD as determined by their personality, their genetic loading and their exposure to early environmental adversities.
In this article, we evaluate the ‘reasons-as-causes’ versus ‘trap of meaning’ hypotheses for MD by examining detailed interviews with 630 twins who experienced an episode of MD in the last year, from a longitudinally assessed population-based registry. These interviews contained rich contextual information that permitted us to rate, with high reliability, the level of understandability (LOU) of their episodes. Furthermore, we have for this cohort detailed information about a range of empirically validated genetic and environmental indices of risk for MD. How would these causes for being at high risk for MD relate to the reasons for their current episode as reflected in our measures of understandability?
Participants derive from two inter-related investigations in Caucasian twin pairs in VATSPSUD,4 a sample ascertained from a population-based twin register in the Commonwealth of Virginia. Female–female (FF) twin pairs, born 1934–1974, became eligible if both members responded to a mailed questionnaire in 1987–1988, the response rate to which was ~64%. This sample was interviewed four times over 10 years, with cooperation rates ranging from 85 to 92%. Male–male and male–female (MMMF) pairs (born 1940–1974) were ascertained directly from registry records by a phone interview with a response rate of 72%. The second interview wave, conducted up to 5 years later, had an 83% response rate. The mean (s.d.) age and years of education of the twins at the first FF (FF1) and second MMMF (MM2) interview were 29.3 (7.7), 13.5 (2.1) and 37.0 (9.1), 13.6 (2.6), respectively. The FF1 and MM2 interviews were completed largely face–to-face while the other interview waves were conducted mostly by telephone. Signed informed or verbal consent was, respectively, obtained before all face-to-face and telephone interviews.
At each interview, we assessed the occurrence over the last year of 14 individual symptoms, representing the disaggregated nine ‘A criteria’ for MD in DSM-IV. For each symptom, we eliminated responses arising from physical illness or medication. The respondents aggregated their symptoms over the last year into co-occurring syndromes reporting months of onset and remission. The diagnosis of MD was made using DSM-IV criteria without the bereavement exclusion. Respondents were asked to report their ‘worst’ episode in the last year. It was that episode on which we focused in our review.
This project began by LJH and KSK together designing a simple scale (Table 1) that assessed the understandability of the depressive episode based on a review of the detailed contextual information available from these interviews. LJH then reviewed and prepared vignettes for all cases from the MM2 and FF1 interviews that met criteria for one or more episodes of MD in the past year with the goal of rating the understandability of the depressive episode. These interviews were very detailed and included a descriptive summary by the interviewer of the nature, context and key findings from the interview. Interview sections that were consulted included: (i) the subject’s determination as to whether the worst MD episode came ‘out of the blue’ or was attributed to something that happened, and if the latter, a description of what happened; (ii) a range of relationship and social support items, including quality of relationships with spouse/partner, children, other relatives and friends; number of confidants (and whether spouse/partner was one); and degree of involvement in clubs/organizations; (iii) employment status and work difficulties; (iv) financial status and financial problems; (iv) health status and any difficulties; (v) an extensive list of 15 categories of personal and network stressful life events (SLEs) that occurred in the last year and were dated to the nearest month, with a focus in our review of those that occurred during or shortly before the month of onset of the worst depressive episode of that year. In the MM2 but not FF1 interview, the interviewer coded every SLE on a four-point scale of long-term contextual threat.5
Table 2 contains examples of the vignettes prepared by LJH—two for each category. Some vignettes were randomly selected for inter-rater reliability trials. In the MM2 sample, 95 cases were jointly rated and the resulting weighted κ6 was +0.89 (95% confidence interval = 0.85–0.94). We also jointly rated 37 cases from the FF1 sample with similar results: weighted κ = +0.86 (0.77–0.95).
A total of 866 twins were selected as eligible for this study as they met criteria for MD at some point in the year before interview. In all, 630 were included in this study with the most common reason for exclusion being that the onset of the worst episode was before the last year, so we lacked detailed information about proximal stressors.
Lifetime MD was assessed at personal interview by DSM-III-R criteria in the cotwin and by family history using the Family History-Research Diagnostic Criteria in parents.7 Neuroticism was assessed by the short form of Eysenck’s Personality Questionnaire. 8 Generalized anxiety disorder was diagnosed using DSM-III-R9 criteria requiring 1 month minimum duration. Phobia was diagnosed using an adaptation of DSM-III criteria10 requiring one or more unreasonable fears that interfere with the respondent’s life. Panic disorder was defined using DSM-III-R criteria, except, because of the rarity of fully syndromal panic disorder,11 we allowed up to 30 min for symptoms to maximize. We calculated the risk for new depressive onsets within the 1-year prevalence window assessed by our interviews.
For early life risk factors, we examined self-report measures of parental warmth as measured by a shortened version of the Parental Bonding Instrument12 and childhood sexual abuse (defined as genital contact and/or attempted or completed intercourse). 13
Depending on the whether the dependent variables— all liability indices for MD—were dichotomous or continuous in distribution, we used, respectively logistic or linear regressions in PROC LOGISTIC and PROC REG in SAS14 with our five-point LOU scale as the key predictor variable. Our initial analyses included age, sex and a dummy variable representing the MMMF versus FF sample as covariates. Because of the highly skewed nature of the distribution, the variable number of prior episodes was reduced to a four category variable: zero, 1–2, 3–10 and > 10 prior MD episodes. Two-tailed P-values were reported. Because our sample contained both members of only 31 pairs, we did not formally correct for the correlational structure of the data.
Of the 630 individuals for whom the level of understandability (LOU) ratings were available, 11.9% were rated not understandable, 19.8% a little understandable, 30.8% somewhat understandable, 26.3% quite understandable and 11.1% totally understandable.
In Kendler et al.15 and many other samples,16–18 the association between SLEs and the onset of episodes of MD declines with increasing number of depressive episodes. This has sometimes been called ‘kindling’.16 The occurrence of such stressful events immediately before the depressive onset formed part of the information used for our LOU ratings. Given the potential importance of controlling for prior depressive episodes, we first examined their association with understandability. Number of prior episodes strongly and inversely predicted LOU (χ2 = 17.4, df = 1, P < 0.0001, odds ratio per category = 0.74, 95% confidence intervals = 0.65–0.85). In addition, age but neither sex nor FF versus MMMF cohort significantly predicted LOU with older age associated with lower LOU. Therefore, number of prior episodes and age were used as covariates for all subsequent analyses.
As seen in Table 3, we next examined the ability of our measures of LOU to predict 12 individual risk indices for MD organized into six groups.
The personality trait with the strongest association with risk for MD, mediated substantially by shared genetic factors, is neuroticism.19,20–24 LOU was unrelated to levels of neuroticism in our sample.
Epidemiological and twin studies show that MD and anxiety disorders together form a highly comorbid group of internalizing disorders which share a common genetic diathesis.25–27 No relationship was found between LOU in our participants and their risk for generalized anxiety disorder, panic disorder or phobias.
In Kendler et al.13,29,30 and other samples,31,32 the risk for MD was elevated in subjects exposed to low levels of parental warmth, childhood sexual abuse or childhood parental loss. However, LOU did not predict exposure to any of these adversities.
Familial/genetic factors have been conclusively shown to substantially contribute to liability to MD.2 LOU in this sample was unrelated to risk for MD in the father or mother. However, contrary to the prediction of the reasons-as-causes model, LOU was significantly and positively related to risk for MD in the cotwin.
As our sample was longitudinal, we were able to assess risk for subsequent depressive episodes, a self-evident index of liability to MD. LOU did not predict risk for future episodes of MD.
The goal of this paper was to evaluate empirically, in a large epidemiological sample of individuals with depressive episodes in the last year, two hypotheses about the relationship between reasons and causes for MD: ‘reasons as causes’ and ‘trap of meaning.’ Reasons were assessed by a reliable five-point scale reflecting the LOU of the depressive episode in the context of a rich description of the individual’s stressors and social resources in the last year. For causes, we examined 12 clinical, personality, genetic and environmental risk factors shown in this and other samples to predict episodes of MD.
Consistent with the trap of meaning but not the reasons–as-causes hypothesis, 11 of the 12 variables, including such prominent risk factors as levels of neuroticism, age at onset, comorbidity with anxiety disorders and history of MD in parents, were unrelated to LOU. The one remaining variable (risk for MD in cotwin) was, contrary to the predictions of the reasons-as-causes hypothesis, positively associated with ratings of LOU.
The reasons-as-causes hypothesis—that reasons can not only provide us understanding but also explanation—is deeply intuitive. But our results contradict its central prediction. Individuals with plausible reasons for developing depression did not have fewer scientifically validated causes for MD than individuals without such plausible reasons.
Although our results seem puzzling and may arouse skepticism, they are not without precedent. An older literature on ‘reactive’ or ‘situational’ MD shows that situational and non-situational depressed patients do not differ in personality,33 recovery rates34 or familial loading.35,36 A prior study in this sample found no relationship between many of the same indices of liability to MD and the level of long-term contextual threat of precipitating SLEs.37 Congruent with our results in cotwins, two prior studies found higher rates of familial loading for MD in relatives of either situational MD38 or cases of MD with onset associated with high levels of psychosocial stressors.39 In this same sample, we evaluated the understandability of the origin of phobias (for example, did they arise in response to severe traumas, mild traumas or without known environmental precipitants?) and found that this LOU had no relationship with genetic or personality risk factors for phobias.40
While our results contradicting the predictions of the reasons-as-causes hypothesis are clear-cut, their interpretation is not. We suggest one methodological, two empirical and one philosophical explanation for these findings. The methodological explanation is simply that our measure of LOU, while highly reliable and face-valid, is either just noise—signifying nothing— or too superficial to reflect anything useful about human psychological processes. We knew a good deal about the key ‘surface features’ of our subject’s lives— their health, interpersonal relationships, finances, job stressors and satisfaction, losses and conflict. We inquired specifically about what they thought caused their depression. So, we globally assessed both the array of stressors they had confronted and the social resources they had for coping with those stressors. However, we do not know in detail either their inner psychological workings or whether the specific details of their depressive symptoms relate thematically to the stressors that they experienced. So, for some of the cases that we rated as ‘un-understandable,’ a more in-depth inquiry might have yielded meaningful psychological connections between their life experiences and their depression.
We have two responses to this legitimate concern. First, as our vignettes in Table 2 illustrate, the kind of interview we conducted actually captured a large proportion of the losses, stressors and humiliations that reflect our common human understanding of the psychological origins of depression. We do not claim to have captured all of the relevant experiences, but to argue that our assessments are just noise seems implausible. Second, this interpretation is hard to square with the finding that, although number of prior episodes of depression were not mentioned in the case summaries and therefore had no role in our judgment of understandability, we found a strong inverse association between understandability and the number of past MD episodes. In light of this finding, it is difficult to argue that our measure of LOU is not indexing something important about depression.
We suggest two plausible empirical explanations for our findings. First, the adversity that the subjects report as occurring before their depressive episode (and which formed a substantial part of our understandability ratings) might not actually causally contribute to the depression. Instead, these stressful events and associated financial, interpersonal and medical difficulties could result either from risk factors that are shared with MD or reflect early signs of the depressive episode that was missed by our interview. This interpretation is consistent with some prior findings. A number of the risk factors for MD, including SLEs, low social support and marital conflict, are modestly heritable.41 The genetic risk factors for SLEs also impact on risk for MD.42 The association between event exposure and depressive onsets declines when controlling for genetic background43,44 showing that some proportion of the event-depression association is not causal but results from genetic factors which contribute both to risk for MD and to selection into stressful environments. However, these same studies have shown that this is not the entire story. Even when controlling entirely for genetic effects (studying monozygotic twins discordant for life event exposure), stressful events do still causally impact on risk for MD.43,44
Our second empirical explanation for these findings relates to the phenomenon of kindling. Subjects with high levels of risk to MD may have had earlier depressive episodes that were quite understandable. However, by the time we have seen them, their mind/brain had ‘learned’ to easily shift into a depressive state with little or no external stressors.15 Such cases characterized by high number of causes for MD but low understandability would certainly attenuate any evidence for the reasons-as-causes hypothesis. In our sample, 268 subjects denied earlier depressive episodes. If our findings resulted entirely from kindling, we should see the pattern of results predicted by the reasons-as-causes hypothesis in this first-onset subsample. We did not. In this subsample, our ratings of LOU did not predict lower rates of exposure to any of our 12 liability indices.
The philosophical interpretation of our findings is that reasons and causes may represent two incommensurable ways of knowing about human behavior. The distinction between these two approaches was clearly articulated by Wilhelm Dilthey, who, in Germany in the latter part of the nineteenth century, rejected the idea that everything useful about human behavior could be obtained through causes and explanation as studied by the natural sciences (Naturwissenschaften).45 Instead, he proposed a separate discipline of the human sciences (Geisteswissenschaften) where reasons and understanding were the dominant methodology. Dilthey’s conceptualization substantially influenced the psychiatrist/philosopher Jaspers1 and had a central role in his conceptual framework for psychopathology.
While our findings are contrary to the predictions of the reasons-as-causes hypothesis, they are entirely consistent with the trap of meaning hypothesis.3 While our empathic abilities may be indispensable in permitting us to see the world from another person’s perspective, they may not be a good guide to the causes of (at least) MD. While it may seem sensible to stop looking for objective causes of depressive illness when confronted with an intuitively plausible set of reasons, our results support the contention of Lyketsos and Chisolm3 that this approach may not optimally serve our patients.
As with any study with predominantly negative results, our findings should be interpreted in the context of the statistical power of our experiment. We performed linear and logistic regression with our relatively normally distributed five category predictor variable—LOU. Assuming as we did a two-tailed α level of 5% for linear regressions, we had, with our sample of 630 subjects,~70% power to detect a small effect size (r = 0.10) and over 99% power to detect anything larger.46 For logistic regression, we had over 80% power to detect ORs of ≤0.7 with the rarest of our risk factors—panic disorder and childhood sexual abuse—and 80% or more power at an odds ratio of ≤0.8 for all other variables. Thus, we were very well powered in this study to detect relationships between reasons and causes that were of moderate or large effect size but could have missed some subtle relationships. However, of our 11 non-significant findings, six were in the direction predicted by the reasons-as-causes hypothesis and five were not. These results are not consistent with an underlying trend in favor of the reasons-as-causes hypothesis that was missed due to inadequate power.
Supported in part by NIH grants MH-40828 and MH/AA/DA-49492. Dr James L Levenson provided helpful comments on earlier versions of this manuscript.
Conflict of interest
The authors declare no conflict of interest.