Using time-dependent Cox models and twin modeling, we sought to clarify the causal relationship between MD and CAD in a large population-based sample of Swedish twins in middle to late adulthood. Six findings are noteworthy. First, the lifetime association between MD and CAD in this sample was modest (OR, ~1.3) and did not differ substantially in men and women. Second, in more informative time-dependent analyses, CAD onset was associated with a nearly 3-fold increased risk for depressive onset in that year and a nearly 2-fold increase in subsequent years. The long-term effect of CAD on risk for MD did not attenuate over time. Because the reliability of the dating of CAD onset from hospital data may be limited, we repeated these analyses using the diagnosis of MI. Both associations strengthened, which suggested that we were more likely underestimating than overestimating the temporal MD-CAD association with our hospital-based CAD diagnoses.
Third, given an onset of MD, the risk for CAD onset was increased 2.5-fold in that year and much more modestly (OR, ~1.2) in subsequent years. The ongoing increased risk for CAD after MD onset did not attenuate over time. Although modest, this future risk for CAD was strongly related to the severity and recurrence of MD. Indeed, elevated future CAD risk was confined to individuals with recurrent episodes of MD or those who meet more than the minimal number of diagnostic criteria.
Fourth, the temporal pattern of the prediction of MD onset from CAD onset differed across sexes. In men, the increased risk for MD was much greater in the year of CAD onset than in subsequent years. Women had a smaller concurrent spike in risk for MD in the year of CAD onset, and the subsequent risk was of nearly the same magnitude. These differences became even more striking when we examined only MI. In men, the spike in concurrent risk for MD in the year of MI onset was far greater than the subsequent risk (5.40 vs 1.95); in women, the difference was much more modest (2.65 vs 2.29).
Fifth, when genetic risk factors were added to these Cox models, there was consistently strong evidence for “within-disorder” genetic effects. Genetic risks for MD and CAD were, respectively, robust predictors of onset of MD and CAD. In contrast, “cross-disorder” effects were generally small and nonsignificant.
Sixth, our twin modeling provided the best picture of the genetic and environmental sources of comorbidity for lifetime MD and CAD. The results were relatively complex, with many of the model parameters showing variance by year of birth. Consistent with our lifetime association findings, the overall magnitude of the comorbidity between MD and CAD was modest. In confirmation of our Cox results, differences were found in the source of this comorbidity by sex. In women, comorbidity was primarily owing to shared genetic effects, and this finding held across all age groups. In men, a similar pattern was seen only in the younger members of the sample. In the older men, genetic factors were, if anything, negatively correlated between the 2 disorders. Given the low prevalence rates for MD in these older men, this negative correlation should probably be regarded with some skepticism. Both our Cox and twin models, and the previous literature,11
show that genetic factors are more potent in early-than in late-onset CAD. In men, therefore, it is the earlier onset and more genetically influenced forms of CAD that have a positive genetic correlation with MD.
INTEGRATION OF THE SURVIVAL AND TWIN MODELS
How can we integrate the Cox and twin-modeling results to develop a broader view of the etiologic interrelationship between MD and CAD and the way that that relationship is modified by sex? Unlike the twin models, the Cox model provided us with a temporally dynamic picture of the MD-CAD relationship. Two features of these results are noteworthy. First, we observed an important asymmetry in our prediction of enduring risk. The onset of CAD predicted ongoing risk for MD (HR, 1.75) much more strongly than the onset of MD affected long-term risk for CAD (HR, 1.17). Any complete understanding of the causes of this key comorbidity will have to explain this important finding.
Second, the concurrent association of MD and CAD was consistently stronger in men than in women. A similar pattern was not seen with the enduring risks. Putting these results together, we see that, compared with women, a larger proportion of the MD-CAD comorbidity in men arose from etiologic processes that were short-acting rather than enduring. This factor is of interest because genetic influences on comorbidity are likely to have long-lasting effects. Environmental effects, in contrast, can be short-or long-lived. Therefore, our Cox and twin models are congruent in pointing toward one key sex difference in MD-CAD comorbidity. While our twin models show that genetic factors are more important in MD-CAD comorbidity in women than in men, our Cox models show that enduring effects of each disorder on each other (which are likely genetic) are relatively more important in women than in men. Our Cox models show that short-term effects of MD on CAD risk and CAD on MD risk are more potent in men than in women. As we take all parts of our sample into consideration, our twin models show that environmental effects (which likely have short-term effects) play a greater role in MD-CAD comorbidity in men than in women.
The methods of this study were sufficiently different from those of most of the extensive previous literature on MD and CAD to render direct comparisons difficult. However, it is instructive to compare our findings with a recent study that addressed the MD-CAD relationship using a different set of methods.35
Surtees et al35
assessed MD in 19 649 English subjects aged 41 to 80 years and followed them up for a median of 8½ years, and reported CAD-related deaths. First, they found that MD conveyed a 2.7-fold increased risk for CAD-related death, a figure somewhat higher than our concurrent HR (2.43). Our overall results, which combine our concurrent and persistent effects, are more in line with the OR of 1.60 estimated in a recent meta-analysis.8
Second, like us, Surtees et al35
found no overall difference in MD-CAD association across sexes, and this was not found in the previous meta-analysis.8
Third, broadly consistent with our own findings, Surtees et al35
reported a much greater risk for CAD when MD was reported to be present at the time of assessment vs in the past.
Our results do not agree with the one previous twin study of this issue,5
which found a stronger association between CAD and MD (OR, 4.03) and genetic correlation (+0.42) than we did. These differences could arise from several methodologic differences including the assessment of MD (diagnosis at personal interview vs symptom count), assessment of CAD (hospital summaries and death certificates vs self-report questionnaire), and age of the cohort (mean, 42 vs 57 years). In particular, onset of CAD before age 42 years is rare and atypical. Our results are broadly consistent with previous findings on the heritability of cardiac death,11
in Swedish twins, and sex differences in heritability of precursors of CAD.26
Our results have 2 implications for gene-finding efforts. First, given the modest genetic correlation, only a minority of risk genes for one of these disorders, that is, MD or CAD, are likely to affect risk for the other disorder. Although perhaps rare, such genes would be of great value in providing insight into the underlying pathophysiology of comorbidity. Second, if the goal is to uncover genes that affect MD-CAD comorbidity, it would be better to study women or early-onset CAD in men.
Results of the present study must be interpreted in the context of 4 potential methodologic limitations. First, results are limited to Swedish twins and may not extrapolate to other ethnic groups. Second, in the upper age ranges of our cohort, a substantial percentage of the sample had died or were too ill to be interviewed. This attrition was nonrandom because MD37
predispose to premature death. Such attrition is more likely to attenuate than exaggerate the MD-CAD association. Our Mx models are more sensitive to this bias than our Cox models, in which all comparisons occur within 5-year cohorts. However, we were able to increase the generalizability of these models by the inclusion in all the analyses presented here of co-twins of interviewed twins who were eligible for the SALT Study interview but did not complete an interview either because they were too ill or they refused. The addition of these twins produced little change in parameter estimates. Third, in individuals with MD and CAD onset in the same year, we did not know which preceded the other. Thus, we have less ability to infer a causal relationship in our estimates of concurrent onset than of subsequent onset. Fourth, the validity of our conclusions rests substantially on the quality of the CAD diagnoses in the Swedish IDR and CODR. Previous studies have shown high positive predictive values for the diagnosis of MI (96%38
) and other CAD diagnoses (81%)38
in the IDR and for CAD in the CODR (95%40
). Sensitivity has also been studied and found to be high for the diagnosis of MI in the IDR (94%39
) and for CAD in the CODR (94%41
In conclusion, although the MD-CAD relationship across the lifespan is modest, time-dependent models reveal stronger associations. The sustained effect of CAD onset on MD risk is much stronger than vice versa. The effect of MD on CAD is largely acute, and the longer term effects are apparently mediated via depressive recurrence. When examined separately, in men, environmental effects, which are often acute, have a large role in MD-CAD comorbidity, whereas in women, chronic effects, which are in part genetic, are more important. In men, genetic sources of MD-CAD comorbidity are more important in younger members of the sample.