Available online at www.sciencedirect.com
General Hospital Psychiatry 30 (2008) 407 – 413
Major depression as a risk factor for chronic disease incidence: longitudinal analyses in a general population cohort☆ Scott B. Patten, M.D., Ph.D. a,⁎, Jeanne V.A. Williams, M.Sc. a , Dina H. Lavorato, M.Sc. a , Geeta Modgill, B.A. a , Nathalie Jetté, M.D., Ph.D. b , Michael Eliasziw, Ph.D. a a
Department of Community Health Sciences, Faculty of Medicine, University of Calgary, Calgary, AB, Canada T2N 4N1 b Department of Clinical Neurosciences, University of Calgary, Calgary, AB, Canada T2N 4N1 Received 24 February 2008; accepted 1 May 2008
Abstract Objective: Cross-sectional studies have consistently reported associations between major depression (MD) and chronic medical conditions. Such studies cannot clarify whether medical conditions increase the risk for MD or vice versa. The latter possibility has received relatively little attention in the literature. In this study, we evaluate the incidence of several important chronic medical conditions in people with and without MD. Method: The data source was the Canadian National Population Health Survey (NPHS). The NPHS included the Composite International Diagnostic Interview Short Form to assess past-year major depressive episodes. The NPHS also collected self-report data about professionally diagnosed long-term medical conditions. A longitudinal cohort was interviewed every 2 years between 1994 and 2002. Proportional hazards models were used to compare the incidence of chronic conditions in respondents with and without MD and to produce age-, sex- and covariate-adjusted estimates of the hazard ratios. Results: The adjusted hazard ratios associated with MD at baseline interview were elevated for several long-term medical conditions: heart disease (1.7), arthritis (1.9), asthma (2.1), back pain (1.4), chronic bronchitis or emphysema (2.2), hypertension (1.7) and migraines (1.9). The incidences of cataracts and glaucoma, peptic ulcers and thyroid disease were not higher in respondents with MD. Conclusion: A set of conditions characterized particularly by pain, inflammation and/or autonomic reactivity has a higher incidence in people with MD. © 2008 Elsevier Inc. All rights reserved. Keywords: Major depressive episode; Depressive disorders; Longitudinal studies; Arthritis; Back pain; Bronchitis; Cataracts; Emphysema; Glaucoma; Heart disease; Hypertension; Migraines; Thyroid disease; Peptic ulcers
1. Introduction The association of chronic medical conditions with major depression (MD) has been well established by crosssectional studies [1–6]. While useful for descriptive purposes, cross-sectional estimates cannot clarify temporal relationships and are therefore difficult to interpret. In a few instances, longitudinal methods have been employed. Brown et al. [7] used administrative data to determine whether the ☆
This analysis was based on data collected by Statistics Canada. However, the analyses and interpretations presented do not reflect those of Statistics Canada. ⁎ Corresponding author. Tel.: +1 403 220 8752; fax: +1 403 270 7307. E-mail address:
[email protected] (S.B. Patten). 0163-8343/$ – see front matter © 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.genhosppsych.2008.05.001
incidence of diabetes is increased in people with MD, confirming earlier reports from the Baltimore Epidemiologic Catchment Area Follow-up Study [8]. A related analysis indicated that the incidence of MD was not increased in people with type II diabetes [9]. Migraines are another condition where longitudinal studies have been helpful in clarifying well-known crosssectional associations [3,4,10–12] with MD. In a longitudinal investigation, Breslau et al. [13] found that MD increased the incidence of migraines and also that migraines increased the incidence of MD. An effect of migraine on depression incidence, but no effect of depression on migraine incidence, was seen in another prospective study conducted in Baltimore [12].
408
S.B. Patten et al. / General Hospital Psychiatry 30 (2008) 407–413
Depressive symptoms (as opposed to depressive disorders) were found to be associated with subsequent hypertension incidence in two studies. One of these studies analyzed data from the National Health and Nutrition Examination Survey [14], and the other analyzed data from the CARDIA study [15]. The only study to examine hypertension incidence in association with depressive disorders was the Baltimore study [16]. Here, an adjusted odds ratio of 2.16, suggesting an elevated risk of hypertension, was reported; however, the association did not achieve statistical significance. A literature of cross-sectional studies has reported associations between back pain and depression [3,4,10]. Currie and Wang [17] used longitudinal data from two cycles of the Canadian National Population Health Survey (NPHS) to confirm whether there is an elevated incidence of MD in people with chronic back pain and also an elevated incidence of back pain in people with MD. The objective of the current study was to compare ageand sex-adjusted incidences of several long-term medical conditions in people with and without MD. Migraines and hypertension were considered to be of particular importance in view of the inconclusive nature of the literature for these conditions. Another objective was to replicate the elevation in back pain incidence reported by Currie and Wang [17]. On exploratory analyses, the incidence of several other conditions in respondents with and without MD was assessed.
2. Method The NPHS is a longitudinal study based on a nationally representative community sample assembled by Statistics Canada (Canada's national statistical agency) in 1994. Detailed information about the methods employed in this study may be found on the Statistics Canada Web page (www.StatCan.ca). The longitudinal cohort included 17,276 participants, but the current analysis was restricted to n=15,254 respondents who were over the age of 12 years at the time of the initial 1994 interview. The youngest of these respondents would have been approximately 18 years old in 2002 at the end of the follow-up interval examined in this study. The NPHS cohort has been interviewed every 2 years, such that five interviews are available for the period 1994–2002. The NPHS interview includes a series of items concerned with long-term medical conditions. Study participants were read a list of chronic medical conditions and asked whether they had been diagnosed with one of these conditions by a health professional. The wording of the relevant item was: “Now I would like to ask about certain chronic health conditions that you may have. We are interested in long-term conditions that have lasted, or are expected to last, 6 months or more and that have been diagnosed by a health professional.” This was followed by a series of specific queries (e.g., “Do you have high blood pressure?”).
In the case of back pain, the NPHS item referred to “back problems” and drew a distinction between other related conditions in the survey. The wording of the item was: “Remember, we are interested in conditions diagnosed by a health professional. Do you have back problems, excluding fibromyalgia and arthritis?” The methodological approach of eliciting self-reported but professionally diagnosed chronic conditions is employed in the US Behavioral Risk Factor Survey and in the National Health and Nutrition Examination Survey. Several studies have explored the validity of the approach. Most of these validation studies have focused on hypertension and diabetes. The validity of diabetes selfreports is not considered here, as the NPHS dataset was not large enough to assess the incidence of this condition. Martin et al. [18] reported that the sensitivity of selfreported professional diagnoses of hypertension in relation to a medical record review was 83% and that the specificity was 81%. El Fakiri et al. [19] reported kappa coefficients ranging from 0.63 to 0.51 (depending on the ethnic group) for self-reported and medical-record-confirmed diagnoses of hypertension. When biometric data rather than medical records were used as a validation standard, self-reports were found to be highly specific, but lower levels of sensitivity have been observed (e.g., 34.5% in one study [20] and 49% in another [21]). However, these studies measured blood pressure only during a single assessment. Vargas et al. [22] used six measures taken on two occasions, being more consistent with clinical practice, and reported much more favorable (75%) sensitivity and (95%) specificity for self-report. Similarly, a Spanish study [23] evaluated 79 members of a cohort of university graduates who self-reported that they had hypertension. Additional data collection confirmed the diagnosis in 82.3% of cases. In a sample of 48 participants who reported that they did not have hypertension, this was confirmed in 85.4%. A broader set of conditions has been associated with higher levels of agreement between selfreport and clinical diagnoses (myocardial infarction, asthma and congestive heart failure, among others) than for hypertension [24]. For example, a concordance rate of 86.9% for osteoarthritis and a concordance rate of 96.1% for rheumatoid arthritis were reported by Barlow et al. [25] in a sample of rheumatology outpatients. The NPHS interview included the Composite International Diagnostic Interview Short Form (CIDI-SF) [26] for MD, which assesses past-year major depressive episodes. The CIDI-SF is scored with a predictive probability based on the number of symptom-based criteria fulfilled during the same 2-week period in the past year and also requires that depressed mood or loss of interest or pleasure, or both, be reported. The instrument was scored at the 90% positive predictive value cutpoint in these analyses. In addition to these measures, the NPHS collected data on height and weight from which body mass index (BMI) could be calculated. Obesity, defined as BMI≥30, was included as
S.B. Patten et al. / General Hospital Psychiatry 30 (2008) 407–413
a covariate in some parts of the analysis. The NPHS also had a family history module that included items concerned with a family history of heart disease in first-degree relatives. In addition, the interview included items assessing smoking and physical activity. Sedentary lifestyle was defined using a formula developed by the Canadian Fitness and Lifestyle Institute (http://www.cflri.ca). Each of the 20 recreational activities evaluated in the NPHS was assigned a metabolic indicator value representing energy expenditures, with a past-year average expenditure of less than 1.5 kcal/kg/day being representative of a sedentary lifestyle pattern. This level of activity corresponds to approximately 30 min of walking as exercise per day. The survey also collected data on mental health care utilization. In this study, having two or more physician visits during the past year was used to describe nonroutine health care use. The NPHS sample is representative of the target population of household residents in Canada. The baseline interview was conducted in person, but most of the follow-up interviews were completed by telephone. The NPHS master files are released to researchers through a national system of regional data centers. However, Statistics Canada asks that researchers adhere to data release guidelines involving minimal allowable cell sizes and coefficients of variation. Coefficients of variation depend on several factors, including the parameters being estimated and the size and characteristics of the at-risk group. In the case of certain conditions, most notably stroke and diabetes, it was not possible to generate releasable incidence comparisons, so these conditions were not included in the analysis. The goal of this study was the assessment of incidence, which implies the new occurrence of a disease. Therefore, for each chronic condition under evaluation, those having that condition at the baseline interview were excluded from incidence analyses for that specific condition. The remaining at-risk population was followed for the ensuing 8 years (unless they developed the outcome, died or were lost to follow-up before this time). After exclusion of respondents with specific chronic conditions at baseline, the remaining “at-risk” respondents were divided into depressed and nondepressed cohorts. This was performed in two ways. First, CIDI-SF-defined MD at the baseline (1994) interview was used to define the depressed cohort. The nonexposed cohort then consisted of “at-risk” respondents without MD at the baseline interview. Second, MD was treated as a timevarying factor, so that MD status at the start of each 2-year incidence interval determined whether a respondent was in the exposed or in the nonexposed cohort during that interval. We modeled the effect of MD on chronic medical condition incidence using a proportional hazards model. Because the NPHS employed interviews at discrete (2-year) time points, a proportional hazards model for grouped time data was used. There were four discrete time intervals in these analyses, defined by the five interviews with the cohort members. The proportional hazards models were fitted as generalized linear models of the binomial family
409
with a complementary log–log link function, as described by Prentice and Gloeckler [27]. Jenkins [28] has outlined procedures for the implementation of these analyses in Stata (http://www.iser.essex.ac.uk/teaching/degree/stephenj/ ec968/pdfs/STB-39-pgmhaz.pdf). Whereas Jenkins describes both parametric and nonparametric alternatives, the analyses presented here are nonparametric. After preliminary tabular and stratified analyses, crude (unadjusted) hazard ratios (HRs) were calculated. Next, age and sex were added to the models. Age was depicted using dummy variables for five age groups: 12–18, 19–25, 26–45, 46–65 and 66+ years. In some analyses, these age intervals were collapsed in order to ensure an adequate number of observations in each age stratum. In some analyses, additional covariates were included, but a detailed etiologically oriented analysis was not conducted. Ageand sex-by-exposure interaction terms were assessed in all analyses. In the proportional hazards models, respondents who were lost to follow-up were censored, so they were included in the analysis as long as they were successfully followed. The NPHS does not collect data from institutionalized respondents, so these were also censored from the analysis at their time of institutionalization. The NPHS collects causeof-death information for those respondents who died during follow-up. In most analyses, these deaths were treated as censored observations. However, in the analysis of heart disease incidence, these were treated as outcome events when the cause of death was cardiac (International Classification of Diseases, Tenth Revision: I20–I25 and I50–I50.9). The NPHS used a multistage stratified design that also included clustering to select eligible households. To correct for bias potentially resulting from this complex survey design, Statistics Canada recommends a bootstrap procedure that uses a set of replicate weights. All analyses were conducted at the Prairie Regional Data Center at the University of Calgary campus, using Stata [28]. The study received approval from the University of Calgary Conjoint Ethics Board.
3. Results The annual prevalence of MD at the baseline interview was 5.7%. The rate of loss to follow-up among the eligible respondents was 20.4% over 8 years. This was slightly higher in those with MD at baseline [25.5%; 95% confidence interval (95% CI)=21.6–29.5] compared to those without (19.1%; 95% CI=18.3–20.0). Initial analyses adopted a definition of MD based on fulfillment of the diagnostic criteria at the time of the 1994 (baseline) interview. Using this definition, Table 1 presents three HRs for each chronic condition included in the analysis: (a) unadjusted HRs, (b) age- and sex-adjusted HRs and (c) estimates adjusted for age, sex and having two or more physician visits during the
410
S.B. Patten et al. / General Hospital Psychiatry 30 (2008) 407–413
Table 1 HRs for chronic condition incidence, by major depression at baseline interview
Migraines a Hypertension Back problems Arthritis/rheumatism Asthma Cataracts/glaucoma Chronic bronchitis/emphysema Heart disease Peptic ulcers Thyroid disease a
Unadjusted HR
95% CI
HR adjusted for age and sex
95% CI
HR adjusted for age, sex and health care use
95% CI
1.9 1.3 1.4 1.5 2.1 0.5 2.2 1.3 1.5 1.4
1.0–3.8 1.0–1.7 1.1–1.8 1.2–1.9 1.6–3.0 0.4–0.8 1.6–2.9 0.9–1.9 1.0–2.3 0.8–2.3
1.5 1.7 1.4 1.9 2.0 0.8 2.4 1.6 1.6 1.5
0.7–3.1 1.3–2.3 1.2–1.8 1.5–2.4 1.5–2.8 0.6–1.2 1.7–3.3 1.1–2.4 1.1–2.5 0.9–2.5
1.4 1.6 1.4 1.7 1.8 0.8 2.1 1.4 1.5 1.5
0.7–2.9 1.2–2.2 1.1–1.7 1.3–2.2 1.3–2.5 0.5–1.1 1.5–2.9 1.0–2.1 1.0–2.2 0.9–2.5
Restricted to respondents under the age of 26 years because of a significant age-by-MD interaction.
CI=0.2–0.7) in the age category 12–25 years to 9.5% (95% CI=7.4–9.7) in the age category 66+ years. The unadjusted HR for MD at the baseline interview was 1.3 (95% CI=1.0– 1.7), which was not significantly elevated (P=.07). However, with adjustment for age and sex, the HR increased to 1.7 (95% CI=1.3–2.3) and was statistically significant (Pb.001). Similarly, with treatment of MD status as a time-varying factor, the unadjusted HR was 1.0 (95% CI=0.8–1.3; P=.87). However, with adjustment for age and sex, the HR increased to 1.4 (95% CI=1.1–1.9), indicating significantly (P=.01) elevated risk. Adjustment for health care use did not substantially alter these results (see Tables 1 and 2). There were 12,995 respondents who did not report back problems at the time of the baseline interview in 1994 and who were therefore considered at risk for incident back problems. Among those with MD at the baseline interview, there was an 11.2% two-year incidence of back problems during the initial follow-up interval (95% CI=8.4–14.1), compared to 7.8% (95% CI=7.2–8.4) in those without MD at baseline. These frequencies represented 1072 new-onset cases during the interval 1994–1996. Incidences in this interval were similar in men (7.4%; 95% CI=6.6–8.2) and in women (8.6%; 95% CI=7.7–9.4). It was lowest in the age category 12–18 years (4.6%; 95% CI=3.2–6.0) and highest in the age category 46–65 years (9.9%; 95% CI=8.7–11.2). Elevated HRs were observed in association with MD, and
preceding year. Table 2 presents the same analyses, but with MD defined as a time-varying characteristic. There were 14,084 respondents who were at risk for incident migraines in 1994. There were 357 respondents who reported new-onset migraines between the baseline interview and the 1996 interview. In the various NPHS cycles, migraine incidence ranged from 2.1% to 4.8%. However, an age-by-MD interaction was observed in preliminary analyses. This interaction suggested that an association between MD and migraines was present only in respondents under the age of 26 years. Therefore, further analyses were restricted to this group. Unadjusted HRs were elevated in both sets (baseline and time varying) of analyses. In the former — but not in the latter — part of the analysis, the HR diminished with adjustment for age, sex and health care use (see Tables 1 and 2). There were 13,581 respondents who were at risk for developing hypertension at the 1994 baseline NPHS interview. In the first 2 years of follow-up, 490 respondents reported new-onset high blood pressure. The 2-year incidence was nearly identical in those who had MD at the baseline interview (3.1%; 95% CI=1.8–4.5) as in those without (3.2%; 95% CI=2.8–3.6). During the initial interval of follow-up, there were similar 2-year incidences in men (3.1%; 95% CI=2.6–3.6) and in women (3.4%; 95% CI=2.8– 3.8), but there was an increase with age from 0.5% (95%
Table 2 HRs for chronic condition incidence, by major depression as a time-varying characteristic
Migraines a Hypertension Back problems Arthritis/rheumatism Asthma Cataracts/glaucoma Chronic bronchitis/emphysema Heart disease Peptic ulcers Thyroid disease a
Unadjusted HR
95% CI
HR adjusted for age and sex
95% CI
HR adjusted for age, sex and health care use
95% CI
2.8 1.0 1.4 1.1 2.2 0.6 2.8 1.1 2.0 1.3
1.6–5.0 0.8–1.3 1.1–1.7 0.9–1.4 1.5–3.1 0.4–1.0 2.1–3.9 0.7–1.6 1.3–3.0 0.8–2.2
2.3 1.4 1.5 1.5 2.0 1.1 3.2 1.4 2.2 1.4
1.3–4.1 1.1–1.9 1.2–1.8 1.2–2.0 1.4–2.9 0.7–1.8 2.3–4.5 0.9–2.1 1.4–3.3 0.8–2.4
2.1 1.3 1.3 1.3 1.7 0.9 2.6 1.2 1.8 1.3
1.2–3.6 1.0–1.7 1.1–1.6 1.0–1.7 1.2–2.4 0.6–1.5 1.9–3.7 0.8–1.8 1.2–2.8 0.8–2.3
Restricted to respondents under the age of 26 years because of a significant age-by-MD interaction.
S.B. Patten et al. / General Hospital Psychiatry 30 (2008) 407–413
these were not altered by adjustment for age, sex or health care use (see Tables 1 and 2). There were 11,724 respondents who did not report arthritis at the baseline interval and were considered at risk for incident arthritis. In the initial follow-up interval, there were 785 new diagnoses of arthritis reported. Among respondents with MD in 1994, there was a 6.4% new incidence by the 1996 interview (95% CI=4.3–8.5), which was similar to that seen in respondents without MD at baseline (6.0%; 95% CI=5.4–6.5). The HRs increased with adjustment for age and sex. The association was not substantially diminished after adjustment for health care use (see Tables 1 and 2). Additional modeling determined that female sex (HR=1.5; 95% CI=1.3–1.7) and obesity (defined as BMI≥30; HR=1.6; 95% CI=1.3–1.9) were associated with arthritis incidence. However, inclusion of obesity did not alter the association of MD with arthritis incidence. For example, when MD was modeled according to its presence at the baseline interview, the adjusted HR from a model including age, sex and obesity was 1.9 (95% CI=1.5–2.4). There were 14,278 respondents who were at risk for asthma because they did not report asthma during the 1994 interview. In the first 2 years of follow-up, there were 283 new cases. In proportional hazards models, the results were nearly identical when MD was defined at baseline and when it was treated as a time-varying factor. In each case, the HR was indicative of an approximate doubling of risk and was not substantially altered by adjustment for age, sex or health care use (see Tables 1 and 2). In view of previous reports that childhood traumatic experiences may be associated with asthma incidence [29], we included this variable in a subsequent set of models. In both sets of models, the HR for childhood stressors was elevated at 1.4 (95% CI=1.1–1.7), but its inclusion did not change the estimated effect of MD. There were 14,682 respondents who were at risk for chronic bronchitis and emphysema in 1994, and 248 new cases had emerged by 1996. Strong associations between MD and the incidences of these conditions were observed, and these persisted after adjustment for age, sex and health care use (see Tables 1 and 2). Smoking was also added to the analyses as a covariate and was found to be strongly associated with incidence (HR=2.1; 95% CI=1.6–2.6). Inclusion of smoking as a covariate, however, did not substantially change the HR for MD: the age-, sex- and smoking-adjusted HR was 2.2 (95% CI=1.6–3.0) using the baseline definition and was 2.9 (95% CI=2.1–4.1) for the time-varying definition. The part of the analysis that was concerned with heart disease incidence was restricted to 5692 respondents over the age of 45 years who did not report heart disease at the baseline interview in 1994. By the first follow-up interview in 1996, 301 respondents self-reported new-onset heart disease. With inclusion of cardiac deaths (International Classification of Diseases, Tenth Revision: I20–I25 and I50–
411
I50.9; see explanation above), the total incidence in the first 2 years of NPHS follow-up was 6.6% (95% CI=5.7–7.4). The crude HR for MD as assessed at baseline was 1.3. With adjustment for age, sex and health care use, the HR increased slightly when the baseline definition of MD was used (see Table 1), but the part of the analysis treating MD as a timevarying characteristic uncovered no association between MD and heart disease incidence (see Table 2). The lack of strong evidence for an association may have been due to uncontrolled confounding. For this reason, we developed additional models that included cardiac risk factors such as age, sex, obesity, physical activity, smoking, high blood pressure, diabetes and family history of heart disease. In these models, no significant association of obesity or smoking with heart disease incidence was found, but age (N65 years; HR=2.8; 95% CI=2.2–3.6), female sex (HR=0.7; 95% CI=0.5–0.8), sedentary lifestyle (HR=0.6; 95% CI=0.5–0.8), hypertension (HR=1.6; 95% CI=1.2–2.0), diabetes (HR=2.2; 95% CI=1.6–3.2) and family history of heart disease (HR=2.2; 95% CI=1.8–2.7) were associated with the incidence of heart disease. In a model adjusting simultaneously for these variables, the HR for MD (as defined at baseline) was 1.7 (95% CI=1.0–2.8; P=.03). However, there was no association when the time-varying definition was used (HR=1.0; 95% CI=0.6–1.8). There were 14,646 respondents who were at risk for ulcers in 1994, and 212 of these reported having a diagnosis of ulcers 2 years later in 1996. A significant association that persisted after adjustment for age, sex and health care use was observed (see Tables 1 and 2). There was no association between MD and thyroid disease or cataracts and glaucoma (see Tables 1 and 2). Since longitudinal epidemiologic data can contribute to etiologic judgment, the strength of association is of some interest [30]. Unfortunately, the CIDI-SF does not provide a retrospective rating of the severity of past-year episodes. However, the instrument does include an assessment of episode duration, which has some intrinsic value as a metric for severity and is also correlated with episode severity [31,32]. Table 3 fails to show conclusive evidence that HRs
Table 3 Unadjusted HRs, by duration of past-year MD episode 95% CI Depressed for 95% CI Depressed for 2–12 weeks 13–52+ weeks a Migraines a Hypertension Back problems Arthritis/rheumatism Asthma Chronic bronchitis/ emphysema Heart disease Peptic ulcers
1.7 1.0 1.3 1.2 2.1 1.9
0.7–3.8 0.7–1.4 1.0–1.7 0.8–1.7 1.4–3.1 1.3–2.9
1.8 2.0 1.5 2.2 2.4 2.5
1.0 2.2
0.5–2.0 1.6 0.9–5.3 2.2
0.4–7.8 1.2–3.3 1.0–2.1 1.5–3.3 1.3–4.5 1.3–5.0 0.9–3.1 0.5–10.6
a Restricted to respondents under the age of 26 years because of a significant age-by-MD interaction.
412
S.B. Patten et al. / General Hospital Psychiatry 30 (2008) 407–413
are increased to a greater extent in association with longer episodes, but this may be due to loss of precision as a result of the stratification employed. For most of the conditions, the point estimates of the HRs are higher in those reporting longer episodes.
4. Discussion This analysis identified associations between MD and the incidence of a variety of medical conditions. The associations appear not to be due to confounding by age or sex. The associations observed do not appear to be an artifact of health care use, as they persisted in each case after adjustment for an indicator of this variable. As noted above, the literature has been inconclusive as to whether MD increases the risk of migraines. The NPHS data provide evidence that an association with migraine incidence exists, but that the strength of association depends on age and was only evident in young respondents in this study. The association was more evident when MD was treated as time varying than when MD was defined according to the baseline interview, but the two analyses were otherwise consistent, with the same direction of association being seen in both analyses. In the case of hypertension, the NPHS provides evidence of an association, although this was not evident in unadjusted analyses. The emergence of an association after adjustment for age and sex is not surprising, since hypertension is more common in older age groups whereas MD is more common in young people. Confounding by age is expected to occur and is expected to be in a negative direction in these circumstances. MD was associated with an increased incidence of arthritis and peptic ulcer disease. Like migraines and back problems, these conditions are typically painful. These findings raise the possibility of a nonspecific relationship between MD and painful conditions. Each of these conditions exists on a broad spectrum of severity and may not be consistently diagnosed by health professionals. By magnifying pain or by increasing the rate of consultation with health professionals, MD may increase the likelihood of being diagnosed with these conditions. This could introduce bias that could result in overestimation of the relevant associations, but the results of the study did not specifically support this possibility, since adjustment using an indicator of health care use did not generally alter the strength of association observed. These and other medical conditions may not always be accurately reported on self-report measures [33]. This could result in nondifferential misclassification bias, which would tend to be in the direction of the null value for the HR [34]. The conditions found to be associated with MD in these analyses are diverse. Several of them are associated with inflammation (heart disease, chronic lung disease and asthma). Arthritis could also be added to this list, although
the vast majority of arthritis cases detected would have been osteoarthritis rather than rheumatoid arthritis. There have been reports of elevated levels of inflammatory markers in MD (see review by Glassman and Miller [35]), so it is possible that physiological mechanisms may contribute to the observed associations. The NPHS did not collect detailed information about the timing of events and cannot fully clarify this aspect of the epidemiology. However, by excluding those respondents who already had the various medical conditions at baseline, it was possible to clarify the temporal relationships to a large extent. The measure of MD employed in the NPHS was brief relative to the full version of the CIDI employed in many psychiatric epidemiologic studies. This may have introduced some misclassification, probably also nondifferential, and may potentially have resulted in a dilution of the strength of some of the observed effects. Another threat to the validity of these estimates is the issue of loss to follow-up. The rate of loss to follow-up was only slightly lower in the MD cohort, so substantial bias is unlikely to result [36–38]. Cross-sectional associations between medical conditions and MD are well known, and there may have been a tendency to assume that these associations are due to an effect of the medical conditions on the incidence of depression. These results show that, for many chronic conditions, effects in the other direction also occur. For completeness, it should be added that the cross-sectional associations can theoretically arise from effects on illness persistence, as well as on mortality. MD must increasingly be viewed either as a chronic disease risk factor or as an indicator of other factors that are in turn related to the risk of developing chronic medical conditions. Acknowledgments Dr. Patten is a health scholar at the Alberta Heritage Foundation for Medical Research (www.ahfmr.ab.ca). Dr. Eliasziw is a senior scholar at the Alberta Heritage Foundation for Medical Research. This work was supported by a grant from the Canadian Institutes for Health Research. References [1] Wells KB, Golding JM, Burnam MA. Psychiatric disorder in a sample of the general population with and without chronic medical conditions. Am J Psychiatry 1988;145:976–81. [2] Moldin SO, Scheftner WA, Rice JP, Nelson E, Knesevich MA, Akiskal H. Association between major depressive disorder and physical illness. Psychol Med 1993;23:755–61. [3] Gagnon LM, Patten SB. Major depression and its association with long-term medical conditions. Can J Psychiatry 2002;47:167–73. [4] Patten SB. Long term medical conditions and major depression in the Canadian population. Can J Psychiatry 1999;44:151–7. [5] Lindeman S, Hämäläinen J, Isometsä E, Kaprio J, Poikolainen K, Heikkinen M, et al. The 12-month prevalence and risk factors for major depressive episode in Finland: representative sample of 5993 adults. Acta Psychiatr Scand 2000;102:178–84.
S.B. Patten et al. / General Hospital Psychiatry 30 (2008) 407–413 [6] Braam AW, Prince MJ, Beekman ATF, Delespaul P, Dewey ME, Geerlings SW, et al. Physical health and depressive symptoms in older Europeans. Results from EURODEP. Br J Psychiatry 2005;187:35–42. [7] Brown LC, Majumdar SR, Newman SC, Johnson JA. History of depression increases risk of type 2 diabetes in younger adults. Diabetes Care 2005;28:1063–7. [8] Eaton WW, Pratt L, Armenian H, Ford DE, Gallo J. Depression and risk for onset of type II diabetes. Diabetes Care 1996;19:1097–102. [9] Brown LC, Majumdar SR, Newman SC, Johnson JA. Type 2 diabetes does not increase risk of depression. Can Med Assoc J 2006;175:42–6. [10] Patten SB, Beck CA, Kassam A, Williams JVA, Barbui C, Metz LM. Long-term medical conditions and major depression: strength of association in the general population. Can J Psychiatry 2005;50: 195–202. [11] Molgat CV, Patten SB. Comorbidity of major depression and migraine — a Canadian population-based study. Can J Psychiatry 2005;50:832–7. [12] Swartz KL, Pratt LA, Armenian HK, Lee LC, Eaton WW. Mental disorders and the incidence of migraine headaches in a community sample. Arch Gen Psychiatry 2000;57:945–50. [13] Breslau N, Lipton RB, Stewart WF, Schulz LR, Welch KMA. Comorbidity of migraine and depression. Neurology 2003;60:1308–12. [14] Jonas BS, Franks P, Ingram DD. Are symptoms of anxiety and depression risk factors for hypertension? Longitudinal evidence from the National Health and Nutrition Examination Survey I Epidemiologic Follow-up Study. Arch Fam Med 1997;6:43–9. [15] Davidson K, Jonas BS, Dixon KE, Markovitz JH. Do depression symptoms predict early hypertension incidence in young adults in the CARDIA study? Arch Intern Med 2006;160:1495–500. [16] Meyer CM, Armenian HK, Eaton WW, Ford DE. Incident hypertension associated with depression in the Baltimore Epidemiologic Catchment Area Follow-up Study. J Affect Disord 2006;83:127–33. [17] Currie SR, Wang JL. More data on major depression as an antecedent risk factor for first onset of chronic back pain. Psychol Med 2005;35: 1275–82. [18] Martin LM, Leff M, Calonge N, Garrett C, Nelson DE. Validation of self-reported chronic conditions and health services in a managed care population. Am J Prev Med 2000;18:215–8. [19] El Fakiri F, Bruijnzeels MA, Hoes AW. No evidence for marked ethnic differences in accuracy of self-reported diabetes, hypertension, and hypercholesterolemia. J Clin Epidemiol 2007;60:1271–9. [20] Molenaar EA, Van Ameijden EJ, Grobbee DE, Numans ME. Comparison of routine care self-reported and biometrical data on hypertension and diabetes: results of the Utrecht Health Project. Eur J Public Health 2007;17:199–205. [21] Goldman N, Lin IF, Weinstein M, Lin YH. Evaluating the quality of self-reports of hypertension and diabetes. J Clin Epidemiol 2003;56: 148–54. [22] Vargas CM, Burt VL, Gillum RF, Pamuk ER. Validity of self-reported hypertension in the National Health and Nutrition Examination Survey III, 1988–1991. Prev Med 1997;26:678–85.
413
[23] Alonso A, Beunza JJ, Gado-Rodriguez M, Martinez-Gonzalez MA. Validation of self reported diagnosis of hypertension in a cohort of university graduates in Spain. BMC Public Health 2005; 5:94. [24] Tisnado DM, Adams JL, Liu H, Damberg CL, Chen WP, Ashlee F, et al. What is the concordance between the medical record and patient self-report as data sources for ambulatory care. Med Care 2006;44: 132–40. [25] Barlow JH, Turner AP, Wright CC. Comparison of clinical and selfreported diagnoses for participants on a community-based arthritis selfmanagement programme. Br J Rheumatol 1998;37:985–7. [26] Kessler RC, Andrews G, Mroczek D, Ustun B, Wittchen HU. The World Health Organization Composite International Diagnostic Interview Short-Form (CIDI-SF). Int J Methods Psychiatr Res 1998;7: 171–85. [27] Prentice R, Gloeckler L. Regression analysis of grouped survival data with application to breast cancer data. Biometrics 1978;34:57–67. [28] Stata [computer program], version 9.0]. College Station (Tex): Stata Corporation; 2005. [29] Goodwin RD, Fergusson DM, Horwood LJ. Asthma and depressive and anxiety disorders among young persons in the community. Psychol Med 2004;34:1465–74. [30] Hill AB. The environment and disease: association or causation. Proc Royal Soc Med 1965;58:295–300. [31] Spijker J, de Graaf R, Bijl RV, Beekman ATF, Ormel J, Nolen WA. Duration of major depressive episodes in the general population: results from The Netherlands Mental Health Survey and Incidence Study (NEMESIS). Br J Psychiatry 2002;181:208–13. [32] Spijker J, de Graaf R, Bijl RV, Beekman ATF, Ormel J, Nolen WA. Determinants of persistence of major depressive episodes in the general population. Results from the Netherlands Mental Health Survey and Incidence Study (NEMESIS). J Affect Disord 2004;81: 231–40. [33] Kriegsman DMW, Penninx BWJH, Van EijK JTM, Boeke AJP, Deeg DJH. Self-reports and general practitioner information on the presence of chronic diseases in community dwelling elderly. J Clin Epidemiol 1996;49:1407–17. [34] Kleinbaum DG, Kupper LL, Morgenstern H. Information bias. Epidemiologic research. New York: Van Nostrand Reinhold; 1982. p. 220–41. [35] Glassman AH, Miller GE. Where there is depression, there is inflammation… sometimes! Biol Psychiatry 2007;62:280–1. [36] Beaudet MP, Swain L, Catlin G. The National Population Health Survey — its longitudinal nature. Health Reports 1999;10:69–82. [37] de Graaf R, Bijl RV, Smit F, Ravelli A, Vollebergh WA. Psychiatric and sociodemographic predictors of attrition in a longitudinal study. Am J Epidemiol 2000;152:1047. [38] Badawi MA, Eaton WW, Myllyluoma J, Weimer LG, Gallo J. Psychopathology and attrition in the Baltimore ECA 15-year follow-up 1981–1996. Soc Psychiatry Psychiatr Epidemiol 1999;34:91–8.
