lable at ScienceDirect

Journal of Psychiatric Research 45 (2011) 1012e1018

Contents lists avai

Journal of Psychiatric Research

journal homepage: www.elsevier .com/locate/psychires

The association between antidepressant use and depression eight yearslater: A national cohort study

Ian Colman a,*, Yiye Zeng a, Anushka Ataullahjan a, Ambikaipakan Senthilselvan a, Scott B. Patten b,c

a School of Public Health, University of Alberta, Edmonton, CanadabDepartment of Psychiatry, University of Calgary, Calgary, CanadacDepartment of Community Health Sciences, University of Calgary, Calgary, Canada

a r t i c l e i n f o

Article history:Received 18 August 2010Received in revised form28 January 2011Accepted 3 February 2011

Keywords:Depressive disorderAntidepressive agentsTreatment outcomeLongitudinal studies

* Corresponding author. School of Public Health,University Terrace, 8303 112 Street, Edmonton, Alber780 492 8578; fax: þ1 780 492 0364.

E-mail address: ian.colman@ualberta.ca (I. Colman

0022-3956/$ e see front matter � 2011 Elsevier Ltd.doi:10.1016/j.jpsychires.2011.02.001

a b s t r a c t

Investigations of the effects of antidepressant treatment for individuals with major depression havefocused on short-term outcomes in individuals that meet very specific criteria; however, there is limitedknowledge about long-term outcomes associated with antidepressant use in general population samples.This study aimed to investigate the long-term outcomes associated with antidepressant use by focusingon 486 depressed adults in a prospective observational Canadian cohort in 1998/99. We used logisticregression to investigate the association between antidepressant use and depression status 8 years later.Non-random allocation to treatment was accounted for by a propensity-for-treatment model whichincluded thirteen predictors of antidepressant use, including: severity of depressive symptoms, previousepisodes of depression (from 1994 to 1997), physical health condition, social support and socio-demo-graphic characteristics. 29% of individuals with major depression reported antidepressant use. Afteradjusting for propensity for treatment in 1998/99, and antidepressant use from 2000 to 2007, depressedindividuals who reported antidepressant use in 1998/99 were less likely to be depressed in 2006/07compared to those who did not report antidepressant use (OR ¼ 0.36, 95% CI: 0.15e0.88). Amongstindividuals with symptoms of major depression, those reporting use of anti-depressants at baselineexhibited improved long-term outcomes in comparison to those who did not report treatment.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

The use of antidepressant medications for the treatment ofdepression has been rising steadily in recent years (Helgason et al.,2004; Middleton et al., 2001; Patten and Beck, 2004), though themajority of individuals with depression remain untreated (Colmanet al., 2008a, 2006; Patten and Beck, 2004). In the last thirty years,an overwhelming number of randomized controlled trials havedemonstrated the efficacy of antidepressant medications for thetreatment of depression (Fournier et al., 2010). However, these trialshave tended tohave relatively short follow-upperiods, oftenonly sixweeks (Barbui and Hotopf, 2001; Geddes et al., 2003). In addition,these trials often have stringent criteria that exclude patients withcommon comorbid symptoms or suicidal ideation, consequentlypainting a picture of antidepressant efficacy in unrealistically

University of Alberta, 3-50Dta, Canada T6G 2T4. Tel.: þ1

).

All rights reserved.

homogeneous patient populations (Barbui and Hotopf, 2001; Leonet al., 2003). There is a paucity of evidence regarding the long-term effectiveness of antidepressant use (Rouillon, 2004).

Population-based samples, followed longitudinally, may behelpful in clarifying whether the short-term efficacy demonstratedin clinical trials extends to long-term effectiveness, especially giventhe low likelihood of long-term randomized controlled trials.Observational studies, however, suffer from one crucial drawbackin comparison with randomized controlled trials: individuals whoreceive treatment may differ in clinical or social factors from thosewho do not, and these differences may affect prognosis, therebyconfounding any observed differences in outcome (D’Agostino,1998). Naturalistic study designs must therefore be analyzedusing methods that can address these selection-to-treatmentelements, in order to estimate the actual impact of treatment thatwould have been generated by a randomized design.

The propensity score approach is one method that can be usedfor this purpose (Rosenbaum and Rubin, 1983). This methodreduces all of the covariates that might predict treatment intoa single score, which quantifies the probability of being treated. Iftwo individuals have the same probability of being treated (i.e., the

I. Colman et al. / Journal of Psychiatric Research 45 (2011) 1012e1018 1013

same propensity score), but one has been treated and the other hasnot, then it is as if they have been randomly assigned to treatment(assuming that no unmeasured confounders exist) (D’Agostino,1998; Rosenbaum and Rubin, 1983).

Using these methods, one recent study found that individuals inmid-adulthood with severe symptoms of depression or anxiety whoreported antidepressantoranxiolyticusewere significantly less likelyto suffer from the same symptoms ten years later (Colman et al.,2008a); however, this finding has not been replicated. The primaryobjective of this study was to investigate long-term outcomes asso-ciated with self-reported antidepressant use on individuals withmajor depression in a longitudinal cohort of adult Canadians.

2. Materials and methods

2.1. Sample

The National Population Health Survey (NPHS) is a nationwidelongitudinal study conducted by Statistics Canada, which started in1994/95 and included health and other health-related information,such as economic, social, demographic, occupational and environ-mental data. At study inception, 17,276 individuals were randomlyselected using a stratified two-stage sample design. The cohort isrepresentative of the Canadian population, and has been followed-up every two years. The first cycle of interview in 1994/95 hada response rate of 83.6%, while the follow-up rates of the third cyclein 1998/99 and the most recent cycle in 2006/07 were 88.3% and77.0%, respectively (Statistics Canada, 2007). The focus of this studywas 486 individuals aged 16 years or older who experienceda major depressive episode in 1998/99, and had complete dataregarding factors associated with use of anti-depressants.

2.2. Depression status

Major depression in the NPHS is captured by the CompositeInternational Diagnostic Interview-Short Form (CIDI-SF). The CIDI-SF, a 10-min interview, has been found to have 93% classificationaccuracy for a major depressive episode compared with the fullCIDI (Kessler et al., 1998), an hour-long interview that can identifydepressive episodes consistent with the Diagnostic and StatisticalManual of Mental Disorders (DSM-IV) (American PsychiatricAssociation, 1994). For each individual, the CIDI-SF producesa predictive probability of a major depressive episode in the past 12months (Kessler et al., 1998). An individual with a predictiveprobability of 90% or higher was considered as having majordepression (a score of 5 or higher on a 0-8 scale) (Kessler et al.,1998). This corresponds to DSM-IV criteria for a major depressiveepisode: 5 of 9 depressive symptoms in a 2-week period during thepast year, including either loss of interest or depressed mood(American Psychiatric Association, 1994).

2.3. Treatment of depression

The primary exposure of interest was self-reported use of anti-depressants in 1998/99. In each cycle of the survey, surveymembers were asked for information on prescription and over-the-counter medications. With regards to antidepressant use, surveymembers were asked “In the past month, did you take anti-depressants such as Prozac, Paxil or Effexor?” (Statistics Canada,1999) Individuals who responded positively to this question in1998/99 were considered to be using anti-depressants. Self-reportof antidepressant use for depressed individuals has been estab-lished as having high accuracy (Kwon et al., 2003).

A secondary exposure of interest was counseling by a healthprofessional (primarily physicians and psychologists). In 1998/99,

survey members were asked whether they had seen a healthprofessional about their emotional or mental health in the previous12 months. Individuals who responded positively to this question in1998/99 were considered to be counseled by a health professional.

2.4. Confounding variables

Numerous factors are associated with both antidepressanttreatment and prognosis of depression. Notably, individuals withmore severe symptoms of depression are more likely to seektreatment and to be treated with psychotropic medications(Andrews et al., 2001; Burns et al., 2003; Colman et al., 2008b,2006; Mojtabai et al., 2002; Spijker et al., 2001), and these indi-viduals are also more likely to have a poorer prognosis (Kennedyet al., 2003; Spijker et al., 2004). Similarly, individuals witha history of depression or long-standing symptoms are more likelyto be treated (Spijker et al., 2001; Starkes et al., 2005; Colman et al.,2008b, 2006), and are more likely to also have a poor prognosis(Kennedy et al., 2003; Spijker et al., 2004; Mueller et al., 1999).These factors confound the treatmenteoutcome relationship inamanner such that it often appears that thosewho are treated havea worse prognosis (i.e., confounding by indication). Consequently,information was collected on several potential confounders.

Severity of symptoms was captured by including the CIDI-SFscore as well as a validated distress scale that includes feelings ofsadness, nervousness, restlessness, hopelessness, worthlessnessand feelings that “everything was an effort” over the precedingmonth (Kessler et al., 2002).

History of depression was derived using data from the 1994/95and 1996/97 cycles of the NPHS. Individuals who had majordepression at either of these time points according to the CIDI-SFwere considered to have a history of depression. The number ofweeks of depression an individual had experienced in the previousyear was also included in the propensity score.

In addition, suicidal behavior is associated with treatment-seeking (Mojtabai et al., 2002; Burns et al., 2003; Colman et al.,2008b; Starkes et al., 2005). NPHS respondents are asked aboutany suicide attempts in the previous 12 months. Those whoresponded positively in 1994/95, 1996/97, or 1998/99 wereconsidered to have a history of suicidal behavior.

Individuals in poor physical health are also more likely to seekhelp during episodes of depression (Andrews et al., 2001; Colmanet al., 2008b; Mojtabai et al., 2002). Physical health was rated ona 5-point scale in 1998/99; the presence of a chronic medicalcondition was also reported (Statistics Canada, 1999).

Finally, several demographic features are associated with treat-ment for depression, and may be related to prognosis; these wereincluded in the propensity score calculation. For example, those whoare female (Andrews et al., 2001; Mojtabai et al., 2002), divorced,separated, or widowed (Andrews et al., 2001; Colman et al., 2008b;Kisely et al., 2000; Mojtabai et al., 2002), unemployed (Colman et al.,2008b; Kisely et al., 2000),more educated (Starkes et al., 2005), older(Andrews et al., 2001; Kisely et al., 2000; Mojtabai et al., 2002), andhave less social support (Colman et al., 2008b; Mojtabai et al., 2002)are more likely to be treated for depression.

2.5. Statistical analysis

We used propensity score techniques to account for non-random allocation for treatment and subsequent confounding byindication. The propensity score is the predicted probability ofantidepressant use, conditional on all available observed covariates(Rubin, 1997). All potential confounders are included in a logisticregression model to estimate the probability of antidepressant useregardless of statistical significance or multicollinearity. Since the

I. Colman et al. / Journal of Psychiatric Research 45 (2011) 1012e10181014

goal of the model is to calculate a probability of antidepressant use(i.e., a propensity for treatment), redundant covariates do not harmthe model and over-parameterization is not a concern (Conniffeet al., 2000). In addition, including all available potentialconfounders ensures that the group that reports antidepressant useand the group that does not are comparable.

The propensity score was computed based on all observationswith complete data in 1998/99 so that it wasmore representative ofthe general population. The SURVEYLOGISTIC procedure was usedto predict the probability of reporting use of anti-depressants (i.e.,propensity score) with incorporation of bootstrap weights (seebelow). The individuals were grouped into 5 groups based onpropensity score quintiles. If no differences exist between cova-riates within quintiles of the propensity score between the treatedand untreated group, then any overall differences in the covariatescan be considered “ignorable” in an analysis controlling forpropensity score quintiles (Rubin, 1997).

To distinguish possible long-term effects of antidepressant useduring the initial episode outside of its association with continuingtreatment, self-reported antidepressant use from 2000 to 2007wasalso controlled in the model. An individual was considered to haveused anti-depressants recently if he/she had ever reported anti-depressant use in 2000/01, 2002/03, 2004/05 or 2006/07.

The primary analysis used logistic regression to calculate theassociation between reported antidepressant use in 1998/99 anddepression in 2006/07. Covariates in the model included thepropensity score and recent antidepressant use. The standarderrors for all estimates were calculated using the bootstrap method(Rao and Wu, 1988). All estimates were weighted to adjust forunequal selection probabilities and cluster sampling. Coefficients ofvariance (CVs) were calculated using the bootstrap method toestimate the sample error.

The secondary analysis looked at antidepressant use in 1998/99and depression in 2006/07, but used multinomial logistic regres-sion to investigate outcomes associated with being in one of threegroups: (1) use of anti-depressants, (2) counseled by a healthprofessional but no antidepressant use, or (3) no treatment. For thisanalysis we calculated a second propensity score for antidepressantuse or being counseled by a health professional using the sameconfounding variables listed above.

SAS 9.2 was used for all analyses.

Table 1Factors associated with antidepressant use among 486 individuals with depression in 19

Variables measured in 1998/99

CIDI-SF depression score (mean)Number of weeks depressed (mean)Depression history (%) No previously reported episodes

One previously reported episodeTwo previously reported episodes

Suicide attempt history (%)a YesDistress (mean)Gender (%) MaleAge (mean)Self-rated physical health (mean)Social support (mean)Chronic health condition (%) YesMarital Status (%) Divorced, Separated, WidowedEducation (%) Some post-secondary educationEmployment Status (%) Employed

UnemployedNot in Labor Force

*Statistically significant at p < 0.05.**Statistically significant at p < 0.01.

a Results can not be shown due to confidentiality.

3. Results

The prevalence of major depression in 1998/99 among thoseaged 16 or older was 4.5%. There were 486 individuals withcomplete data to calculate the probability of antidepressant use.Amongst these depressed individuals, 29.1% were using anti-depressants. Another 26.5% reported being counseled by a healthprofessional.

Table 1 illustrates several differences between those who wereusing anti-depressants and those who were not. Those who wereusing anti-depressants had higher levels of distress, had beendepressed for longer, and were more likely to have a history ofdepression. In addition, they were more likely to be female, inpoorer physical health, not employed, and to be divorced, separatedor widowed. A history of suicide attempts was included in thepropensity calculation; however, suicide attempts were rare.Statistics Canada prohibits reporting small numbers in order toprotect confidentiality, so these numbers are therefore not reportedexplicitly.

All 13 variables listed in Table 1 were included in a logisticregressionmodel that calculated propensity for antidepressant use.Themodel explained 27% of the variation in antidepressant use. Theindividuals were grouped into quintiles of the propensity score; allcovariates were balanced within each quintile between those usinganti-depressants and those not.

Of the 486 individuals with a major depressive episode, thedepression status of 321 individuals was followed-up eight yearslater in 2006/07 (66%). Follow-up among those who reportedantidepressant use at baseline was 62%; 68% of those who didwho did not report antidepressant use at baseline weresuccessfully followed. 61.7% of the non-response was attribut-able to loss to follow up, while 15.9% of the non-responseresulted from deceased individuals and institutionalized indi-viduals, and another 22.4% of the respondents did not providecomplete depressive symptom information. Table 2 comparesthose with follow-up information to those without. Individualswho did not have follow-up data did not differ from those withcomplete data with respect to CIDI-SF score or antidepressantuse in 1998/99. However, those who did not complete follow-uphad a longer duration of symptoms and a higher severity on thedistress scale in 1998/99.

98/99.

Antidepressant use in 1998/99 p-value

Yes (29.1%) No (70.9%)

6.6 6.1 <0.01 **

18.7 10.8 <0.01 **

43.5 70.6 <0.01 **

30.3 21.426.3 8.0

0.9610.9 8.2 <0.01 **

22.1 37.4 0.01 *

41.9 38.7 0.131.8 2.3 <0.01 **

3.8 3.9 0.3185.0 77.0 0.0835.3 20.8 <0.01 **

72.2 67.1 0.437.9 62.2 <0.01 **

4.0 4.458.1 33.3

Table 2Comparison of those with outcome data at eight-year follow-up to those without outcome data among 486 individuals with depression in 1998/99.

Variables measured in 1998/99 Depression status measured in 2006/07 p-value

Complete data (66%) Incomplete data (34%)

CIDI-SF depression score (mean) 6.2 6.4 0.34Antidepressant use (%) 27.2 32.5 0.34Number of weeks depressed (mean) 11.4 16.3 0.01 *

Depression history (%) No previously reported episodes 63.2 61.8 0.73One previously reported episode 24.6 22.8Two previously reported episodes 12.2 15.4

Suicide attempt history (%) y Yes 0.18Distress (mean) 8.3 10.3 <0.01 **

Gender (%) Male 32.3 34.2 0.75Age (mean) 38.6 41.5 0.19Self-rated physical health (mean) 2.3 1.9 <0.01 **

Social support (mean) 4.0 3.8 0.04 *

Chronic health condition (%) Yes 76.6 84.4 0.10Marital Status (%) Divorced, Separated, Widowed 23.2 28.3 0.36Education (%) Some post-secondary education 73.3 59.8 0.02 *

Employment Status (%) Employed 60.4 45.3 <0.01 **

Unemployed 5.0 3.1Not in Labor Force 34.6 51.6

*Statistically significant at p < 0.05.**Statistically significant at p < 0.01.yResults can not be shown due to small numbers reporting.

Table 3Association between antidepressant use in 1998/99 and depression in 2006/07among 321 individuals with depression in 1998/99.

Effect of antidepressant useon depressioneight years later OR (95% CI)

Unadjusted model 0.94 (0.44, 2.01)Model adjusted for propensity quintiles 0.70 (0.30, 1.61)Model adjusted for propensity

quintiles and antidepressant usefrom 2000 to 2007

0.36 (0.15, 0.88)

Results in bold are significant at p < 0.05.

I. Colman et al. / Journal of Psychiatric Research 45 (2011) 1012e1018 1015

Adequate data were available on 321 individuals in order toassess the presence of depression in 2006/07. In unadjustedmodels, depressed individuals who reported antidepressant usein 1998/99 were equally likely to have a major depressiveepisode in 2006/07 as those who did not report antidepressantuse (OR ¼ 0.94, 95% CI: 0.44e2.01; see Table 3). Among thoseusing anti-depressants at baseline, 77.9% used anti-depressantsduring the follow-up period between 2000 and 2007, while33.0% of those who did not use anti-depressants at baselinereported antidepressant use during the follow-up period. Afteradjusting for propensity for antidepressant use in 1998/99, andantidepressant use from 2000 to 2007 (see Table 2), depressedindividuals who received antidepressant use in 1998/99 wereless likely to be depressed in 2006/07 compared to those whodid not use anti-depressants (OR ¼ 0.36, 95% CI: 0.15e0.88).Notably, recent antidepressant use was strongly and positivelyassociated with major depression in the model. The individualswho reported antidepressant use during the previous six yearswere much more likely to have major depression in 2006/07(OR ¼ 5.83, 95% CI: 2.36e14.36).

In the comparison of antidepressant use, counseling by a healthprofession and no treatment (see Table 4), those who used anti-depressants at baseline were significantly less likely to bedepressed eight years later (OR ¼ 0.32, 95% CI: 0.12e0.85)compared to those who were not treated after adjusting forpropensity for treatment and use of anti-depressants during thefollow-up period. However, no differences in depression at follow-up were noted between those who reported seeing a healthprofessional at baseline and thosewhowere not treated (OR¼ 0.93,95% CI: 0.39e2.22).

4. Discussion

4.1. Highlights and related findings

Our results from a national Canadian cohort find that depressedindividuals who reported antidepressant use in 1998/1999 wereless likely to display depressive symptoms eight years later.Numerous randomized controlled trials have shown that antide-pressant treatment is efficacious for individuals suffering fromamajor depressive episode, particularly those suffering from severesymptoms (Fournier et al., 2010). However, these trials often have

strict inclusion criteria that exclude patients with suicidal ideationand comorbid illnesses, leaving trial subjects generally unrepre-sentative of the patient population commonly seen in clinicalpractice (Barbui and Hotopf, 2001; Leon et al., 2003). Observationalstudies present an opportunity to demonstrate potential effec-tiveness of anti-depressants in general populations, particularly ifthey account for confounding by indication by using propensity-for-treatment models. One notable study used a propensity scoreanalysis on observational data to study 285 depressed patients overa 20-year period, and found individuals using higher levels of anti-depressants were significantly more likely to recover from symp-toms in the short term (Leon et al., 2003). Given that there is littledata on long-term outcomes of antidepressant use (Rouillon, 2004),another used propensity-for-treatment techniques to study 204depressed members of a national British birth cohort, and foundthat individuals using anti-depressants or anxiolytics were signif-icantly less likely to be suffering from symptoms of depression tenyears later (Colman et al., 2008a), similar to the results observedhere in this Canadian sample.

Although maintenance use of anti-depressants after an initialdepressive episode has been shown to reduce the likelihood ofa relapse episode (Geddes et al., 2003), the majority of individualswith depression do not continue antidepressant treatment in thelong term, even if they are suffering from chronic symptoms(Bockting et al., 2008; Colman et al., 2008a). Consequently, ourpositive results are unlikely to be explained by long-term use ofanti-depressants. In order to confirm this, we controlled for anti-depressant use during the follow-up period in our study to ensurethat antidepressant use at baseline was not just an indicator ofsubsequent or ongoing use. There is currently little evidence to

Table 4Association between depression in 2006/07 and antidepressant use or counseling by a health professional in 1998/99 (compared to no treatment)among 321 individuals with depression in 1998/99.

Effect of antidepressant use on depressioneight years later OR (95% CI)

Effect of counseling bya health professional on depressioneight years later OR (95% CI)

Unadjusted model 1.13 (0.48, 2.68) 1.61 (0.72, 3.60)Model adjusted for propensity quintiles 0.58 (0.24, 1.38) 0.97 (0.44, 2.16)Model adjusted for propensity quintiles and

antidepressant use from 2000 to 20070.32 (0.12, 0.85) 0.93 (0.39, 2.22)

Results in bold are significant at p < 0.05.

I. Colman et al. / Journal of Psychiatric Research 45 (2011) 1012e10181016

suggest that the neural mechanisms, such as neurogenesis in thehippocampus, that are influenced by antidepressant use continueafter withdrawal of the medication (DeRubeis et al., 2008).Consequently, the effects observed in our study appear point to theimportance of receiving initial treatment that maximizes the like-lihood of complete remission of symptoms. Periods of partialremission are relatively common after treatment (Cornwall andScott, 1997), and have been associated with increased hopeless-ness, suicide, life dissatisfaction, and decreased psychosocial func-tioning (Viinamaki et al., 2008). During periods of partial remission,these symptoms often go untreated (Cornwall and Scott, 1997) andare associated with significantly higher rates of relapse episodes ofdepression (Pintor et al., 2003).

The association observed in this study between baseline anti-depressant use and reduced risk of depression eight years latermaybe a reflection of treatment-seeking behavior rather than an effectof treatment with anti-depressants. Perceived need for treatmentand treatment-seeking behavior were not measured in our studybut are strongly associatedwith treatment for depression (Mojtabaiet al., 2002). Those who have a positive attitude toward mentalhealth treatment are more likely to pursue treatment whensuffering from symptoms of mental illness (Mojtabai et al., 2002).We were not able to measure perceived need for treatment, atti-tudes toward mental health treatment or treatment-seekingbehavior in our study; however, it may be that those who reportbeing treated for depression at baseline are more likely to perceivea need for treatment and seek treatment during subsequentepisodes (Colman et al., 2008a).

Notably, our study found that reported antidepressant useduring the follow-up period was associated with a poor outcome.This is likely due to the selection factors that led us to usepropensity score techniques in our analysis of baseline antide-pressant use. In other words, it is likely that those who reportantidepressant use during the follow-up period are suffering frommore severe symptoms and are likely to have a poor outcome at theend of the follow-up period. However, after adjusting for this, wefind that those who reported antidepressant use during their initialepisode had a better long-term outcome. This suggests that thelonger depressive symptoms persist before treatment the worsethe long-term prognosis may be. Supporting this concept isevidence from other studies that have shown individuals who havehad a longer duration of depressive symptoms before treatment aremore likely to have relapse episodes during several years of follow-up (Kennedy et al., 2003; Mueller et al., 1999).

4.2. Strengths and limitations

It is not possible to make inferences about causality based onour study. Randomized controlled trials provide much strongerevidence for causality, particularly with large samples, as therandomization process ensures that the groups are comparable onall factors. Although our propensity-for-treatment model balancedseveral key determinants of antidepressant use, there are many

other factors that were not measured in this study that could alsobe associated with both treatment and prognosis. An importantadditional unmeasured factor that could be associated with bothtreatment and prognosis is common comorbid psychiatric illnesssuch as generalized anxiety disorder. The NPHS does not assessanxiety disorders. The inclusion of the distress scale, whichincludes nervousness and restlessness, should have captured someof the comorbidity introduced by anxiety disorders, but not all.However, those with comorbid disorders are more likely to seektreatment (Andrews et al., 2001; Mojtabai et al., 2002; Spijker et al.,2001) and more likely to have a poor prognosis (Ormel et al., 1993).Not accounting for this in our design biases our results toward thenull, suggesting that the presented results are conservative.

This study did not have sufficient data to assess long-termoutcomes associated with evidence-based psychotherapy inter-ventions for depression. We attempted a crude measurement ofpossible exposure to psychotherapy through contact to a healthprofessional, but datawere not available to assess what, if any, non-pharmaceutical treatment might have been offered by the healthprofessional. Nevertheless, there is strong evidence supportinglong-term prevention of depression for non-pharmaceutical treat-ments. In particular, cognitive behavioral therapy has been shownto prevent relapse of depressive symptoms over several years afterthe initial episode (Bockting et al., 2009; Fava et al., 2004), bychanging the way individuals who have had a depressive episodeprocess depression-related material from an absolutist view toa relativist view (Teasdale et al., 2001). There is some evidence tosuggest that cognitive behavioral therapy is as or more effectivethan medication in preventing recurrence of symptoms over thelong term (Hollon et al., 2005). Other non-pharmaceutical treat-ments may have similar long-term prevention effects. For example,a recent review showed that community-based social support(“befriending”) has a significant long-term effect on reduction ofdepressive symptoms (Mead et al., 2010). Nevertheless, suchpatients would have been included in our untreated group andtherefore bias our results toward a more conservative finding.

Another limitation is that we do not consider long-term sideeffects that may be associated with antidepressant use. Forexample, the use of selective serotonin reuptake inhibitors (SSRI)has been associated with many long-term side effects includingosteoporosis, cardiovascular side effects, and alopecia (Moret et al.,2009). In addition, SSRI use has been associated with an increasedrisk of suicidal behavior in adolescents (Barbui et al., 2009).

This study used the CIDI short form instead of the full CIDI.Although the CIDI-SF has a 93% classification accuracy for a majordepressive episode compared with the full CIDI (Kessler et al., 1998),individuals could have beenmisclassifiedwith respect to the primaryoutcome. Such non-differential misclassification biases resultstoward the null, suggesting the results we present are conservative.

Finally, this study, as with any longitudinal study, did not havecomplete data on all individuals at the eight-year follow-up point.Those without follow-up data were no different from those withfollow-up data with regards to their CIDI-SF score; however, they

I. Colman et al. / Journal of Psychiatric Research 45 (2011) 1012e1018 1017

did have a longer duration of symptoms at baseline and higherscores on the distress scale, suggesting they may have had a higherdegree of severity and consequently a poorer prognosis. There wasno difference between these two groups with respect to antide-pressant use at baseline, so any effects of higher severity patientsbeing lost does not appear to be differential with respect to anti-depressant use and consequently is unlikely to bias the associationbetween antidepressant use and subsequent depression.

The limitations to this study are outweighed by several strengths.The sample for this study was a large population-based sample.Population-based samples are particularly helpful as the resultsobtained from such samples are relevant to the general population,not just the small group that meet inclusion criteria for antidepres-sant RCTs (Barbui andHotopf, 2001; Leon et al., 2003). This study is ofparticular value in that data on a variety of factors that are associatedwith depression treatment and prognosis were available includinginformation on antidepressant use throughout the follow-up period.

We confirm the findings of an earlier British study, which used anage-homogenous sample, combined anxiolytics with anti-depres-sants, and used a general screen for symptoms of depression andanxiety to show long-term positive outcomes associated withpsychotropic medication use (Colman et al., 2008a). However, thisCanadian sample provided adults of all ages, studied currentlyprescribed anti-depressants, and used a structured assessment fordepression which has been widely used for epidemiological studiesand conforms to widely used diagnostic criteria (Kessler et al., 1998).These strengths underscore the provocative findings, which suggestthat treatment for major depression may have significant long-termbenefits.

Contributors

1. Conception and design, or acquisition of data, or analysis andinterpretation of data (Ian Colman, Yiye Zeng, Anushka Ataul-lahjan, Ambikaipakan Senthilselvan, Scott Patten).

2. Drafting the article or revising it critically for important intel-lectual content (Ian Colman, Yiye Zeng, Anushka Ataullahjan,Ambikaipakan Senthilselvan, Scott Patten).

3. Final approval of the version to be published (Ian Colman, YiyeZeng, Anushka Ataullahjan, Ambikaipakan Senthilselvan, ScottPatten).

Role of the funding source

The project was supported by the Canadian Institutes of HealthResearch (CIHR), grant numberMOP93750. Dr. Colman is supportedby a Population Health Investigator Award from the Alberta Heri-tage Foundation for Medical Research (AHFMR) and a New Inves-tigator Award from CIHR. Dr Patten is supported by a Senior HealthScholar Award from AHFMR. The sponsors had no role in the designand conduct of the study; collection, management, analysis, andinterpretation of the data; and preparation, review, or approval ofthe manuscript.

Ethical approvalThis study was approved by the Health Research Ethics Board of

the University of Alberta.

Conflict of interestAll authors declare that they have no conflict of interest.

Acknowledgments

The authors would like to acknowledge Ms. Irene Wong ofStatistics Canada for her assistance with data access and use. The

researchandanalysis arebasedondata fromStatisticsCanadaand theopinions expressed do not represent the views of Statistics Canada.

References

American Psychiatric Association. Diagnostic and statistical manual of mentaldisorders. Washington: American Psychiatric Association; 1994.

Andrews G, Issakidis C, Carter G. Shortfall in mental health service utilisation. TheBritish Journal of Psychiatry 2001;179:417e25.

Barbui C, Hotopf M. Forty years of antidepressant drug trials. Acta PsychiatricaScandinavica 2001;104:92e5.

Barbui C, Esposito E, Cipriani A. Selective serotonin reuptake inhibitors and risk ofsuicide: a systematic review of observational studies. Canadian Medical Asso-ciation Journal 2009;180:291e7.

Bockting CL, ten Doesschate MC, Spijker J, Spinhoven P, Koeter MW, Schene AH.Continuation and maintenance use of antidepressants in recurrent depression.Psychotherapy and Psychosomatics 2008;77:17e26.

Bockting CL, Spinhoven P, Wouters LF, Koeter MW, Schene AH. Long-term effects ofpreventive cognitive therapy in recurrent depression: a 5.5-year follow-upstudy. Journal of Clinical Psychiatry 2009;70:1621e8.

Burns T, Eichenberger A, Eich D, Ajdacic-Gross V, Angst J, Rossler W. Which indi-viduals with affective symptoms seek help? Results from the Zurich epidemi-ological study. Acta Psychiatrica Scandinavica 2003;108:419e26.

Colman I, Wadsworth ME, Croudace TJ, Jones PB. Three decades of antidepressant,anxiolytic and hypnotic use in a national population birth cohort. The BritishJournal of Psychiatry 2006;189:156e60.

Colman I, Croudace TJ, Wadsworth ME, Kuh D, Jones PB. Psychiatric outcomes 10years after treatment with antidepressants or anxiolytics. The British Journal ofPsychiatry 2008a;193:327e31.

Colman I, Croudace TJ, Wadsworth MEJ, Jones PB. Factors associated with antide-pressant, anxiolytic and hypnotic use over 17 years in a national cohort. Journalof Affective Disorders 2008b;110:234e40.

Conniffe D, Gash V, O’Connell PJ. Evaluating state programmes: "natural experiments"and propensity scores. The Economic and Social Review 2000;31:283e308.

Cornwall PL, Scott J. Partial remission in depressive disorders. Acta PsychiatricaScandinavica 1997;95:265e71.

D’Agostino Jr RB. Propensity score methods for bias reduction in the comparison ofa treatment to a non-randomized control group. Statistics in Medicine1998;17:2265e81.

DeRubeis RJ, Siegle GJ, Hollon SD. Cognitive therapy versus medication fordepression: treatment outcomes and neural mechanisms. Nature Reviews.Neuroscience 2008;9:788e96.

Fava GA, Ruini C, Rafanelli C, Finos L, Conti S, Grandi S. Six-year outcome ofcognitive behavior therapy for prevention of recurrent depression. The Amer-ican Journal of Psychiatry 2004;161:1872e6.

Fournier JC, DeRubeis RJ, Hollon SD, Dimidjian S, Amsterdam JD, Shelton RC, et al.Antidepressant drug effects and depression severity: a patient-level meta-analysis. The Journal of American Medical Association 2010;303:47e53.

Geddes JR, Carney SM, Davies C, Furukawa TA, Kupfer DJ, Frank E, et al. Relapseprevention with antidepressant drug treatment in depressive disorders:a systematic review. Lancet 2003;361:653e61.

Helgason T, Tomasson H, Zoega T. Antidepressants and public health in Iceland. Timeseries analysis of national data. The British Journal of Psychiatry 2004;184:157e62.

Hollon SD, DeRubeis RJ, Shelton RC, Amsterdam JD, Salomon RM, O’Reardon JP, et al.Prevention of relapse following cognitive therapy vs medications in moderateto severe depression. Archives of General Psychiatry 2005;62:417e22.

Kennedy N, Abbott R, Paykel ES. Remission and recurrence of depression in themaintenance era: long-term outcome in a Cambridge cohort. PsychologieMedicale 2003;33:827e38.

Kessler RC, Andrews G, Mroczek D, Ustun B, Wittchen H- U. The world healthorganization composite international diagnostic interview short-form(CIDI-SF). International Journal of Methods in Psychiatric Research 1998;7:171e85.

Kessler RC, Andrews G, Colpe LJ, Hiripi E, Mroczek DK, Normand SL, et al. Shortscreening scales to monitor population prevalences and trends in non-specificpsychological distress. Psychologie Medicale 2002;32:959e76.

Kisely S, Linden M, Bellantuono C, Simon G, Jones J. Why are patients prescribedpsychotropic drugs by general practitioners? Results of an international study.Psychologie Medicale 2000;30:1217e25.

Kwon A, Bungay KM, Pei Y, Rogers WH, Wilson IB, Zhou Q, et al. Antidepressant use:concordance between self-report and claims records. Medical Care 2003;41:368e74.

Leon AC, Solomon DA, Mueller TI, Endicott J, Rice JP, Maser JD, et al. A 20-yearlongitudinal observational study of somatic antidepressant treatment effec-tiveness. The American Journal of Psychiatry 2003;160:727e33.

Mead N, Lester H, Chew-Graham C, Gask L, Bower P. Effects of befriending ondepressive symptoms and distress: systematic review and meta-analysis. TheBritish Journal of Psychiatry 2010;196:96e101.

Middleton N, Gunnell D, Whitley E, Dorling D, Frankel S. Secular trends in antide-pressant prescribing in the UK, 1975e1998. Journal of Public Health andMedicine 2001;23:262e7.

Mojtabai R, Olfson M, Mechanic D. Perceived need and help-seeking in adults withmood, anxiety, or substance use disorders. Archives of General Psychiatry2002;59:77e84.

I. Colman et al. / Journal of Psychiatric Research 45 (2011) 1012e10181018

MoretC, IsaacM,BrileyM.Problemsassociatedwith long-termtreatmentwithselectiveserotonin reuptake inhibitors. Journal of Psychopharmacology 2009;23:967e74.

Mueller TI, Leon AC, Keller MB, Solomon DA, Endicott J, Coryell W, et al. Recurrenceafter recovery from major depressive disorder during 15 years of observationalfollow-up. The American Journal of Psychiatry 1999;156:1000e6.

Ormel J, Oldehinkel T, Brilman E, vanden Brink W, vanden Brink W. Outcome ofdepression and anxiety in primary care. A three-wave 3 1/2-year study ofpsychopathology and disability. Archives of General Psychiatry 1993;50:759e66.

Patten SB, Beck C. Major depression and mental health care utilization in Canada:1994 to 2000. Canadian Journal of Psychiatry 2004;49:303e9.

Pintor L, Gasto C, Navarro V, Torres X, Fananas L. Relapse of major depression aftercomplete and partial remission during a 2-year follow-up. Journal of AffectiveDisorders 2003;73:237e44.

Rao JNK, Wu CFJ. Resampling inference with complex survey data. Journal of theAmerican Statistical Association 1988;83:231e41.

Rosenbaum P, Rubin DB. The central role of the propensity score in observationalstudies for causal effects. Biometrika 1983;70:41e55.

Rouillon F. Long term therapy of generalized anxiety disorder. European Psychiatry2004;19:96e101.

Rubin DB. Estimating causal effects from large data sets using propensity scores.Annals of Internal Medicine 1997;127:757e63.

Spijker J, Bijl RV, de Graaf R, Nolen WA. Care utilization and outcome of DSM-III-Rmajor depression in the general population. Results from the Netherlandsmental health survey and incidence study (NEMESIS). Acta Psychiatrica Scan-dinavica 2001;104:19e24.

Spijker J, de Graaf R, Bijl RV, Beekman AT, Ormel J, Nolen WA. Determinants ofpersistence of major depressive episodes in the general population. Resultsfrom the Netherlands mental health survey and incidence study (NEMESIS).Journal of Affective Disorders 2004;81:231e40.

Starkes JM, Poulin CC, Kisely SR. Unmet need for the treatment of depression inAtlantic Canada. Canadian Journal of Psychiatry 2005;50:580e90.

Statistics Canada. National population health survey cycle 3 (1998e1999). Ottawa:Public Use Microdata Files Documentation; 1999.

Statistics Canada. National population health survey cycle 7 (2006e2007). Ottawa:Public Use Microdata Files Documentation; 2007.

Teasdale JD, Scott J, Moore RG, Hayhurst H, Pope M, Paykel ES. How does cognitivetherapy prevent relapse in residual depression? Evidence from a controlledtrial. Journal of Consulting and Clinical Psychology 2001;69:347e57.

Viinamaki H, Hintikka J, Tolmunen T, Honkalampi K, Haatainen K, Koivumaa-Honkanen H. Partial remission indicates poor functioning and a high level ofpsychiatric symptoms: a 3-phase 6-year follow-up study on major depression.Nordic Journal of Psychiatry 2008;62:437e43.