Time to recovery in episodes of majordepression from 4 community samples.
Weibull and lognormal distributionsfitted to time to relapse after an episode of major depression from 6 naturalisticfollow-up studies.16- 21
Proportion of time in depressionaverted (95% uncertainty interval) during 1 year after the onset of indexepisode (A) and during 5 years after the end of the index episode (B) in thosein contact with health services.
Vos T, Haby MM, Barendregt JJ, Kruijshaar M, Corry J, Andrews G. The Burden of Major Depression Avoidable by Longer-term Treatment Strategies. Arch Gen Psychiatry. 2004;61(11):1097–1103. doi:10.1001/archpsyc.61.11.1097
Major depression is the largest single cause of nonfatal disease burden
in Australia. Effective drug and psychological treatments exist, yet are underused.
To quantify the burden of disease currently averted in people seeking
care for major depression and the amount of disease burden that could be averted
in these people under optimal episodic and maintenance treatment strategies.
Modeling impact of current and optimal treatment strategies based on
secondary analysis of mental health survey data, studies of the natural history
of major depression, and meta-analyses of effectiveness data. Monte Carlo
simulation of uncertainty in the model.
The cohort of Australian adults experiencing an episode of major depression
in 2000 are modeled through “what if” scenarios of no treatment,
current treatment, and optimal treatment strategies with cognitive behavioral
therapy or antidepressant drug treatment.
Main Outcome Measure
Disability-Adjusted Life Year.
Current episodic treatment averts 9% (95% uncertainty interval, 6%-12%)
of the disease burden of major depression in Australian adults. Optimal episodic
treatment with cognitive behavioral therapy could avert 28% (95% uncertainty
interval, 19%-39%) of this disease burden, and with drugs 24% (95% uncertainty
interval, 19%-30%) could be averted. During the 5 years after an episode of
major depression, current episodic treatment patterns would avert 13% (95%
uncertainty interval, 10%-17%) of Disability-Adjusted Life Years, whereas
maintenance drug treatment could avert 50% (95% uncertainty interval, 40%-60%)
and maintenance cognitive behavioral therapy could avert 52% (95% uncertainty
interval, 42%-64%), even if adherence of around 60% is taken into account.
Longer-term maintenance drug or psychological treatment strategies are
required to make significant inroads into the large disease burden associated
with major depression in the Australian population.
As in other industrialized countries, depression is the most commonmental disorder in Australia.1 It is the largestsingle cause of disability and is responsible for 6.2% and 9.8% of years livedwith disability in men and women, respectively. It is the third largest causeof disease burden in Disability-Adjusted Life Years (DALY) in women and rankedeighth in men.2 Treatment guidelines recommendantidepressant (AD) drugs and/or a specific, effective psychological therapyfor major depressive disorder.3,4 Cognitivebehavioral therapy (CBT) and interpersonal therapy are the psychological treatmentsthat have the best-documented efficacy.4 However,only 59% of respondents identified as having major depression in the 1997National Survey of Mental Health and Wellbeing (NSMHWB) had sought any helpfor their problem, and 35% reported consulting a relevant health professionaland receiving medication or psychological treatment.1
The high burden of depression and the poor use of efficacious treatmentsmean that there is considerable potential for health gain. A previous analysisquantified the cost-effectiveness of an ideal mix of interventions for affectivedisorders in Australia and concluded that evidence-based health care is supportedon grounds of efficacy as well as cost-effectiveness.5 However,it did not report the scope of health gain as a proportion of the overalldisease burden of major depression that can be achieved by offering effectivetreatment to all people with depression who seek care from health services.Further limitations were that no separate conclusion could be made about theimpact of different treatment options and that by limiting analyses to a 1-yearperiod the impact of long-term treatment strategies could not be evaluated.This article quantifies the impact of treatment strategies on the diseaseburden due to major depression. In particular, it answers the following questions:(1) what is the proportion of the depression burden averted by current treatment?(2) what is the potential of episodic drug and psychological treatment optionsto further reduce this burden? and (3) what is the potential of longer-termmaintenance drug and psychological treatment options to further reduce thisburden?
The impact of evidence-based psychological and drug treatment strategiesis modeled as a change in DALY. Separate estimates are presented for short-termtreatments directed at episodes, including a short continuation phase andlonger-term maintenance treatments during 5 years of follow-up. Data are derivedfrom existing surveys and routine health information collection systems inAustralia as well as findings on the epidemiology of depression and its treatmentsin the international literature. The analysis starts with a description ofthe epidemiology of depression and current health service use patterns inAustralia. The next step is an evaluation of the impact of effective treatmentsby translating outcome measures from meta-analyses of trials into a changein DALY. The main comparisons are between the amount of depression experiencedunder current and expanded treatment options vs the hypothetical disease burdenin the absence of treatment. Our analysis applies to Australian adults whoexperienced an episode of major depression in the year 2000 and sought carefrom health services.
We derive parameters on the prevalence of major depression and treatmentpatterns from the 1997 NSMHWB1 and apply theseto 2000 population figures. The main outcome of the survey was the 1-yearprevalence, ie, people qualifying for a diagnosis of major depression in the12 months prior to the survey. An additional question on the recency of symptomsallows identification of respondents with current prevalence, ie, having hadsymptoms in the last 2 weeks, the minimum duration of an episode. Of the surveyrespondents identified as having major depression as defined by the International Classification of Diseases, 10th Revision (ICD-10),6 58.9% had consulted a psychologist,psychiatrist, and/or general practitioner for a mental health problem, whereas35.1% fulfilled our criteria for potentially having received evidence-basedtreatment: consulting a health professional at least 3 times and having hadmedication and/or CBT (“learning how to change thoughts, behaviors,and emotions”).5
We grade the severity of prevalent cases of depression from the NSMHWBby the number of standard deviations from the mean mental component scoreof the 12-item Short-Form Health Survey (SF-12)7 intonormal (≥45), mild (35-44.9), moderate (25-34.9), and severe (<25).Disability weights for mild (0.14), moderate (0.35), and severe (0.76) depression,which were used in the Australian Burden of Disease Study2 andderived from a Dutch study,8 are assumed toapply to these categories.
Next, we use data from international follow-up studies on the naturalhistory of major depression to mathematically describe the variation in durationof episodes and the time to the next episode. While there are many naturalisticstudies of the duration of major depressive disorder episodes in clinicalsamples, there are few follow-up studies of major depressive disorder in communitysamples.9- 13 The4 US studies show a similar pattern of recovery over time after the startof an episode. The median time to recovery in the 4 studies ranged from 8to 12 weeks, and at 1 year between 3% and 11% of cases had not yet recovered.The figures from Kendler et al10 have beenadjusted for the 7% of excluded cases with an onset of more than 1 year priorto study. The fifth study from the Netherlands11 reportsa considerably longer duration of episodes. Inclusion of subsyndromal depressionand dysthymia in life chart histories is a possible explanation for this higherestimate. From the data reported in the US studies,9,10,12,13 wefit a lognormal distribution14 that has thelowest sum of squared differences between modeled and observed time to recoverystarting from a minimum duration of 2 weeks specified in the definition ofan episode of major depression (Figure 1).
Major depression is a chronic episodic disorder, and hence for our modelingpurposes it is important to describe the pattern of time to a next episodeafter a previous episode. During a few decades of follow-up, major depressionis reported as a recurrent disorder in 80% of cases.15 Weassume that during a lifetime at least 90% of affected individuals experiencea recurrence. Six naturalistic follow-up studies16- 21 reporton the risk of relapse during periods varying between 6 months and 2 yearsafter cessation of drug treatment for an acute episode of major depression.We fit lognormal and Weibull distributions22 thatgive the best fit as determined by the lowest sum of squared differences betweenmodeled and observed data points (Figure 2).We decide to use the lognormal distribution because it gives a slightly betterfit. In a Monte Carlo simulation model, we use the lognormal distributionsdescribing the length of episodes and the time between episodes to estimatethe mean number of episodes and the mean time depressed during a 6-month anda 5-year period after an episode.
We separately evaluate drug treatment for episodes of major depressionplus a continuation phase after remission of symptoms and maintenance treatmentof 5 years after remission of an episode, CBT treatment of major depressiveepisodes, and a maintenance variant of CBT with booster sessions during aperiod of 5 years.
A meta-analysis reporting on 48 trials estimated an effect size (ES)of 0.55 (95% confidence interval, 0.40-0.70) for selective serotonin reuptakeinhibitors over placebo.23 No differences werefound between 4 different selective serotonin reuptake inhibitors. Meta-analysesexamining the efficacy of selective serotonin reuptake inhibitors and tricyclicantidepressants consistently show no significant differences between the 2drug classes.23- 26 Therefore,we assume the same efficacy for all antidepressants.
From the figures presented in a recent meta-analysis27 ofthe odds ratios of relapse in 26 maintenance drug studies and 7 continuationdrug studies, we derive a pooled relative risk of 0.416 (95% confidence interval,0.312-0.555) for relapse with continuation AD drug treatment and 0.437 (95%confidence interval, 0.394-0.485) for maintenance AD drug treatment.
A meta-analysis of cognitive therapy reports a pooled ES of 0.82 from48 studies.28 On closer inspection, severalstudies included in this systematic review do not fit the stated inclusioncriteria. Our own meta-analysis of CBT interventions, including many of thesame studies as well as a few additional studies, gives a random effects ESof 0.77 (95% confidence interval, 0.44-1.10), close to the Gloaguen pointestimate but with wider confidence intervals. We use these figures in ourmain analyses. Excluding 2 outlier studies (by the same author) with particularlyhigh ES estimates reduces the Q statistic for heterogeneity from 50.8 (df = 16, P< .001)to 22.3 (df = 13, P = .051).In a separate sensitivity analysis, we recalculate the model using the ES(0.54; 95% confidence interval, 0.29-0.79) calculated after excluding these2 outliers.
Although the effect of AD drugs ceases when treatment is stopped, thereis evidence for a prolonged effect of CBT beyond the treatment period. Froma review of naturalistic longer-term follow-up studies (ranging from 1.5 to4 years) after randomized controlled trials17,20,29- 31 thatwere set up to compare CBT with AD drugs in the acute phase, we calculatea lower risk of relapse after CBT (relative risk, 0.64; 95% CI, 0.51-0.79).
Maintenance CBT is described in 2 trials. The first compares CBT maintenancewith AD drug maintenance and during 1 year of follow-up found no differencein relapse.32 The other reports on a trialof maintenance CBT after acute CBT.21 At a2-year follow-up, the groups who had maintenance CBT had 25% relapse (5/20)compared with 80% (16/20) in the group receiving case management only afterCBT (AD drugs were tapered off and discontinued in both groups). The scantyevidence from these 2 trials suggests similar impact of maintenance strategieswith AD drugs or CBT.
Several meta-analyses with a large overlap in the included studies reportdiscontinuation rates of between 27% and 39%, with 3% to 6% lower rates forselective serotonin reuptake inhibitors in comparison with tricyclic antidepressants.23,24,26,33 However,because most trials are of short duration, representing what is possible withmotivated patients and physicians, adherence rates may be lower than reportedin the controlled trial literature. Adherence in 4 studies of primary careranges from 50% to 66%.34- 37 Wedecide to model drug adherence ranging uniformly between the recorded adherencelevel in trials and an estimated lower level of 50% adherence in communitysettings.
We have found one community study of the attrition rate of CBT for depressionin which volunteers were recruited via the local media for a 12-week courseof CBT. The total dropout rate was 47%, with almost half of those droppingout in the first 3 weeks.38 As with AD drugs,we model adherence ranging between the estimate reported in trials (81%)39 and a lower estimate of 50% in community settings.
The health benefit of interventions is measured in DALY, which is thesum of a nonfatal component determined by the severity-weighted time livedwith depression and a fatal component, years of life lost, calculated as thestream of life lost because of suicide.
As described elsewhere,40 we use 2 methodsto translate ESs from trial literature into a reduction in the disabilityweights. Briefly, the first method relies on an estimate of disability weightchange for each SD change in severity of depression, which we call the conversionfactor.41 Because the ES quantifies the impactof an intervention in SD units, health gain in DALY units can be calculatedas the product of the ES, the conversion factor, and the duration spent inthe health state. The second (survey severity) method applies the ES to themental component score of the SF-12 across eligible respondents in the mentalhealth survey, after which the difference in average disability weight withand without treatment is calculated. Results from both methods are incorporatedin our uncertainty analyses and hence broaden the uncertainty ranges aroundthe results presented.
Reductions in disability weight are only applied to the time from thecommencement of the intervention, ie, taking into account that there is alag to treatment-seeking after the onset of symptoms. A UK study found a median10-week interval between onset and care-seeking for patients with an affectivedisorder.42 We cannot assume a similar lagbecause the proportion of cases with a duration shorter than 10 weeks in acommunity sample is greater than the total proportion not seeking care inthe NSMHWB. Instead, based on expert consultation, we decide to model a lagvarying between 2 and 6 weeks.
As ESs are calculated from continuous measures and are not calculatedon an intention-to-treat basis, we apply the full nonadherence rate as a reductionin impact. For cases not adherent with treatment, no reduction in disabilityweight is modeled.
From the point prevalence of depression in the NSMHWB, a UK estimateof the relative risk for suicide of 20.4,43 andobserved suicide deaths in Australia in 2000, we derive suicide deaths attributableto depression by age and sex. We assume that a relative risk of 1.8 from Swedishroutine data collection systems44 applies totime lived with depression while not effectively taking AD drugs. In the absenceof long-term studies, we assume that suicide rates are similar in patientsreceiving CBT as in those taking AD drugs.
From these estimates, we derive suicide rates in those currently receivingeffective treatment and those ineffectively treated. As in the AustralianBurden of Disease Study,2 the years of lifelost associated with a death are calculated as the cohort life expectancyfor each age and sex category. We then divide the sum of years of life lostfor suicide in treated and untreated depression by the person-years of depressionin 2000.
The size of the burden averted by current treatment strategies requiresa back-calculation of the burden if no treatment were given. This is doneby applying the ES estimates for CBT and AD drugs to the mean disability weightof respondents in the NSMHWB receiving these treatments, taking into accountthe estimated lag to treatment and level of adherence.
We use simulation-modeling techniques and present uncertainty rangesinstead of point estimates that reflect all the main sources of uncertaintyin the calculations. Details of the parameters and distributions for the uncertaintyassumptions are shown in Table 1. Theprobability distributions around the input variables are based on (1) standarderrors quoted in or calculated from the literature, (2) a range of parametervalues quoted in or calculated from the literature, or (3) expert advice.We use the @RISK software (Palisade Corp, Newfield, NY), which allows multiplerecalculations of a spreadsheet each time choosing a value from uncertaintydistributions defined for input variables. We run a Monte Carlo simulationand calculate 95% uncertainty intervals for our output variables (boundedby the 2.5 and 97.5 percentiles of the 4000 values generated).
To identify the main sources of uncertainty affecting our results, weregress the values of each of the input variables against results in eachof the iterations of our simulation modeling. We report on input variableswith a regression coefficient greater than 0.2 or less than –0.2. Allresults are presented to 2 significant digits only.
The fitted lognormal distribution for the duration of episodes (correspondingto a normal distribution with a mean of 2.049 and SD of 1.599) has a meanof 27.9 weeks, resulting in an average duration of episodes of 29.9 weeksafter adding the minimum 2 weeks of duration. In combination with the fittedlognormal distribution of time to next episode (corresponding to a normaldistribution with a mean of 2.353 and SD of 3.876), the modeled mean numberof episodes during 5 years of follow-up after an episode is 2.4 and the meanproportion of time spent in major depression is 20.8%. The mean proportionof time spent with depression during 6 months after an episode is 19.5%.
The mean disability weights for mental health survey respondents onevidence-based treatment (0.429), those consulting health professionals butnot receiving evidence-based treatment (0.364), and those not consulting physicians(0.282) indicate that those with more severe disease are more likely to seekcare and to be offered potentially effective treatments. We attribute a reductionin disability weight from 0.490 (the hypothetical level of severity withouttreatment) to 0.429 for current treatment strategies.
In the year 2000, we estimate that 555 male and 198 female suicide deathsare attributable to major depression in Australia (or 30% of all suicides).Per person-year lived with major depression, the suicide risk is 0.8% in menand 0.3% in women. For both sexes combined, the risk of suicide in those takingmedication or receiving CBT is 0.26%, and in those not treated it is 0.47%.This translates on average across all ages into an annual loss of 0.093 yearsof life lost if treated and 0.167 years of life lost without treatment, ie,a net health gain of 0.074 years of life lost that we attribute to treatmentper year lived with depression.
During the first year after the onset of an episode of major depression,current treatment strategies avert 10% (95% uncertainty interval, 6%-12%)of the burden experienced by those in contact with health services. Treatmentduring the episode with an additional 6 months’ continuation treatmentcan raise this proportion to 28% (95% uncertainty interval, 19%-39%) withCBT and 24% (95% uncertainty interval, 19%-30%) with AD drugs (Figure 3A).
If all those seeking care for an episode are offered 5 years of maintenancetreatment, 52% (95% uncertainty interval, 42%-64%) of the burden can be avertedwith CBT and 50% (95% uncertainty interval, 40%-60%) with AD drugs comparedwith 13% (95% uncertainty interval, 10%-17%) under a scenario in which episodictreatment continues (Figure 3B).
The results for CBT are not very sensitive to the choice of ES for CBT(0.54 vs 0.77). The lower ES estimates bring the estimates of burden avertedby CBT down by less than 2 percentage points. Similarly, the proportion ofburden averted by AD drugs is only modestly sensitive to the assumed ES. Alteringthe ES for AD drugs by 25% alters results by 3 percentage points for episodictreatment and less than 1 percentage point for maintenance treatment.
The main sources of uncertainty in the model are the assumed treatmentdiscontinuation rates, the method of calculating a reduction in disabilityweight, and, to a lesser extent, the ESs.
Prevention of suicide contributes to almost a third of the amount ofhealth gain in DALY for each of the 4 intervention scenarios in comparisonwith no treatment. In the episodic treatment scenarios, reduction in severityis the main impact of treatments, whereas in maintenance treatment the impacton preventing relapse contributes more to overall health gain than reductionof severity while depressed (Table 2).
Our results strongly support longer-term treatment strategies for depression.Despite assuming rates of adherence to treatment of around 60%, we estimatethat half of depression experienced during 5 years after an episode of majordepression can be averted. The main reasons for this favorable outcome arethat maintenance treatment prevents relapses and that relapses that do occurare being treated from the start rather than after a lag time to seeking appropriatecare. Because the vast majority of people with depression experience multipleepisodes over a lifetime and are particularly prone to relapses shortly afteran index episode, there are convincing arguments for treating all depressionas a chronic disorder and not just those with recurrent or more severe episodesas recommended in current treatment guidelines.3,4
We have made a conscious choice to simplify our modeling by using averages,eg, for the severity of episodes and the duration of the index episode, andby modeling all ages and both sexes together. Some of these decisions do notdo justice to the great complexity and variation in the manifestation of depression.However, each added complexity requires more epidemiological input data withassociated uncertainty and is limited by the lack of efficacy data for differentdurations, severities, sex, and age. We believe we have struck a reasonablebalance. The model takes enough of the complexities into account but stillis simple enough for others to scrutinize and apply to other situations.
Elsewhere, we discuss the difficulties we encountered in translatingtrial findings into a health benefit in DALY terms.40 Tosome extent, we were able to incorporate this into our uncertainty analysisby using the range of results between 2 different methods of determining healthbenefit. The difference in burden averted between current practice and alternativetreatment options is less affected because the same imperfect method is usedfor each treatment scenario. More accurate measurements of change in healthstatus that can be attributed to interventions require further developmentalwork, such as the use of general quality-of-life outcome measures in trialsand more sensitive disability weights in DALY.
Our analyses are enhanced by the use of local epidemiological information.We had to rely on the 1997 NSMHWB as the only and most recent source for muchof the epidemiology of depression in Australia. Regular updates of the surveyare needed to sustain this kind of analysis in the future. Because this hasbeen the only community prevalence study in Australia, we are unable to incorporatetemporal trends in the occurrence of depression. However, the time horizonduring which we calculate our results is 5 years at most, and hence resultsare not much affected by the assumption of stable incidence of major depressiveepisodes. It would be very useful if a future survey identifying people withdepression in the community endeavored to follow up people over time to examineif our modeled assumptions of duration, time to next episode, and proportionof time with depression can be replicated in the Australian context.
The studies from which we derived our mathematical descriptions of theaverage duration of episodes and time to next episode are few and of relativelysmall size. Our 20% estimate of the average time with major depression during5 years of follow-up is higher than that from a clinical study in the UnitedStates, which found that 15% of time was spent with depressive symptoms atthe level of major depression during 9 years of follow-up.45 Ourresults are rather insensitive to this finding because the treatment impactmeasures applied to a 15% or 20% amount of depression during follow-up givesimilar estimates of the proportion of depression burden averted.
We have limited our analyses to major depression, ignoring that duringfollow-up time many people will spend time with subsyndromal symptoms or dysthymia.45 If we assume that treatments are also effective forthese types of depression, this means that we have underestimated the trueimpact of treatments.
The measures of efficacy of maintenance treatment strategies are derivedfrom studies of people who responded to treatment during an episode, and henceit is not evident that these would apply equally to all people with depressionas we have modeled. However, our results are not very sensitive to the estimatesof ES for AD drugs or CBT, and thus our conclusions would not alter even ifthe effectiveness of treatment in primary care cases is estimated to be asmuch as 25% higher or lower. The information we used from 2 European studies43,44 to determine the risk of suicideis not so strong. However, the inclusion of years of life lost from suicidein the analyses is important because it constitutes almost a third of theoverall health benefits. Our Australian estimates of suicide are high in comparisonwith a US estimate of suicide risk in people followed up after a diagnosisof depression.46 However, if we take into accountthat suicide rates in young adults are 30% higher in Australia (based on analysisof deaths reported to the World Health Organization, available at http://www3.who.int/whosis/mort) and that we estimated the risk of suicide only during depression andnot for all follow-up time, our estimates are only marginally higher (by 12%in men and 20% in women) than the US estimates.
Despite the limitations associated with a lack of data on the courseof depression and the impact of treatments, our results suggest that onlyby treating depression as a chronic episodic disorder with longer-term treatmentstrategies is it possible to make a meaningful reduction in the large burdenof depression in Australia. Similarities in community survey findings on theepidemiology of major depression in the US47,48 andAustralia1 and the predominantly US studieson the impact of treatments used in our model make it likely that our resultsalso have relevance to depression in the United States.
Psychological and drug treatments have similar impact on reducing thedepression burden, giving clinicians a choice of treatments. Additional informationon cost-effectiveness is needed to complement these results and inform prioritysetting.
Submitted for Publication: December 8, 2003;final revision received March 25, 2004; accepted April 22, 2004.
Correspondence: Theo Vos, MD, MSc, Schoolof Population Health, University of Queensland, Herston Rd, Herston QLD 4006,Australia (firstname.lastname@example.org).
Funding/Support: This study was conducted aspart of the Assessing Cost Effectiveness (ACE)–Mental Health projectfunded by the Australian Department of Health and Ageing, Canberra, Australia,and the Victorian Department of Human Services, Melbourne, Australia.
Acknowledgments: Unit record data of the NationalSurvey of Mental Health and Wellbeing were obtained from the Australian Bureauof Statistics, Canberra, Australia, and Gavin Andrews, MD, provided a revisedscoring algorithm to determine ICD-10 and DSM-IV diagnoses.