What Is Depression?

A major depressive episode requires five of nine symptoms in the same two-week period. At least one of them has to be low mood or loss of interest or pleasure, lasting most of the day, nearly every day. Four of the other seven are significant weight change or a change in appetite, insomnia or sleeping too much, agitation or slowing that someone else can see, and fatigue. The rest are worthlessness or misplaced guilt, trouble concentrating or deciding, and recurrent thoughts of death or suicide, or a suicide attempt.

The symptoms have to be new or clearly worse than before, and they have to cause clinically significant distress or impairment. Two editions of the American Psychiatric Association's Diagnostic and Statistical Manual of Mental Disorders (DSM), DSM-IV in 1994 and DSM-5 in 2013, list the same nine symptoms. DSM-5-TR in 2022 kept them, rewording the suicide symptom and an exclusion for psychotic disorders.

One 2013 change was to delete the bereavement exclusion, and the Substance Abuse and Mental Health Services Administration counts it among DSM-5's more controversial changes. The exclusion had advised clinicians not to diagnose major depression in the first two months after the death of a loved one, unless certain serious symptoms were present. Suicidal thoughts and psychotic symptoms were among them.

The American Psychiatric Association pointed out that no other stressor, such as losing a job, blocked the diagnosis. The administration's report gives further reasons for the change: depression after a death has similar risk factors and course to other depression, and responds to the same treatments.

Jerome Wakefield and Michael First argued that there was no scientific basis for the removal, and that some normal grief could be misdiagnosed as major depression. DSM-5-TR later added prolonged grief disorder. In adults, it covers grief that still causes clinically significant distress or impairment more than 12 months after the death. DSM-5-TR says both diagnoses are given when both are met.

Alongside the episodic form there is persistent depressive disorder: low mood most of the day, more days than not, for at least two years in adults. It also needs two of six further symptoms, with no gap longer than two months. It merged what used to be dysthymia and chronic major depression.

Five of nine, one of which must be from a fixed pair, allows 227 different ways to qualify. Three of the criteria are two-sided. You can meet the sleep item by sleeping far too little or far too much, and the appetite item by eating far too little or far too much. The psychomotor item can be met by agitation or by slowing. Count those poles separately and the number goes to 945. Count the sub-parts of most compound items too, as Coomans and Fried did, and it reaches 10,377.

So one person can have low mood, insomnia, poor appetite, agitation and guilt. Another can have lost interest and pleasure, and be sleeping far too much, eating far more, slowed down and unable to concentrate. Both meet the criteria for major depression, and they share no symptom.

Zimmerman and colleagues add that nine combinations, a relatively small number, accounted for more than 40% of the patients diagnosed.

Fried and Nesse used the baseline data from STAR*D (Sequenced Treatment Alternatives to Relieve Depression), one of the largest clinical trials of depression. In that trial, the 501 profiles held by a single person mean that roughly one patient in seven had a symptom picture shared with nobody else. Controlling for how severe people were did not reduce the spread.

Fried has also shown that seven common depression rating scales between them contain 52 distinct symptoms. The scales overlap by about a third on average (0.36, on a measure from 0 for no overlap to 1 for full overlap). He concludes that research results may depend on which scale a study used.

The melancholic subtype does not narrow the number of ways to qualify. On Coomans and Fried's counts, depression with melancholic features can be met in 10,999 to 341,737 ways, more than major depression itself.

Fried and Nesse conclude that depression is not one coherent condition with a single cause. They weigh this against the view that it is one disease with varied symptoms. That view, they say, remains possible but is increasingly unlikely, because no biomarker for it has been replicated.

Three American national surveys of adults put lifetime prevalence of major depression at about 13% to about 21%.

The surveys used different interviews, and NESARC-III applied DSM-5, the diagnostic manual's 2013 edition, which no longer excludes bereavement. NESARC-III's authors wrote that dropping the exclusion could explain a small amount of the increase since NESARC-I, not all of it, and that other national reports support a real one. NESARC-I and NESARC-III both found women at higher risk; in NESARC-III, men's odds of past-year depression were half women's.

Whether depression is becoming more common is not settled. Compton and colleagues found past-year prevalence among US adults rising from 3.33% in 1991–92 to 7.06% in 2001–02, on the same criteria. The Stirling County Study in Atlantic Canada found current depression stable at about 5% in samples from 1952, 1970 and 1992, on its own measures. Ormel and colleagues, reviewing the evidence in 2022, found little evidence that prevalence has risen.

People with depression die at roughly twice the rate of people without it. A 2025 review in World Psychiatry pooled 268 cohort studies and 10.8 million people with depression. It put all-cause mortality at a relative risk of 2.10 (95% confidence interval 1.87 to 2.35). Against people matched for their other medical conditions, the relative risk was still 1.29.

Suicide carries the highest relative risk in that review, 9.89, but the authors note, citing earlier studies, that the excess deaths are mainly from natural causes. A Danish register study shows both.

Osler and colleagues, 2025: 330,577 adults diagnosed with major depression in Danish hospital registers from 1970 to 2021, and five matched people for each. Cumulative risk of death ten years after diagnosis.

In the first year after diagnosis, suicide was the leading cause of death among the Danish patients, at 0.9%, and it was also the leading cause before age 65.

For about thirty years, many papers and textbooks repeated that 15% of people with depression die by suicide. As Bostwick and Pankratz put it in 2000, the figure came from Guze and Robins's 1970 meta-analysis of patients with affective illness and was then applied to all depression.

Bostwick and Pankratz argued that the figure used a less appropriate measure: suicides as a share of the patients who had died, called proportionate mortality. They recalculated it as case fatality, suicides as a share of everyone diagnosed, and split it by treatment setting.

Blair-West and colleagues worked from population figures: given how common major depression is and how many suicides occur, the lifetime risk cannot be 15%. Their 1999 estimate was 3.4%. By sex, it was about 7% for men and 1% for women. Boardman and Healy modelled 212 suicides in one English health district against community prevalence rates. They got 2.4% for any affective disorder, and 1.1% for cases without substance misuse, personality disorder or non-affective psychosis that had never reached mental health services.

The risk is highest in the first year after diagnosis and among people who have been in hospital. In the Danish register study, the lifetime risk of suicide was 11.2% for people diagnosed with depression by hospital services, against 1% in the matched population.

Download this conclusion: the 15% figure

This section as one image: the old 15% figure, the lifetime risks of 8.6%, 4.0% and 2.2% by treatment setting, and the 2.4% to 3.4% that population models give.

Download the card (PNG, 1,080 × 1,350)

Depression ranked second among all causes of years lived with disability worldwide in the 2019 Global Burden of Disease estimates, and thirteenth for total burden. Its total burden counts no deaths, because the estimates file suicide under injuries. Depression's figures are modelled: estimated prevalence multiplied by disability weights. The weights come from surveys asking people which of two described health states is healthier, and from questions placing them on a scale from perfect health to death.

Course has been followed both in the general population and in clinic patients, though the studies define recovery and recurrence differently. Eaton and colleagues followed 92 people in Baltimore through their first lifetime episode, in a study with 23 years of follow-up. Around half recovered and never had another episode. Of the 78 who recovered, 35 later recurred. Roughly 15% of the 92 had no year free of it. Median episode length was twelve weeks, though the first episode lasted a median of 20 weeks.

The National Institute of Mental Health's Collaborative Program on the Psychobiology of Depression followed 318 patients in its clinical studies for ten years. Nearly two-thirds had at least one recurrence.

A 2000 review of family, adoption and twin studies concluded that major depression results from both genetic and environmental influences, not from either alone. The research on causes asks how much of depression comes from genes, how much from circumstance, and whether effort changes it.

Twin studies put heritability at 37%, from that review's meta-analysis by Sullivan, Neale and Kendler, who called this probably the lower bound. They put the likely range at 31% to 42%. A Swedish national twin study put it at 42% for women and 29% for men. Shared family environment contributes almost nothing; the rest is environment specific to the individual, together with measurement error. Measured directly from genetic data, common variants account for about 9%, around a quarter of the twin estimate.

Large genome-wide studies found 44 associated variants in 2018, then 102 in 2019. A genetic risk score from the 2018 study ranked a person with depression above one without 57% of the time. A coin does it 50% of the time.

Before genome-wide methods, the field tested single plausible genes in samples far smaller than today's. One of the best-known results was Caspi and colleagues' in 2003. Carriers of one or two short versions of a stretch of the serotonin transporter gene, 5-HTTLPR, had more depression after stressful life events than people with two long versions.

Culverhouse and colleagues tested Caspi's result in 38,802 people of European ancestry. They found no subgroup or definition in which the interaction was statistically significant, while stress itself was a strong risk factor. Border and colleagues tested eighteen candidate genes, in samples of 62,138 to 443,264 people. They found no clear evidence for any of them, alone or with the environment. Their results, they concluded, suggest that the many associations reported in that literature are likely to be false positives.

Moncrieff and colleagues published an umbrella review in 2022 covering six areas of serotonin research. They concluded there is no consistent evidence of an association between serotonin and depression, and no support for the idea that depression is caused by lowered serotonin.

Jauhar, Cowen and more than thirty co-authors replied in 2023 that Moncrieff's conclusion was overstated, and that the review was not a standard umbrella review. They also said its certainty ratings used a modified version of a standard system, added after the review's protocol was set. Moncrieff and colleagues answered that they had synthesised all existing studies in the main areas, and that the onus is on others to make a convincing case. Whether antidepressants work is a separate question, tested by the trials in the next section.

Stressful life events predict the onset of depression. In a 1999 study of female twins by Kendler and colleagues, the comparison within identical twin pairs held their shared genes constant. Even within those pairs, the odds of a new episode were 3.58 times higher in the month of a stressful event. The authors still judged about a third of the overall link non-causal.

Brown and Harris found depression more common among working-class women in Camberwell, in London. By the University of Essex's account, they tied the difference to life events and to vulnerability factors, such as having no one to confide in.

A procedural replication in Calgary, published in 1982, found no link between new episodes in the past year and social class, employment or the number of children at home. But it confirmed the links with severe life events and difficulties, and with lacking intimacy with a partner. It concluded that the role of social factors is community-specific, and that the causal role of events and difficulties remains uncertain.

Behavioural activation is a therapy that uses scheduled activities to help people reconnect with what they find rewarding. A 2026 review gathered 105 trials of it, with 13,933 patients. In 61 comparisons with control conditions in adult outpatients, its standardised mean difference was 0.67. The effect was still significant 12 months after randomisation.

Against other therapies, in 27 comparisons, the difference was not significant. Across the 61 control comparisons, heterogeneity was high and the prediction interval crosses zero, so a new trial could find no benefit. Self-guided behavioural activation also worked, with a smaller effect of 0.36.

In 2018, Cipriani and colleagues pooled 522 double-blind trials covering 116,477 adults. The trials covered 21 antidepressants, and all 21 beat placebo. The odds ratios for response, meaning symptom scores fell by at least half, ran from 2.13 for amitriptyline down to 1.37 for reboxetine. The overall drug-placebo difference came to a standardised mean difference of 0.30. Nine per cent of the trials were rated high risk of bias, 73% moderate, and the authors graded the certainty of the evidence as moderate to very low.

Irving Kirsch and colleagues worked from the Food and Drug Administration's own data on four newer antidepressants. They put the pooled difference at 1.8 points on the Hamilton scale. Drug groups improved by 9.6 points and placebo groups by 7.8. The National Institute for Health and Care Excellence (NICE) had defined a clinically significant benefit as a 3-point difference or a standardised mean difference of 0.50. The pooled difference, a standardised mean difference of 0.32, falls short of both.

Fountoulakis and Möller recalculated Kirsch's data and got 2.18 or 2.68 points. Venlafaxine and paroxetine came out above 3. Kirsch and colleagues replied that the recalculation weighted the trials wrongly; done correctly, no single drug reached NICE's criterion. The recalculation's own standardised effect size matched theirs, substantially below the criterion.

Turner and colleagues showed why the published literature looks better than the drugs are. Of 74 trials registered with the Food and Drug Administration (FDA), 31% were not published. By the FDA's reading 51% were positive; in the published record 94% appeared positive. The effect size in print was 0.41, against 0.31 in the FDA's data for all trials, published and unpublished alike.

The headline figure of the STAR*D (Sequenced Treatment Alternatives to Relieve Depression) trial came from Rush and colleagues in 2006. It is a cumulative remission rate of 67% after up to four sequential treatment steps. The trial's report called that figure theoretical: it assumed that the people who dropped out would have remitted at the same rate as those who stayed. In Pigott and colleagues' reanalysis, 53.7% dropped out.

Pigott and colleagues reanalysed the patient-level data in 2023, following the trial's protocol and, where it was silent, the trial's other papers. The protocol specified the Hamilton scale, given by blinded research assessors, rather than the self-report scale behind the 67%. Following the other papers, they excluded patients the published analysis had included. They got 35.0%, or 41.3% when patients with no final Hamilton score are also counted if their last self-report score showed remission.

Rush and colleagues replied that the reanalysis dropped 941 patients. The difference in remission rates, they said, was most likely the result of excluding hundreds with low symptom scores when they left. They also pointed to a separate study, co-authored by Pigott's senior author, that found 60% remission over 12 months of antidepressant treatment.

A third analysis, by Sakurai and colleagues in 2024, used the self-report scale and weighted for dropout. It estimated 53.8% at 90 days and 74.5% at six months. At a year, the estimate was 87.5%. One dataset, three numbers, none of them from a placebo-controlled trial, because STAR*D had no placebo arm.

Download this conclusion: STAR*D

This section as one image: the 67% STAR*D reported, the 35.0% a reanalysis found on the measure the trial's protocol specified, and what each assumed.

Download the card (PNG, 1,200 × 630)

Cuijpers and colleagues have measured how far the literature on psychotherapy for depression overstates its effects. Their 2010 meta-analysis found an overall effect size of 0.67, falling to 0.42 once publication bias was accounted for. In 2019 they adjusted for risk of bias, publication bias and waiting-list controls together, and left out trials from non-Western countries. The pooled figure fell from an unadjusted 0.70 to 0.31. Only 23% of the studies were at low risk of bias.

Munder and eight co-authors reanalysed the same effect sizes, treating waiting lists as the right comparison and removing outliers. They found about 0.70 against waiting lists and 0.31 against usual care. Cuijpers and colleagues replied in the same journal.

A 2021 network meta-analysis of 331 trials by Cuijpers and colleagues compared psychotherapies. Cognitive behavioural therapy, behavioural activation, interpersonal therapy and psychodynamic therapy did not differ significantly from each other; only non-directive supportive counselling did worse.

A 2024 network meta-analysis in the BMJ compared exercise with active controls such as usual care. It covered 218 studies. Their 495 arms held 14,170 participants.

Only one of the 218 studies met Cochrane's criteria for low risk of bias. The authors rated their confidence low for walking or jogging and very low for every other kind of exercise. They conclude that exercise is an effective treatment; how large its effect is remains uncertain.

Davies and Read's 2019 review of the withdrawal literature reported a weighted average incidence of 56%. Across its studies, incidence ranged from 27% to 86%. In four large studies, 46% of those affected chose the most severe rating offered. Jauhar, Hayes and colleagues replied that these estimates leaned on convenience-sample surveys, two of them of people who reported withdrawal needing treatment. Such surveys, they wrote, cannot be extended to everyone taking antidepressants.

Henssler and colleagues pooled 79 studies in 2024. They found at least one symptom in 31% of patients stopping an antidepressant, against 17% of those stopping a placebo. Severe symptoms were rarer: 2.8% against 0.6%.

The numbers are still disputed, but the guidance moved. In 2019 the Royal College of Psychiatrists said some patients can suffer more severe symptoms that can last much longer. It asked NICE to update its advice, which suggested most people should be able to come off antidepressants over four weeks. NICE now says the dose should be reduced in stages.

In the United States, outpatient treatment for depression rose from 0.73 people in every 100 in 1987 to 2.33 in 1997. By 2007 it was 2.88, about four times the 1987 rate.

Ormel, Hollon, Kessler, Cuijpers and Monroe reviewed this in 2022 and called it the treatment-prevalence paradox. In their account, treatments have improved and become far more available since the 1980s, but the prevalence of depression has not decreased.

They found little evidence that more misdiagnosis or more new cases had hidden a real fall. They concluded that three explanations probably account for most of the paradox:

• published trials overstate how well treatments work;
• treatments work considerably less well in ordinary clinics than in trials;
• treatment's impact differs substantially between non-recurrent cases and chronic or recurrent ones, which make up most of the depression in a population at any one time.

This does not mean the treatments do nothing: in trials, antidepressants beat placebo and psychotherapy beats control conditions, by modest amounts.

In 2023 Mulder and Jorm compared countries in the Organisation for Economic Co-operation and Development (OECD). Countries prescribing more antidepressants, they found, did not have less sadness, worry or unhappiness. For worry, the link came close to statistical significance in the opposite direction: more prescribing, more worry. Countries with a higher Human Development Index, which combines income, education and life expectancy, had less of all three.

Major depression is diagnosed from a checklist of five of nine symptoms over two weeks. The criteria can be met in 227 ways. Because some symptoms can run in opposite directions, such as sleeping too little or too much, two people with the same diagnosis can share no symptom.

People with depression die at about twice the rate of others, mainly of natural causes, though suicide carries the highest relative risk. For depression and other affective disorders in the community, models put the lifetime risk of suicide at 2.4% to 3.4%. That is far below the old figure of 15%.

Genes and circumstance both play a part. Twin studies put heritability at 37%, probably a lower bound, and stressful events raise the odds of an episode even within identical twin pairs. Moncrieff and colleagues' 2022 review found no support for the idea that low serotonin causes depression; its critics call that conclusion overstated.

Treatment helps, modestly. Antidepressants beat placebo by a standardised mean difference of 0.30. Psychotherapy's effect falls from 0.70 to 0.31 once trial quality, waiting-list comparisons and publication bias are allowed for and non-Western trials are left out. Exercise helps, by an amount that weak trials leave uncertain.

Yet as treatment has spread, the prevalence of depression has not fallen. Ormel and colleagues' 2022 review concluded that overstated trial results, weaker results in ordinary clinics and a different impact on chronic and recurrent cases probably account for most of that.

Download the conclusion

The verdict on treatment as one image, with the three numbers that frame it. Antidepressants beat placebo by a standardised mean difference of 0.30. Psychotherapy's effect is 0.70 before allowing for bias and 0.31 after. The image is 1,200 by 630 pixels, the size social networks use for link previews, and anyone may download it and post it.

Download the card (PNG, 1,200 × 630)