PHQ-9 accuracy: what a depression screening score can and can’t tell you

Scoring 11 on the depression check-in, with “Moderate depression symptoms” and a suggestion to talk to a professional, can leave you wondering whether it means you have depression. It is also common to score 8, “Mild depression symptoms”, two weeks later, or to get a different number from another depression test. So how far can you trust the score?

In short, the PHQ-9 is a well-studied screening tool. Researchers who pooled its studies describe it as the most commonly used tool for screening for depression in primary care (Levis et al., 2019), and when studies were pooled, a score of 10 picked out 85% of people whose major depression was confirmed in a clinician’s diagnostic interview and correctly ruled out 85% of people without it (Negeri et al., 2021). Still, screening is not diagnosis. In primary care, the researchers calculated that only about half of the people who score 10 or more would have major depression (Levis et al., 2019), and because the questions ask about the past two weeks, it is natural for the score to move from one time to the next.

This guide covers how the accuracy of a screening test is judged, how the cut-off of 10 performs when studies are pooled, why a score of 10 or more may not mean depression and when a low score misses it, why the score moves when you take it again, why question 9 is read on its own, and how to read and use your result. Where your score on the mind.io depression check-in (PHQ-9) sits among Korean adults is in average PHQ-9 scores, and what depression actually looks like is in depression symptoms.

If you have thoughts of harming yourself, do not wait until you finish reading: call or text 988 now in the US or Canada, call Samaritans on 116 123 in the UK or Ireland, or find a line at findahelpline.com elsewhere.

A screening test such as the PHQ-9 does not confirm an illness; it is a first step that picks out people who may need a closer look. Its accuracy is therefore judged against a diagnostic interview. Sensitivity is the share of people with major depression confirmed in the interview who score at or above the cut-off; specificity is the share of people without it who score below. The interviews used as the yardstick differ too: semistructured interviews are run by trained clinicians who add their own judgement, while fully structured interviews are read from a script, which raises standardisation but may cost some accuracy, as the researchers explain (Levis et al., 2019).

The PHQ-9 was first validated in the United States with 6,000 patients (8 primary care clinics and 7 obstetrics-gynaecology clinics). Of these, 580 primary care patients were interviewed by telephone by a mental health professional within 48 hours to confirm whether they had major depression (41 did); against those interviews, a score of 10 or more had a sensitivity and a specificity of 88% each. Internal consistency, how closely the items move together, was .89 in primary care and .86 in obstetrics-gynaecology (Kroenke et al., 2001).

The study also described what each score range tended to mean. Scores of 0 to 4 almost always meant no depressive disorder, 5 to 9 mostly meant none or a milder, subthreshold depression, and 15 or more usually meant major depression. Scores of 10 to 14 held a mix, which is why the researchers called the 10 to 15 range a “gray zone”. The likelihood of major depression rises steeply with the score: a score of 0 to 4 was only 0.04 times as likely in someone with major depression as in someone without it, while 20 to 27 was 36.8 times as likely. The cut-offs of 5, 10, 15 and 20 were also chosen for the practical reason that clinicians can remember them easily (Kroenke et al., 2001). So where your score sits matters more than whether it crossed a line.

The mind.io depression check-in (PHQ-9) uses this PHQ-9 as published. The nine items follow the original English wording, with Pfizer’s official translations for the other languages; each is answered from 0 to 3, the total of 0 to 27 is read with the same five ranges, and from 10 the result suggests talking with a professional. mind.io has not studied how accurate it is as an online check-in, though: every figure in this guide comes from PHQ-9 studies in clinics and communities.

The original study confirmed major depression in 41 people, by telephone interview. Later work pooled studies from many countries to check its accuracy again. The first meta-analysis of individual participant data (58 studies, 17,357 people) found that in the 29 studies (6,725 people) that used a semistructured interview, a score of 10 had a sensitivity of 88% and a specificity of 85% (Levis et al., 2019). In the larger update (100 studies, 44,503 people, 4,541 with major depression), the 47 studies (11,234 people) with a semistructured interview gave a sensitivity and a specificity of 85% each, and 10 was the cut-off with the best combined accuracy. Among people not already diagnosed or in treatment, the ones screening is meant for, sensitivity stayed the same and specificity rose to 89% (Negeri et al., 2021).

The yardstick matters as well. In the same update, against fully structured interviews read from a script, the sensitivity of a score of 10 fell to 64%, and against the MINI, a brief structured interview, to 74%, while specificity stayed similar at 88% and 89% (Negeri et al., 2021). The questionnaire was the same; what changed was the interview that decided who “really” had depression. The researchers think it may be because fully structured interviews classify more people with mild symptoms as having major depression (Levis et al., 2019).

It also matters who was studied. These analyses included only people aged 18 or over and left out people already identified as depressed or recruited from psychiatric care, since screening aims to find cases not yet recognised (Levis et al., 2019). Specificity was a little higher in people aged 60 or over and in men, but not by enough to justify different cut-offs by age or sex. Most languages and countries were represented by few studies, so the researchers could not analyse them separately (Negeri et al., 2021).

Studies within single countries sometimes found a different best cut-off. In Korea’s 2011 national psychiatric epidemiology survey of 6,022 adults, major depression confirmed with a scripted diagnostic interview (the Korean CIDI) was 2.5% (150 people), and the cut-off that best balanced sensitivity and specificity was 5 (sensitivity 89.9%, specificity 84.1%); that study did not report results for a cut-off of 10. The researchers noted that East Asian studies often find low cut-offs of 5 to 7, and suggested this may reflect a cultural reluctance to report depressive symptoms; studying the community rather than patients may also have played a part (Kim et al., 2023). In a Korean study of 567 psychiatric outpatients checked with the MINI, on the other hand, 10 was the appropriate cut-off (Shin et al., 2019).

In Japan, among 284 primary care patients (93 with major depression on the MINI-Plus interview), a score of 10 had a sensitivity of 90.5% and a specificity of 76.6% (Muramatsu et al., 2018, from the abstract); in a study that interviewed 104 of 598 internal medicine outpatients at a rural general hospital with the MINI and weighted the results, a cut-off of 5 had a sensitivity of 86% and a specificity of 85% (Inagaki et al., 2013, from the abstract); and among 521 new outpatients of a university hospital’s general medicine department checked with the MINI, the best balance came at 11 (sensitivity 76.2%, specificity 80.6%) (Suzuki et al., 2015). For French, a study comparing the questionnaire with a psychiatrist’s diagnosis in 212 internal medicine inpatients at the Geneva University Hospitals in Switzerland reported that it missed about half of the cases of major depression (Carballeira et al., 2007, from the abstract), while a Canadian study of 960 patients with systemic sclerosis who answered the English or the Canadian French version found small differences on three items but total scores that could be compared as they were (Arthurs et al., 2012).

So the best cut-off shifts a little with who is tested and against which interview. mind.io uses 10 in all four languages, the cut-off set out by the original study and confirmed by the meta-analyses. But a cut-off is a trade-off: raising it rules out more people without depression but misses more people with it, and lowering it does the opposite (Negeri et al., 2021). A score just below the line, such as 9, is worth taking seriously too.

A sensitivity and specificity of 85% do not mean that 85% of people who score 10 or more have depression. The share of people at 10 or more who really have major depression (the positive predictive value) depends heavily on how common major depression is among the people taking the test. The rarer it is, the more people without depression cross 10 compared with those who have it.

Here are the figures. In the pooled data from semistructured interviews, 14% of people had major depression, and under that condition the researchers calculated that 49% of those scoring 10 or more would have it. In primary care, where 12% had major depression, 22% of patients would score 10 or more, but only about half of them would have major depression (Levis et al., 2019). With an assumed prevalence of 5% to 25%, this share ranged from 23% to 65% (Negeri et al., 2021). In the original sample (7% with major depression), moving the cut-off from 9 to 15 changed it from 31% to 51% (Kroenke et al., 2001), and in the Korean general population study (2.5%), a cut-off of 5 gave 12.6% (Kim et al., 2023). Across whole groups, in 44 studies with 9,242 people, the share scoring 10 or more was 24.6% while major depression confirmed by interview was 12.1%, on average about 2.5 times as many per study (Levis et al., 2020, from the abstract). That is why the share at 10 or more reported by national surveys should not be read as the share of people with depression. How Korean adults score on the PHQ-9 is in average PHQ-9 scores.

This does not mean that someone who scores 10 or more without major depression has nothing going on. In the original study, about half of the 65 people with another, milder depressive disorder also scored 10 or more (Kroenke et al., 2001). One study suggested that conditions such as adjustment disorder or panic disorder may have raised scores (Shin et al., 2019), and a diagnosis also checks for physical causes of depressive symptoms (Kroenke et al., 2001). So a score of 10 or more is best taken as a sign that your situation is worth talking over with a professional, whether or not it turns out to be depression.

Errors go the other way too. A sensitivity of 85% means that about 15 in every 100 people with major depression score below 10 (Negeri et al., 2021), and in some settings, such as hospital inpatients, more cases were missed (Carballeira et al., 2007, from the abstract). If you scored low but a heavy mood has lasted more than two weeks or daily life feels hard, trust that feeling more than the number and ask for help.

Because of false positives and false negatives like these, countries take different views on screening. As of a 2019 paper, screening for depression was recommended in US primary care, but national guidelines in Canada and the UK advised against routine screening, citing false positives, overdiagnosis and the lack of trials showing that screening helps (Levis et al., 2019). A self-check you take alone works best not as a verdict but as a way to start a conversation with someone.

Every PHQ-9 item asks how often a problem has bothered you “over the last 2 weeks”. It measures your state over the past two weeks, not a lasting trait such as personality, so the score is meant to change when your situation does. The PHQ-9 is also used to follow change during treatment (Löwe et al., 2004, from the abstract).

Retaken at a short interval, scores are quite close. In the original study, the score patients gave themselves and a mental health professional’s telephone rating of the same nine symptoms within 48 hours correlated at 0.84, with averages of 5.08 and 5.03 (Kroenke et al., 2001). That compares two different ways of answering, not the same questionnaire taken twice. In a German study of 130 patients on a psychotherapy ward, most of them diagnosed with depression, who answered on paper and on a computer 24 hours apart, the two scores correlated at 0.92 and there was no significant difference between formats, but the second answers tended to come out very slightly lower (d = 0.07) (Erbe et al., 2016). So a difference of 1 or 2 points can come from the day itself.

How big a change means something? In a treatment trial for late-life depression (434 participants, mean age 71), data from subgroups who answered twice exactly 7 days apart put the smallest change that matters for an individual at 5 points, twice the standard error of measurement (Löwe et al., 2004, from the abstract). Changes smaller than 5 points are hard to tell apart from measurement wobble, so the direction over several check-ins says more than a single difference of 1 or 2 points.

On mind.io, band edges can also make a score look as if it jumped. Changing one answer by one step changes the total by 1 point, and the bands split between 4 and 5, 9 and 10, 14 and 15, and 19 and 20. So 9 (“Mild depression symptoms”) and 10 (“Moderate depression symptoms”) are one answer apart, and from 10 the result also suggests talking with a professional. The result page shows no separate note when a score sits near a band edge, so look at the score itself rather than the band name. If an earlier result is saved in the same browser, the result page shows how many points it moved and notes that scores reflect the past two weeks and naturally move up and down, and you can add a calendar reminder to check in again in two weeks.

Question 9 asks how often over the past two weeks you have been bothered by “Thoughts that you would be better off dead or of hurting yourself in some way”. It asks about thoughts of death or self-harm but not directly about suicidal intent (Wu et al., 2020). When the PHQ-9 is used to count diagnostic symptoms, this item alone counts if it is present at all, however briefly (Kroenke et al., 2001).

That is why question 9 alone cannot settle whether someone is at risk. Among 841 patients in a US registry of depression care centres, 41.1% answered question 9 with something other than “Not at all”, but 13.4% were found at risk on a brief electronic version of a scale that asks directly about suicidal thoughts and behaviour (the C-SSRS), which counted suicidal thoughts with intent to act or recent suicidal behaviour. Of those who answered question 9, 28.6% were also found at risk on that scale, and even among those who answered “Not at all”, 2.8% were found at risk on it. The researchers concluded that question 9 alone is not enough to assess suicide risk (Na et al., 2018, from the abstract). A summary of four studies in non-psychiatric settings (US veterans in primary care, patients with coronary artery disease and patients with cancer) found that only 18% to 35% of people who answered question 9 reported suicidal thoughts when asked about them specifically (four studies, as cited in Wu et al., 2020).

Question 9 should not be brushed aside either. In one US health system’s records of PHQ-9s completed at outpatient visits, people who reported such thoughts on more days in question 9 were more likely to make a suicide attempt over the following year, even after overall depression scores were taken into account. Most did not, but attempts also occurred among people who had answered “not at all” (Simon et al., 2013, authors’ manuscript). Records from the US veterans’ health system showed a similar pattern (Louzon et al., 2016, from the abstract). An answer to question 9 neither confirms nor rules out risk. So if you answered it, whatever your total, use it as a reason to tell someone you trust or a professional how you feel.

Research surveys often use the eight items without question 9 (the PHQ-8), so that they are not left with answers to question 9 they cannot follow up. The two totals correlated at 0.996, and their accuracy in finding depression was almost the same (Wu et al., 2020). mind.io asks question 9 and responds to it separately: if you choose anything other than “Not at all”, crisis lines appear right under the question, and the result page shows them at the top whatever your total and suggests talking with a professional. Even with a total under 10, it suggests talking to someone soon instead of checking in again in two weeks, and this answer is never put in a link or sent to analytics.

Whatever your score, if you have thoughts of harming yourself, do not carry them alone: call or text 988 now in the US or Canada, call Samaritans on 116 123 in the UK or Ireland, or find a line at findahelpline.com elsewhere.

First, look at the score and how far it is from the line, not just the band name. A 9 and a 10 have different names and messages, but they are one answer apart. At 9, rather than deciding you are fine, look at which items were high; at 11, read it not as a diagnosis but as a signal to talk to someone.

Second, check in again about two weeks apart, in similar circumstances. The PHQ-9 asks about the past two weeks, so checking more often measures overlapping periods. Look at the direction over several check-ins, and at whether the score moved by 5 points or more, rather than at a single difference of 1 or 2 points (Löwe et al., 2004, from the abstract).

Third, look at how much these problems get in the way of daily life, not only at the score. The original PHQ-9 form has a tenth question on how difficult the problems have made it to do your work, take care of things at home or get along with other people; it is not part of the total, and mind.io does not ask it (Kroenke et al., 2001). The same score can weigh very differently on different lives, so think this part through yourself.

Fourth, if your total is 10 or more, if you answered question 9, or if a hard time has lasted more than two weeks whatever your score, talk to a professional. Outside Korea, your family doctor or a local mental health service is a good place to start; in Korea, a community mental health welfare centre or a psychiatry clinic. Saying “I took this check-in and this is my score” is an easy way to begin. What happens at a first visit is in seeing a psychiatrist for the first time.

What not to use it for: do not use a PHQ-9 score alone to decide that you or someone else has depression, or to dismiss a hard time because the score was low. A diagnosis also looks at physical illness, normal bereavement and any history of mania (Kroenke et al., 2001). The PHQ-9 does not ask about periods of unusually high or elevated mood; if you have had them, read about bipolar disorder and be sure to tell a professional. The meta-analyses in this guide included only people aged 18 or over, so a teenager is better off reading the result with a parent, a school counsellor or a professional. If worry and tension weigh more, the anxiety check-in (GAD-7) can help too. It is not a diagnosis either.

Q. Is the PHQ-9 a trustworthy test? A. As a tool for screening depression symptoms, it has been studied a great deal. In an analysis of individual data from 100 studies and 44,503 people, the studies that used a clinician’s semistructured interview showed that a score of 10 picked out 85% of people with major depression and correctly ruled out 85% of people without it (Negeri et al., 2021). Being a screening tool, though, it does not diagnose on the score alone. The mind.io check-in uses the same nine items and scoring.

Q. Does 10 or more mean I have depression? A. Not on its own. In primary care, the researchers calculated that only about half of the people who score 10 or more would have major depression in an interview (Levis et al., 2019). Still, 10 or more is a sign that your current state is worth talking over with a professional, because other difficulties can be there even when it is not major depression.

Q. Am I fine if I score 9 or less? A. Not necessarily. About 15 in every 100 people with major depression score below 10 (Negeri et al., 2021). If you answered question 9, if a hard time has lasted more than two weeks or if daily life feels too much, it is fine to ask for help whatever your score.

Q. My score is different from last time. Which one is right? A. Each reflects the two weeks before that check-in. A difference of 1 or 2 points is common, and the smallest change that matters for an individual is estimated at about 5 points (Löwe et al., 2004, from the abstract). Look at the direction over several check-ins rather than at one difference.

Q. Another depression test gave me a different result. A. If the test has different questions, time frame or scoring, the numbers cannot be compared directly. Even the same PHQ-9 can differ slightly on a few items between translations, but a study comparing the English and Canadian French versions found that total scores could be compared as they were (Arthurs et al., 2012).

Q. How often can I take it again? A. The PHQ-9 asks about the past two weeks, so about two weeks apart works well. The mind.io result page lets you add a calendar reminder for two weeks later. If you answered question 9, though, the result page suggests talking to someone soon rather than checking in again, and does not offer the reminder.

Q. I am having thoughts of ending my life. A. Please do not carry them alone. Reach out to a crisis line right away: call or text 988 in the US and Canada, call Samaritans on 116 123 in the UK and Ireland, or find a line at findahelpline.com elsewhere. If you are in immediate danger, contact your local emergency services.

References: Arthurs, E., Steele, R. J., Hudson, M., Baron, M., Thombs, B. D., & Canadian Scleroderma Research Group (2012), Are scores on English and French versions of the PHQ-9 comparable? An assessment of differential item functioning, PLoS ONE 7(12), e52028, https://doi.org/10.1371/journal.pone.0052028; Carballeira, Y., Dumont, P., Borgacci, S., Rentsch, D., de Tonnac, N., Archinard, M., & Andreoli, A. (2007), Criterion validity of the French version of Patient Health Questionnaire (PHQ) in a hospital department of internal medicine, Psychology and Psychotherapy: Theory, Research and Practice 80(1), 69-77, https://doi.org/10.1348/147608306X103641; Erbe, D., Eichert, H.-C., Rietz, C., & Ebert, D. (2016), Interformat reliability of the patient health questionnaire: Validation of the computerized version of the PHQ-9, Internet Interventions 5, 1-4, https://doi.org/10.1016/j.invent.2016.06.006; Inagaki, M., Ohtsuki, T., Yonemoto, N., Kawashima, Y., Saitoh, A., Oikawa, Y., Kurosawa, M., Muramatsu, K., Furukawa, T. A., & Yamada, M. (2013), Validity of the Patient Health Questionnaire (PHQ)-9 and PHQ-2 in general internal medicine primary care at a Japanese rural hospital: A cross-sectional study, General Hospital Psychiatry 35(6), 592-597, https://doi.org/10.1016/j.genhosppsych.2013.08.001; Kim, M., Jung, S., Park, J. E., Sohn, J. H., Seong, S. J., Kim, B.-S., Chang, S. M., Hong, J. P., Hahm, B.-J., & Yeom, C.-W. (2023), Validation of the Patient Health Questionnaire-9 and Patient Health Questionnaire-2 in the general Korean population, Psychiatry Investigation 20(9), 853-860, https://doi.org/10.30773/pi.2023.0100; Kroenke, K., Spitzer, R. L., & Williams, J. B. W. (2001), The PHQ-9: Validity of a brief depression severity measure, Journal of General Internal Medicine 16(9), 606-613, https://doi.org/10.1046/j.1525-1497.2001.016009606.x; Levis, B., Benedetti, A., & Thombs, B. D. (2019), Accuracy of Patient Health Questionnaire-9 (PHQ-9) for screening to detect major depression: Individual participant data meta-analysis, BMJ 365, l1476, https://doi.org/10.1136/bmj.l1476; Levis, B., Benedetti, A., Ioannidis, J. P. A., Sun, Y., Negeri, Z., He, C., Wu, Y., Krishnan, A., Bhandari, P. M., Neupane, D., et al. (2020), Patient Health Questionnaire-9 scores do not accurately estimate depression prevalence: Individual participant data meta-analysis, Journal of Clinical Epidemiology 122, 115-128.e1, https://doi.org/10.1016/j.jclinepi.2020.02.002; Louzon, S. A., Bossarte, R., McCarthy, J. F., & Katz, I. R. (2016), Does suicidal ideation as measured by the PHQ-9 predict suicide among VA patients?, Psychiatric Services 67(5), 517-522, https://doi.org/10.1176/appi.ps.201500149; Löwe, B., Unützer, J., Callahan, C. M., Perkins, A. J., & Kroenke, K. (2004), Monitoring depression treatment outcomes with the Patient Health Questionnaire-9, Medical Care 42(12), 1194-1201, https://doi.org/10.1097/00005650-200412000-00006; Muramatsu, K., Miyaoka, H., Kamijima, K., Muramatsu, Y., Tanaka, Y., Hosaka, M., Miwa, Y., Fuse, K., Yoshimine, F., Mashima, I., Shimizu, N., Ito, H., & Shimizu, E. (2018), Performance of the Japanese version of the Patient Health Questionnaire-9 (J-PHQ-9) for depression in primary care, General Hospital Psychiatry 52, 64-69, https://doi.org/10.1016/j.genhosppsych.2018.03.007; Na, P. J., Yaramala, S. R., Kim, J. A., Kim, H., Goes, F. S., Zandi, P. P., Vande Voort, J. L., Sutor, B., Croarkin, P., & Bobo, W. V. (2018), The PHQ-9 Item 9 based screening for suicide risk: A validation study of the Patient Health Questionnaire (PHQ)-9 Item 9 with the Columbia Suicide Severity Rating Scale (C-SSRS), Journal of Affective Disorders 232, 34-40, https://doi.org/10.1016/j.jad.2018.02.045; Negeri, Z. F., Levis, B., Sun, Y., He, C., Krishnan, A., Wu, Y., Bhandari, P. M., Neupane, D., Brehaut, E., Benedetti, A., Thombs, B. D., & the DEPRESSD PHQ Group (2021), Accuracy of the Patient Health Questionnaire-9 for screening to detect major depression: Updated systematic review and individual participant data meta-analysis, BMJ 375, n2183, https://doi.org/10.1136/bmj.n2183; Shin, C., Lee, S.-H., Han, K.-M., Yoon, H.-K., & Han, C. (2019), Comparison of the usefulness of the PHQ-8 and PHQ-9 for screening for major depressive disorder: Analysis of psychiatric outpatient data, Psychiatry Investigation 16(4), 300-305, https://doi.org/10.30773/pi.2019.02.01; Simon, G. E., Rutter, C. M., Peterson, D., Oliver, M., Whiteside, U., Operskalski, B., & Ludman, E. J. (2013), Does response on the PHQ-9 Depression Questionnaire predict subsequent suicide attempt or suicide death?, Psychiatric Services 64(12), 1195-1202, https://doi.org/10.1176/appi.ps.201200587; Suzuki, K., Kumei, S., Ohhira, M., Nozu, T., & Okumura, T. (2015), Screening for major depressive disorder with the Patient Health Questionnaire (PHQ-9 and PHQ-2) in an outpatient clinic staffed by primary care physicians in Japan: A case control study, PLoS ONE 10(3), e0119147, https://doi.org/10.1371/journal.pone.0119147; Wu, Y., Levis, B., Riehm, K. E., Saadat, N., Levis, A. W., Azar, M., Rice, D. B., Boruff, J., Cuijpers, P., Gilbody, S., et al. (2020), Equivalency of the diagnostic accuracy of the PHQ-8 and PHQ-9: A systematic review and individual participant data meta-analysis, Psychological Medicine 50(8), 1368-1380, https://doi.org/10.1017/S0033291719001314.