PHQ-9

PHQ-9 in Mental Health Trials and Economic Evaluation

The Patient Health Questionnaire-9 is the depression module of the Patient Health Questionnaire, and a single total from it can both flag probable major depression and grade how severe the symptoms are. For a health economist the questionnaire matters because trials of collaborative care and computerised cognitive behavioural therapy record depression on this scale, and because NHS Talking Therapies services in England judge recovery partly on it. The total is a symptom count weighted equally across items, with no preference weights attached, so a cost per QALY can be built on PHQ-9 data only through a preference-based measure collected alongside it or through a published mapping function. Mukuria, Franklin and Hinde estimated such functions for the EQ-5D-3L from four depression trials, compared how well different model forms fitted the data and recommended a four-component mixture model.

Items, response scale and scoring of the PHQ-9

Kroenke, Spitzer and Williams describe the Patient Health Questionnaire as a self-administered version of the PRIME-MD diagnostic instrument for common mental disorders, and the PHQ-9 as its depression module. Each of the nine items corresponds to one of the nine DSM-IV criteria for major depression. The developers' published form asks respondents how often, over the last 2 weeks, they have been bothered by each problem.

Every item is scored from 0, labelled "not at all", to 3, labelled "nearly every day", and the item scores are added. With nine items and a top score of 3 each, 9 × 3 = 27, so the total runs from 0 to 27. Higher totals mean more frequent depressive symptoms. The developers' screener site states that the PHQ-9 and its translations need no permission to reproduce, translate, display or distribute. The site credits Spitzer, Williams, Kroenke and colleagues with developing the measures under an educational grant from Pfizer.

Kroenke and colleagues proposed four severity thresholds. The bands below follow directly from them.

Validation in primary care and obstetrics-gynaecology clinics

The original validation drew on 6,000 patients who completed the PHQ-9 in 8 primary care clinics and 7 obstetrics-gynaecology clinics. A subsample of 580 patients was reinterviewed by a mental health professional using a structured interview, which served as the criterion standard. Against that interview, a PHQ-9 score of 10 or more had a sensitivity of 88% and a specificity of 88% for major depression.

Construct validity was tested against the 20-item Short-Form General Health Survey (SF-20), self-reported sick days, clinic visits and symptom-related difficulty. As PHQ-9 severity rose, functional status fell on all six SF-20 subscales, while sick days, health care use and symptom-related difficulty increased. The authors concluded that the PHQ-9 is a reliable and valid measure of depression severity as well as an aid to diagnosis.

That dual role matters in economic work. The threshold of 10 is a diagnostic cut point chosen for its sensitivity and specificity, while the continuous total is a severity score. A trial that reports the proportion above 10 and a trial that reports mean scores are answering different questions, and each feeds a model differently.

PHQ-9 in NHS Talking Therapies outcome reporting

NHS Talking Therapies for anxiety and depression, the programme formerly called Improving Access to Psychological Therapies (IAPT), uses the PHQ-9 as its routine depression measure. The NHS England manual for the programme (version 7, first published June 2018 and updated March 2024) sets a clinical cut-off and a reliable change index for each measure in its Table 9. For the PHQ-9 the clinical cut-off is 10 and the reliable change index is 6. For the GAD-7, the default anxiety measure, the cut-off is 8 and the reliable change index is 4.

The programme's outcome definitions rest on these numbers. The NHS England service standards describe reliable recovery as applying to people who were above the clinical cut-off before treatment, fell below it after treatment and also reliably improved, and the manual states that a person has reliably recovered when they meet the criteria for both recovery and reliable improvement. The manual asks for recovery, reliable improvement, reliable recovery and reliable deterioration to be calculated on a pair of measures, the PHQ-9 with the anxiety measure suited to the presenting problem, and falls back on the PHQ-9 with the GAD-7 when that anxiety measure is missing. The manual also notes that the GAD-7 was originally developed for generalised anxiety disorder only.

For mental health economics these service definitions supply ready-made binary outcomes. An evaluation can report cost per person reliably recovered, which is a cost-effectiveness ratio in natural units. Such a ratio cannot be compared with a cost per QALY threshold. It also ignores the size of change beyond the thresholds, so on the PHQ-9 alone a person who starts at 11 and ends at 5 meets the recovery and reliable change conditions, while a person who starts at 26 and ends at 14 does not, despite twice the improvement. The paired anxiety measure also has to be taken into account.

Turning PHQ-9 scores into utilities for cost-utility analysis

A cost-utility analysis needs a utility for each health state, anchored at 1 for full health and 0 for dead, so that time can be weighted into QALYs. The PHQ-9 total has no such anchoring, and an equal-weighted sum says nothing about how the public values one symptom against another. A trial that collected only the PHQ-9 therefore needs a mapping function, a statistical model that predicts utility on a preference-based instrument from the PHQ-9 score and possibly other covariates.

Mukuria, Franklin and Hinde estimated mapping functions for the PHQ-9 and GAD-7 to the EQ-5D-3L using data from four depression trials of collaborative care or computerised cognitive behavioural therapy. Their estimation samples held 5,583 observations with both EQ-5D-3L and PHQ-9 and 3,942 with both EQ-5D-3L and GAD-7. They compared adjusted limited dependent variable mixture models (ALDVMM) with ordinary least squares and Tobit regression, found that the mixture models predicted better, and recommended a four-component ALDVMM for economic evaluations that hold PHQ-9 data, GAD-7 data or both.

The target instrument matters. Mukuria and colleagues note that utilities from the EQ-5D-3L and the EQ-5D-5L are not equivalent, so a mapping estimated to the three-level version and scored with a three-level value set does not give five-level utilities. Converting between the two EQ-5D versions is a separate step, usually called a crosswalk, and should not be confused with mapping from a clinical measure.

Worked example of mapped QALYs in a therapy trial (illustrative)

All numbers in this example are illustrative and the mapping equation is invented for teaching. It is a straight line, whereas the published functions above are mixture models. A trial randomises adults with depression to a course of therapy or to usual care and records the PHQ-9 at baseline and at 12 months. Both arms start at a mean PHQ-9 of 16. At 12 months the therapy arm scores 7 and the usual care arm scores 11. Therapy adds GBP 800 per person.

$$u = 0.92-0.025 \times s$$

where $u$ is the predicted utility and $s$ is the PHQ-9 total.

1. Utilities at each time point. At baseline, 0.025 × 16 = 0.4, so $u$ = 0.92 − 0.4 = 0.52 in both arms. At 12 months, 0.025 × 7 = 0.175 for therapy, giving 0.92 − 0.175 = 0.745, and 0.025 × 11 = 0.275 for usual care, giving 0.92 − 0.275 = 0.645.

2. QALYs over the year. Joining the two time points with a straight line gives the area under the curve.

$$Q = \frac{u_0 + u_1}{2} \times T$$

where $Q$ is QALYs, $u_0$ and $u_1$ are the utilities at baseline and 12 months and $T$ is the length of follow-up in years, here 1. For therapy, (0.52 + 0.745) ÷ 2 = 0.6325 QALYs, and for usual care, (0.52 + 0.645) ÷ 2 = 0.5825 QALYs. The QALY gain is 0.6325 − 0.5825 = 0.05.

3. Cost per QALY.

$$\text{ICER} = \frac{800}{0.05} = 16000$$

where 800 is the incremental cost in GBP and 0.05 is the incremental QALY gain, so the ICER is GBP 16,000 per QALY.

4. Sensitivity to the mapping slope. If a different mapping put the slope at 0.02, baseline utility would be 0.92 − 0.32 = 0.60, the 12-month utilities would be 0.92 − 0.14 = 0.78 for therapy and 0.92 − 0.22 = 0.70 for usual care, and the QALYs would be (0.60 + 0.78) ÷ 2 = 0.69 against (0.60 + 0.70) ÷ 2 = 0.65. The gain falls to 0.04 and the ratio rises to 800 ÷ 0.04 = 20,000, or GBP 20,000 per QALY.

5. Reading the same data as a service outcome. Under the NHS Talking Therapies thresholds, the therapy arm's mean falls by 9 points and ends below 10, which on the PHQ-9 alone meets the cut-off and reliable change conditions. The usual care arm falls by 5 points, short of the reliable change index of 6, and stays above the cut-off.

The same four PHQ-9 numbers give a quarter more cost per QALY under one plausible mapping than under another, while the clinical picture does not change. A sensitivity analysis over the mapping function, or direct EQ-5D data, shows how much of a decision rests on that choice. The mean-based reading in step 5 also hides the spread of individual results, since recovery is defined person by person.

Choices that change a PHQ-9 based estimate

Several analytic choices sit between a PHQ-9 dataset and a QALY figure. Each can move the result without any change in the trial data.

The first is the form of the mapping model. Mukuria and colleagues found that mixture models predicted EQ-5D-3L better than ordinary least squares or Tobit regression, so a model built on a weaker form uses less accurate utilities. The second is the target instrument and value set, since a mapping to the EQ-5D-3L carries its value set with it. The third is which scores enter the mapping, since the published functions predict utility from the PHQ-9, the GAD-7 or both. The fourth is timing, because the QALY depends on how often the PHQ-9 was recorded and how utility is assumed to move between visits. Finally, a model built on the proportion above a cut point discards information that the continuous score holds.

How the PHQ-9 differs from nearby measures

The PHQ-9 is a patient-reported outcome instrument, and depression symptoms are the construct it measures. A generic preference-based measure such as the EQ-5D asks about health-related quality of life across a few broad dimensions and comes with a value set, while the PHQ-9 asks only about depressive symptoms and has none. The GAD-7 is the parallel anxiety questionnaire from the same family of screeners, and NHS Talking Therapies uses it as the default anxiety partner to the PHQ-9 when judging recovery.

The reliable change index of 6 used in NHS Talking Therapies is a statistical threshold for change larger than measurement error. The reliable change index is a different idea from a minimal clinically important difference, which is anchored to what patients regard as a meaningful change, and the two should not be used interchangeably. When the PHQ-9 is used to find cases in an apparently healthy population, its sensitivity and specificity belong to screening economics, where false positives and missed cases carry their own costs.

Limitations and common misreadings of PHQ-9 results

Several misreadings in economic use come from treating the score as something it was not designed to be. Each can bias a model or its interpretation.

• A score of 10 or more is not a diagnosis. With 88% sensitivity and 88% specificity in the original sample, some people above the cut point do not have major depression and some below it do. The false positive rate does not depend on prevalence, but the share of positive results that are false rises as prevalence falls, because the positive predictive value falls.
• The total is not a utility. Rescaling the PHQ-9 to run from 0 to 1 does not make it preference-based, and QALYs built that way have no standing in a cost-utility analysis.
• Mapped utilities carry extra uncertainty. A predicted utility comes with prediction error and model choice on top of trial sampling error. A probabilistic sensitivity analysis that samples only the trial data understates the true uncertainty.
• Service recovery rates are not trial effects. Recovery and reliable recovery rates in NHS Talking Therapies describe people treated in routine care, without a comparison group, so they do not estimate a treatment effect.
• Equal item weights hide differences in burden. Two people with the same total can have different symptom profiles, and a preference-based measure may value them differently.

Sources

• Drummond MF, Sculpher MJ, Claxton K, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Health Care Programmes. 4th ed. Oxford: Oxford University Press; 2015.
• Kroenke K, Spitzer RL, Williams JBW. The PHQ-9: validity of a brief depression severity measure. Journal of General Internal Medicine. 2001;16(9):606-613. doi:10.1046/j.1525-1497.2001.016009606.x
• Mukuria C, Franklin M, Hinde S. Mapping functions for the PHQ-9 and GAD-7 to generate EQ-5D-3L for economic evaluation. European Journal of Health Economics. 2025;26(1):63-70. doi:10.1007/s10198-024-01692-0
• NHS England. NHS Talking Therapies for anxiety and depression manual. Version 7, first published June 2018, updated March 2024. Gateway reference 08101. https://www.england.nhs.uk/wp-content/uploads/2018/06/nhs-talking-therapies-manual-v7.1-updated.pdf
• NHS England. NHS Talking Therapies service standards. Web page, consulted October 2026. https://www.england.nhs.uk/mental-health/adults/nhs-talking-therapies/service-standards/
• Patient Health Questionnaire (PHQ) Screeners. PHQ-9 English form and screener download page (measures developed by Spitzer RL, Williams JBW, Kroenke K and colleagues with an educational grant from Pfizer Inc). Web page, consulted October 2026. https://www.phqscreeners.com/select-screener