EPDS

EPDS in Perinatal Depression Screening and Economic Evaluation

The cut-off chosen on the Edinburgh Postnatal Depression Scale decides how many women an economic model of perinatal depression case-finding sends on for clinical assessment, and so drives much of the model's cost. Cox, Holden and Sagovsky validated the scale in 84 mothers against Research Diagnostic Criteria, reporting in 1987 that it could be completed in about five minutes and had a simple method of scoring. For a health economist the EPDS matters in two places. Its accuracy at that threshold shapes the QALYs as well as the costs of a screening model, and its total has to be converted before it can feed a cost-utility analysis. A cost-effectiveness analysis published in the BMJ in 2009 found that formal identification of postnatal depression, the EPDS included, did not seem to represent value for money for the NHS, and its authors named the potential cost of managing women incorrectly diagnosed as depressed as the major determinant.

Items, recall period and scoring of the EPDS

The EPDS is a short self-report patient-reported outcome measure. Each of its ten items asks how the woman has felt, and the EPDS manual by Cox, Holden and Henshaw tells the administrator to ask her to underline the response that comes closest to how she has felt during the previous 7 days.

Every item has four responses. The manual states that items are scored 0, 1, 2 or 3, with the normal response scoring 0 and the severe response scoring 3, so the total runs between 0 and 30. Cox described the scale in a 2019 editorial as a ten-item self-report questionnaire to facilitate the detection of perinatal depression, translated into over 60 languages.

Scoring is simple in principle but errors are common in practice. The manual cites a study in which between 13.4% and 28.9% of completed scales contained errors, which matters when a trial uses the total as its primary outcome or when a fixed cut-off triggers referral.

The authors reported in 1987 that the EPDS was sensitive to change in the severity of depression over time, which allows the total to serve as a continuous trial outcome as well as a screen.

Cut-offs from the 1987 validation and the DEPRESSD meta-analysis

A screening cut-off trades sensitivity against specificity, and the EPDS has been used at many different thresholds. Two sources dominate the choice.

In the original validation the reference standard was Goldberg's Standardised Psychiatric Interview with Research Diagnostic Criteria. Looking back in 2019, Cox wrote that the 1987 study found a cut-off of 12/13 gave optimal sensitivity of 86% and specificity of 78% for major and minor depression. That estimate came from a sample of 84 mothers.

The DEPRESSD group led by Levis pooled far more data in an individual patient data meta-analysis published in 2020. Data came from 58 of 83 eligible studies, covering 15,557 pregnant and postpartum women, 2,069 of whom had major depression. Combined sensitivity and specificity was maximised at a cut-off of 11 or higher across reference standards, with Youden's J of 0.70 for semi-structured interviews, 0.73 for fully structured interviews and 0.66 for the MINI.

The semi-structured interview results rest on 36 studies with 9,066 participants and 1,330 cases of major depression. Levis and colleagues reported that a cut-off of 13 or higher was less sensitive but more specific than 11, and suggested that 13 or higher could be used to identify women with higher symptom levels. The target condition in the meta-analysis is major depression, whereas the 1987 threshold was chosen for major and minor depression together, so the two recommendations answer different questions.

Use in pregnancy and the self-harm item

The name says postnatal, but the scale is used throughout the perinatal period. Murray and Cox gave the EPDS to 100 women attending an antenatal clinic at 28 to 34 weeks of gestation and then interviewed them with a standardised psychiatric interview, reporting in 1990 that the scale identified all women with major depression by Research Diagnostic Criteria but was less effective in detecting minor depression. The title of their paper used the name Edinburgh Depression Scale. The DEPRESSD meta-analysis pooled pregnant and postpartum women in one analysis.

Item 10 asks about thoughts of self-harm. Qiu and colleagues, also from the DEPRESSD group, noted in 2023 that the item is intended to assess thoughts of intentional self-harm but may also elicit concerns about accidental self-harm, and they tested a nine-item version without it. In 41 studies with 10,906 participants and 1,407 cases of major depression, totals on the nine-item version and the full scale correlated at 0.998. Sensitivity was equivalent at cut-offs of 7 to 12, equivalence was indeterminate at 13 to 15, and specificity was equivalent at every cut-off. A response on item 10 is a clinical signal in its own right, and a protocol that scores the item has to say what happens next when a woman endorses it.

EPDS in the NICE antenatal and postnatal mental health guideline

NICE guideline CG192, Antenatal and postnatal mental health, first published in 2014 and updated in 2020, places the EPDS within its recommendations on recognising mental health problems. Brief depression identification questions come first, and the EPDS is used for assessment and monitoring afterwards.

Recommendation 1.5.3 asks practitioners to recognise that anxiety disorders and depression are under-recognised throughout pregnancy and the postnatal period. Recommendation 1.5.4 says to consider asking the depression identification questions, about feeling down, depressed or hopeless and about little interest or pleasure, at a woman's first contact with primary care or her booking visit and in the early postnatal period, together with the GAD-2 questions on anxiety.

Recommendation 1.5.5 then says that when a woman responds positively, is at risk or raises clinical concern, practitioners should consider using the EPDS or the PHQ-9 as part of a full assessment, or referring her to her GP or, if a severe mental health problem is suspected, to a mental health professional. Recommendation 1.5.8, on later contacts, includes using the EPDS or the PHQ-9 as part of monitoring. NICE therefore places the EPDS as an assessment and monitoring tool behind the brief questions, and a model of NICE-concordant care has a two-stage structure.

The EPDS in economic evaluations of perinatal depression screening

In a screening model, often built as a decision tree, the EPDS cut-off sets the share of true and false positives. The cost of following up each positive screen, and the health effect of treating the true cases, then decides the ICER. Two UK analyses show how sensitive the answer is to those inputs.

Paulden, Palmer, Hewitt and Gilbody compared standardised postnatal depression instruments and generic depression instruments with routine care in a decision model published in 2009. The EPDS at a cut-point of 16 had an ICER of GBP 41,103 per QALY against routine care alone, and the other strategies ranged between GBP 49,928 and GBP 272,463 per QALY. The probability that no formal identification strategy was cost effective was 88% at a threshold of GBP 20,000 per QALY and 59% at GBP 30,000. The authors identified the potential additional cost of managing women incorrectly diagnosed as depressed as the major determinant of cost effectiveness.

Heslin and colleagues reported in 2022 a decision tree for screening in early pregnancy, taking an NHS and personal social services perspective and running between the first antenatal appointment and 3 months after birth. It compared the Whooley questions, the EPDS and the Whooley questions followed by the EPDS with no screening, drawing on a cross-sectional survey of diagnostic accuracy. Each screening option had a probability of being cost-effective of around 30% at GBP 20,000 to GBP 30,000 per QALY, against around 20% for no screening. The EPDS cost more than the other strategies but produced more QALYs. The wider methods of screening economics apply, and the EPDS-specific inputs are its accuracy at the chosen cut-off and the cost of its follow-up assessment.

Turning EPDS scores into QALYs

The EPDS total is an equal-weighted symptom score, not a utility anchored at 1 for full health and 0 for dead, so the total cannot be multiplied by time to give QALYs. A trial reporting only the EPDS supports cost per point of change or cost per case detected, ratios that cannot be judged against a cost per QALY threshold.

A mapping function fills the gap when the EQ-5D was not collected. Camacho and colleagues used 1,068 paired EPDS and EQ-5D-3L observations from the BaBY PaNDA cohort in England to estimate a direct mapping, published online in 2023. They tested six model types with four covariate specifications, found a moderate correlation of minus 0.42 between EPDS score and utility, and reported that a two-part model performed best, followed by ordinary least squares. The algorithm fitted better when utility was above 0.6 and was tested in three external datasets, one of them the RESPOND study of women with diagnosed postnatal depression. The predicted values are EQ-5D-3L utilities and suit a cost-utility analysis that needs three-level values.

Worked example of EPDS cut-offs in a screening model (illustrative)

All numbers in this example are illustrative except the sensitivity and specificity, which are the DEPRESSD estimates against semi-structured interviews. A service screens 1,000 women in the postnatal period, the prevalence of major depression is 10%, and each positive screen leads to a clinical assessment costing GBP 150.

1. Cut-off of 11 or higher. Of 100 women with depression, 100 × 0.81 = 81 screen positive and 19 are missed. Of 900 without depression, 900 × 0.88 = 792 screen negative, so 900 − 792 = 108 are false positives. Positive screens total 81 + 108 = 189, and assessments cost 189 × 150 = 28,350, or GBP 28,350.

2. Cut-off of 10 or higher. True positives are 100 × 0.85 = 85. True negatives are 900 × 0.84 = 756, so false positives are 900 − 756 = 144. Positive screens total 85 + 144 = 229, and assessments cost 229 × 150 = 34,350, or GBP 34,350.

3. Positive predictive value. The positive predictive value at the higher cut-off follows from the formula below.

$$\text{PPV} = \frac{Se \times p}{Se \times p + (1-Sp)(1-p)} = \frac{0.81 \times 0.10}{0.81 \times 0.10 + 0.12 \times 0.90} = \frac{0.081}{0.189}$$

where $Se$ is sensitivity, $Sp$ is specificity and $p$ is the prevalence of depression in the screened group. The result is about 0.43, so 81 of 189 positive screens are true cases. At the lower cut-off the value is 85 ÷ 229, about 0.37.

4. Incremental cost of the lower cut-off. Moving to 10 or higher finds 85 − 81 = 4 extra cases and adds 144 − 108 = 36 false positives. Assessment costs rise by 34,350 − 28,350 = 6,000, which is GBP 1,500 for each extra case found before any treatment cost or QALY gain is counted.

Most women flagged at either cut-off do not have major depression, so the model result depends heavily on what a false positive costs. That is the same driver Paulden and colleagues identified, and a one-point change in threshold moves the follow-up budget by about a fifth.

How the EPDS differs from the PHQ-9 and nearby concepts

CG192 names two instruments for a full assessment. The PHQ-9 is the alternative to the EPDS, with nine items scored over a longer recall period, so its totals and cut-offs are not interchangeable with those of the EPDS. The GAD-7 measures anxiety, which the EPDS does not set out to measure. The two depression identification questions, which Heslin and colleagues call the Whooley questions, come before the EPDS in the NICE pathway.

Depression symptoms are what the EPDS score describes, while perinatal epidemiology studies how often depression occurs in pregnancy and after birth and supplies the prevalence that a case-finding model combines with EPDS accuracy. Within mental health economics the EPDS defines who counts as a case and tracks change in symptoms, and QALYs need a preference-based measure or a mapping.

Limitations and common misreadings

Several errors recur when the EPDS enters an economic analysis. Each changes either the number of women counted as depressed or the value placed on their improvement.

• One threshold for all purposes. The 12/13 figure from the 1987 validation targets major and minor depression in a small sample, while the DEPRESSD threshold of 11 targets major depression in pooled data. A model should name the cut-off and the target condition behind its accuracy inputs.
• Reading a positive screen as a diagnosis. At a prevalence of 10%, as in the worked example, most positive screens are false positives, and treating them as cases inflates both costs and the apparent burden of depression.
• Ignoring item 10 handling. Qiu and colleagues found the nine-item version equivalent in sensitivity at cut-offs of 7 to 12, but dropping the item removes a question about self-harm that a service may need for safety.
• Using the score as a utility. A rescaled EPDS total is not a QALY weight. A mapped EQ-5D-3L value is an estimate with its own error, and Camacho and colleagues found the fit weaker at lower utilities.
• Scoring errors. Error rates of between 13.4% and 28.9% in completed scales, cited in the EPDS manual, can move women across a cut-off and bias a screening yield.

Sources

• Camacho EM, Shields GE, Chew-Graham CA, et al. Generating EQ-5D-3L health utility scores from the Edinburgh Postnatal Depression Scale: a perinatal mapping study. European Journal of Health Economics. 2024;25:319-332. doi:10.1007/s10198-023-01589-4
• Cox J. Thirty years with the Edinburgh Postnatal Depression Scale: voices from the past and recommendations for the future. British Journal of Psychiatry. 2019;214(3):127-129. doi:10.1192/bjp.2018.245
• Cox J, Holden J, Henshaw C. Perinatal Mental Health: The Edinburgh Postnatal Depression Scale (EPDS) Manual. 2nd ed. London: RCPsych Publications; 2014. Chapter 7, Using the Edinburgh Postnatal Depression Scale. https://www.cambridge.org/core/books/abs/perinatal-mental-health/using-the-edinburgh-postnatal-depression-scale/745AC0330C33D48D269DB45437569A7E
• Cox JL, Holden JM, Sagovsky R. Detection of postnatal depression. Development of the 10-item Edinburgh Postnatal Depression Scale. British Journal of Psychiatry. 1987;150:782-786. doi:10.1192/bjp.150.6.782
• Heslin M, Jin H, Trevillion K, et al. Cost-effectiveness of screening tools for identifying depression in early pregnancy: a decision tree model. BMC Health Services Research. 2022;22:774. doi:10.1186/s12913-022-08115-x
• Levis B, Negeri Z, Sun Y, Benedetti A, Thombs BD; DEPRESsion Screening Data (DEPRESSD) EPDS Group. Accuracy of the Edinburgh Postnatal Depression Scale (EPDS) for screening to detect major depression among pregnant and postpartum women: systematic review and meta-analysis of individual participant data. BMJ. 2020;371:m4022. doi:10.1136/bmj.m4022
• Murray D, Cox JL. Screening for depression during pregnancy with the Edinburgh Depression Scale (EDDS). Journal of Reproductive and Infant Psychology. 1990;8(2):99-107. doi:10.1080/02646839008403615
• National Institute for Health and Care Excellence. Antenatal and postnatal mental health: clinical management and service guidance (CG192). Published 2014, updated 2020. Recommendations 1.5.3 to 1.5.8. https://www.nice.org.uk/guidance/cg192/chapter/recommendations
• Paulden M, Palmer S, Hewitt C, Gilbody S. Screening for postnatal depression in primary care: cost effectiveness analysis. BMJ. 2009;339:b5203. doi:10.1136/bmj.b5203
• Qiu X, Wu Y, Sun Y, Levis B, et al. Individual participant data meta-analysis to compare EPDS accuracy to detect major depression with and without the self-harm item. Scientific Reports. 2023;13:4026. doi:10.1038/s41598-023-29114-w