Article Text
PDF
Statistics from Altmetric.com
Mitchell AJ, Coyne JC. Do ultra-short screening instruments accurately detect depression in primary care? A pooled analysis and meta-analysis of 22 studies. Br J Gen Pract 2007;57:144–51.
METHODS
Design:
Systematic review of diagnostic studies with meta-analysis.
Data sources:
MEDLINE, PsycINFO, EMBASE, CINAHL, Science Direct, Ingenta Select, Ovid Full text, Wiley Interscience, Web of Knowledge; search date June 2006.
Study selection and analysis:
Studies of ultra-short depression screening tests (one, two or three questions) for use in primary care from published and non-peer reviewed sources. Exclusions: studies in medical patients, secondary care or nursing homes; studies of visual analogue scales or short tools (5–14 items); studies not Standards for Reporting of Diagnostic Accuracy (STARD) compliant. The meta-analysis involved using raw data to calculate the ratio of the proportion of true diagnoses to the proportion of false diagnoses. Where the ratio was significantly greater than 1, data were pooled to determine sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV).
Outcomes:
Ratio of true cases to false cases, sensitivity, specificity, NPV and PPV for each ultra-short test.
MAIN RESULTS
Twenty two analyses (12 publications) met inclusion criteria. Ultra-short depression screening instruments were significantly better at detecting depression in primary care than expected by chance (ratio of true to false diagnoses 5.5, 95% CI 5.3 to 5.6); there was significant heterogeneity in this analysis (p<0.001; see Notes). Pooled analysis of studies examining single-question tests (8 analyses) found that overall sensitivity was 31.9% and specificity 96%. Pooled analysis of studies of two- or three-question test instruments (14 analyses) found that overall sensitivity was 73.7% and specificity 74.7%. PPV was 55.6% for single-question tests (where prevalence of depression was 10.7%) and 38.3% for two- and three-question tests (prevalence of depression 17.6%). NPV was 92.3% for single-question tests and 93% for two- and three-question tests.
CONCLUSIONS
Ultra-short two- and three-question screening tests are more sensitive than ultra-short one-question screening tests. High false positive rates suggest that ultra-short tools should only be used if there are resources available for further assessment of those who screen positive.
NOTES
The authors used a STARD checklist to assess quality of diagnostic studies, but did not specifically report on which aspects of quality they used to determine inclusion in or exclusion from the review. Four studies were excluded from meta-analysis for being “non-STARD compliant for reasons of incomplete data or inadequate sample size”. There was statistically significant heterogeneity between studies in the overall meta-analysis; meta-analysis may therefore not have been appropriate in this case. PPV and NPV values cannot be directly compared between different types of test as the baseline prevalence of depression was different in the single-question and two- or three-question populations. No significant tests were performed to compare outcomes between the different test types.
Commentary
Screening for depression in primary care was recommended by the US preventive services task force,1 as was screening of high risk patients by NICE in the UK.2 However, these policies have been questioned, in part because the available screening tests may not fulfil the required criteria of precision and acceptability.3
This study has very helpfully pooled the results from screening studies using ultra-short measures (up to three questions), and analysed the performance of single-question tests versus two- or three-item tests. The bottom line is that two- or three-item tests perform better than single items in terms of overall sensitivity (74% vs 32%) but less well in terms of specificity (75% vs 97%). The relatively low prevalence of depression in primary care (around 10%) means however that the positive predictive value (PPV) of even very sensitive and specific instruments will be low.3 In this review, the PPVs were 56% for one item tests and only 38% for two to three item tests. On the other hand, the overall negative predictive values (NPVs) were around 93%, which means a negative result can be taken to indicate the patient does not have depression.
The low PPVs are a problem for those advocating screening, as 6 out of 10 patients positive on a two- or three-item test will turn out to be negative on further assessment, creating extra work for the clinicians and possibly anxiety for the patients. It is essential that resources are in place for further assessment of patients who screen positive, and for effective treatment, which unfortunately is often not the case.4 At present in the UK general practitioners are rewarded through the performance-related Quality and Outcomes Framework (QOF) for screening patients with diabetes mellitus and coronary heart disease for depression. The PPV may be better in these selected populations, where the prevalence may be 2–3 times higher;5 ,6 but this remains to be demonstrated. The main value of ultra-short screening tests for depression is in helping clinicians rule out depression.
Footnotes
Sources of funding: reported as “not applicable”.
Footnotes
Competing interests: Professor Kendrick is Chair of the Mental Health Expert Reference Group for the UK GP Contract Quality and Outcomes Framework, whose responsibility it is to review the mental health indicator for screening for depression among diabetes and heart disease patients in UK primary care.