Depression Screener Cutoff and Predictive Values
A community clinic is considering a new 10-item self-report screener for major depression. In a validation sample of 500 adult primary care patients, 20% meet structured-interview criteria for major depression. At the proposed cutoff of 8, the screener has sensitivity of .80 and specificity of .75. The clinic administrator wants to know how many positive screens would be 'true' cases and whether the cutoff should be lowered to 6, which would raise sensitivity to .90 but lower specificity to .60. The clinic has limited staffing for follow-up diagnostic interviews and is concerned about alarming patients with false positives. Most patients have low income and many are Spanish-speaking; the validation sample was 80% non-Hispanic White and 60% college educated. The clinic plans to use the screener with all adults, including those already in treatment.
1. Calculate and interpret the positive and negative predictive values at cutoff 8, and explain what they mean for the clinic.
Read the worked answer
With prevalence 20%, in 500 patients: 100 true cases, 400 non-cases. Sensitivity .80 detects 80 of 100 cases; 20 are false negatives. Specificity .75 correctly identifies 300 of 400 non-cases; 100 are false positives. PPV = 80/(80+100)=44.4%. NPV = 300/(300+20)=93.75%. So among patients who screen positive, only about 44% actually have major depression; among those who screen negative, about 94% do not. The PPV is modest because the base rate is 20%, even though sensitivity and specificity look acceptable. For the clinic, a positive screen is a trigger for a diagnostic interview, not a diagnosis. With limited follow-up capacity, 180 positive screens would need evaluation, and 100 would be false positives. The NPV means a negative screen is fairly reassuring but not perfect. The judgment is to avoid overinterpreting a positive screen and to ensure a confirmatory assessment.
2. Using decision theory, should the clinic lower the cutoff to 6? What values and practical constraints should guide the choice?
Read the worked answer
At cutoff 6: sensitivity .90, specificity .60. Of 100 cases, 90 are detected; 10 are missed. Of 400 non-cases, 160 are false positives; 240 are correctly negative. PPV = 90/250 = 36%; NPV = 240/250 = 96%. Lowering the cutoff detects 10 more depressed patients but adds 60 false positives and increases the number needing follow-up from 180 to 250. The choice depends on the relative cost of a missed depression versus a false alarm, the clinic's capacity for diagnostic interviews, and whether effective treatment is available. If untreated depression is costly and follow-up can be staffed, the lower cutoff may be justified, but only with a two-stage strategy: brief screen, then diagnostic interview. If follow-up is scarce, a higher cutoff or a more specific tool is preferable. The clinic should not adopt cutoff 6 solely because sensitivity is higher; it must weigh PPV, workload, and patient burden.
3. What validity and fairness concerns arise from applying this screener to the clinic's Spanish-speaking and low-income patients?
Read the worked answer
The validation sample was mostly non-Hispanic White and college educated, so measurement invariance is unknown. A Spanish translation may not be linguistically or culturally equivalent; idioms of distress, literacy, and response styles can differ. Low income and education can affect item comprehension and scores independent of depression. Using English-derived cutoffs may misclassify Spanish-speaking patients, producing false positives or false negatives. Fairness requires translation and back-translation, cognitive interviewing, and validation in the target subgroups, including estimates of sensitivity, specificity, and PPV by language and education. The clinic should also examine whether the construct of depression is measured equivalently and whether access barriers, not only screening, drive outcomes. Until then, results should be interpreted cautiously and confirmed clinically. The judgment is that a screener validated in one population cannot be assumed to perform the same way in another.