Should the district screen 45,000 adults for diabetes? Test performance and programme decisions
The district health authority proposes population-based screening for type 2 diabetes for all residents aged 35–64 - about 45,000 people - using fasting plasma glucose at the locally validated cut-off. The local validation study reported sensitivity 50% and specificity 95%. The best available estimate of undiagnosed diabetes prevalence in this age group is 10%. HbA1c would cost roughly three times as much and is unavailable in most peripheral laboratories. A validated non-laboratory risk questionnaire (age, BMI, waist circumference, family history, physical activity) is available free from the national programme and could be applied by public health midwives; in this population it has sensitivity 90% and specificity 50%. The finance division asks whether the proposal is affordable; the clinical services division asks whether it will do more good than harm.
1. Calculate the positive predictive value and the number needed to screen for the one-step fasting plasma glucose strategy, and explain both to the finance division in plain language.
Read the worked answer
Calculation with a hypothetical cohort of 10,000.
- Diseased: 10% × 10,000 = 1,000. Detected by FPG: 1,000 × 0.50 = 500.
- Not diseased: 9,000. False positives: 9,000 × (1 − 0.95) = 9,000 × 0.05 = 450.
- Total positives: 950.
PPV = 500 / 950 = 52.6%. That is, nearly half of everyone told they have diabetes on the first test will not have it.
Number needed to screen = 1 / (prevalence × sensitivity) = 1 / (0.10 × 0.50) = 20. Twenty adults must be tested to find one true case.
Plain-language framing. 'If we test 10,000 people, we will spend the budget on about 950 positive results, of whom roughly 450 will be false alarms and 500 will be real cases. We will also miss about 500 real cases, because this test is only half-sensitive in our population.' The finance division should also see the downstream cost of 450 confirmatory workups and the reassurance given to 500 people with undiagnosed disease - the reassurance is the greater harm, because it delays diagnosis. The key point is that low PPV here is driven by prevalence and specificity, not by the laboratory.
2. Design a two-step strategy using the risk questionnaire first and the FPG second. Recalculate the yield, and state explicitly what trade-off you are accepting.
Read the worked answer
Step 1: questionnaire. With sensitivity 90% and specificity 50%, the prevalence of true disease among questionnaire-positives is:
(0.10 × 0.90) / [(0.10 × 0.90) + (0.90 × 0.50)] = 0.09 / 0.54 = 16.7%.
So 54% of the population is referred for a blood test, and the pre-test probability in that group has risen from 10% to 16.7%.
Step 2: FPG in questionnaire-positives. PPV = (0.167 × 0.50) / [(0.167 × 0.50) + (0.833 × 0.05)] = 0.0835 / 0.1252 = 66.7%.
Overall, the programme detects 0.90 × 0.50 = 45% of true cases, with overall specificity of about 97.5%.
The trade-off. PPV improves from 53% to 67%, and the number of blood tests falls by about half, but sensitivity falls from 50% to 45%. This is a classic yield-versus-reach trade-off. My judgement is that it is acceptable only if the questionnaire-negative group is not told they are disease-free; they need a clear message about symptoms and about re-assessment, and the risk score should be repeated at intervals. If the district's priority is equity of detection rather than laboratory efficiency, the one-step approach may be preferred so long as confirmatory testing is guaranteed.
Non-negotiable in either design: a positive FPG is not a diagnosis. It must be confirmed on a second occasion unless the patient is symptomatic, and the confirmatory pathway must be budgeted before screening starts.
3. Apply an accepted screening appraisal framework and give your recommendation, including the ethical obligations that attach to a positive result.
Read the worked answer
Appraisal against the standard screening criteria. Strong points: type 2 diabetes is an important condition; its natural history is reasonably understood; there is a detectable latent phase; the test is acceptable and the questionnaire is cheap; treatment is effective. Weak points, in this district: confirmatory capacity in peripheral laboratories; whether diagnosed patients can be retained in care and get medication, eye and foot checks; whether the programme will be continuous rather than a one-off campaign; and whether the information system can recall people.
The test does not satisfy the criteria on its own. Sensitivity of 50% with a PPV of 53% in the highest-prevalence age band is a weak screening instrument. A two-step design is better but still misses over half the cases.
Recommendation. Proceed only as an organised, two-step, continuously audited programme with: confirmatory testing before labelling; a named treatment and follow-up pathway with drug supply; a register with recall; monitoring of PPV, yield, retention at 12 months and equity by sector; and a clear decision point to stop or redesign if the yield or retention is poor. Do not run a one-off 'campaign' with no follow-up - that converts a data problem into a clinical one.
Ethical obligations. Consent for screening is not the same as consent for treatment; people must understand the false-positive and false-negative rates in plain language. There is a duty to recall and investigate every positive result, to communicate negatives without false reassurance, and to protect against labelling, out-of-pocket costs and loss of confidentiality in small communities. Screening must not be introduced where the treatment it promises is unavailable - that is the strongest argument for a phased pilot in two MOH areas before district-wide rollout.