Grader discrepancy, an ungradable non-attender and a research request
An internal audit of 300 consecutive diabetic retina screening episodes compares each grader's decision with arbitration by the programme's clinical lead. Grader A agrees with arbitration in 94% of episodes; Grader B, who graded 640 episodes last quarter, agrees in 81%. Twelve episodes graded by B as 'no retinopathy' were graded by arbitration as referable retinopathy, six of them with maculopathy. The programme screens 130,000 people a year with 70% attendance and a 12-month recall. Separately, a 58-year-old man with two previous 'ungradable' image sets has not attended three repeat appointments; his GP telephones asking for 'whatever result you have'. A university researcher also requests access to 5,000 anonymised images for algorithm development, offering programme staff co-authorship, and wants an answer within two weeks because funding ends in March.
1. How would you manage the 12 misgraded episodes and the remaining 628 graded by Grader B?
Read the worked answer
The 12 come first and they are urgent. Referable retinopathy, particularly maculopathy, is sight-threatening if assessment is delayed. The programme's duty is to get those people seen, not to wait for the next annual recall: the clinical lead contacts ophthalmology to arrange review, and each participant receives a corrected result with an explanation and a named contact. Record this as a screening safety incident if local policy requires it.
The remaining 628 cannot be handled by sampling. An 81% agreement rate is not random noise; it signals a systematic problem with one grader's practice. If you sample and find two more, you have learned nothing about the other 626. The defensible options are:
- Re-grade 100% of B's episodes with a second grader and arbitration of disagreements; or
- If capacity genuinely prevents that, re-grade in a documented risk order (any microaneurysm or haemorrhage noted, long diabetes duration, known maculopathy, poor image quality) with a written commitment and timetable to complete the remainder.
Then ask what caused it. Was B certified and when? Has there been a competence review? Is there a second-read policy for borderline images? Is image quality or camera calibration poor, or is the time allowed per episode too short? Is this one person or one process?
Manage the person fairly. Restrict independent sign-off pending a competence assessment, but do not name or humiliate B; a just-culture approach is what makes the next grader report their own discrepancy instead of hiding it. Notify the programme's clinical governance committee and keep a decision log.
2. The 58-year-old man has missed three repeat appointments and his GP wants 'whatever result you have'. How do you handle this?
Read the worked answer
The result is 'screening not completed', not 'normal'. The greatest risk in this scenario is everyone assuming that silence means a clear result - the GP, the man, and the record.
- Speak to the GP the same day by phone. Explain that three appointments were missed and no grading exists. Share only what is necessary: the fact of non-attendance, the need for screening, and that he should be encouraged to attend. Do not send a 'normal' result in any form, written or verbal.
- Contact the man directly, not only through the GP. Offer a specific appointment rather than an open invitation, and ask what has got in the way - travel, work, caring responsibilities, fear, a previous bad experience, language or literacy. Offer an alternative site, a different time, or transport support if the programme provides it.
- Ask why the images were ungradable twice. Small pupils, cataract, poor camera positioning or tear film problems are correctable; a third identical attempt will likely fail again. Flag the record so the photographer uses a different protocol.
- Set a clear deadline and escalate to the clinical lead if he continues not to attend; consider whether a referral for a dilated fundus examination outside screening is more appropriate, given two ungradable sets.
- If he declines, record informed refusal and tell the GP, so the decision is documented rather than drifting.
- Check the wider pattern - three missed appointments suggest the recall process may not be reaching people. Audit missed repeat appointments.
3. How would you respond to the researcher's request for 5,000 images?
Read the worked answer
The co-authorship offer is a conflict of interest and should be declared and declined as a basis for the decision. Images collected for screening are held for a defined purpose; research is a secondary use and needs its own governance.
- Do not agree informally or under time pressure. A March funding deadline is the researcher's problem, not a reason to bypass approval.
- Check identifiability. Retinal images plus metadata (date, site, diabetes type, age) can re-identify; 'anonymised' is a claim to test, not accept. Pseudonymisation with a controlled key and a data-sharing agreement is usually more honest.
- Requirements: a data protection impact assessment, research ethics committee approval, the declarations required under the Health Research Regulations, a data-sharing agreement covering security, retention, destruction and onward sharing, and a justified sample size.
- Check what participants were told. If screening information did not mention research use, a transparency notice and, where required, consent are needed.
- Ask what comes back. A useful arrangement includes a commitment to share results and any validated algorithm with the programme, and an agreed publication route.
- Take it to the programme board with the data protection officer's view, and give the researcher a realistic timeline.
Refusing until governance is in place is defensible. Releasing identifiable health data because a deadline was tight is not, and the reputational and legal cost outweighs any benefit from co-authorship.