Pilot protocol

Design

Prospective, paired diagnostic accuracy study: every person does both the Thabat check and the MoCA at the same visit.

Setting

Two IFHAS Memory Testing centres, one in Abu Dhabi city and one in Al Ain.

Who takes part

Adults aged 50 to 75 attending IFHAS Memory Testing who consent. Not included: a recorded dementia diagnosis, acute illness on the day, or a check that cannot be done even with help.

Index test

The Thabat check (four activities), on the person's phone, with a helper or at the kiosk. Its result is held from the person and the MoCA rater until the visit ends.

Order

Randomised 1:1: Thabat first or MoCA first, so practice and tiredness balance out.

Low schooling

People with 0–6 years of schooling also do the RUDAS, and an informant completes the GPCOG, as the FCGA guideline names for low education.

Dialect and literacy

Before enrolment, cognitive interviews with 24 people across Gulf, Levantine, Egyptian and Sudanese Arabic and low literacy. Any item misread by 2 or more is reworded and retested.

Primary

The same-visit IFHAS MoCA, given by an IFHAS nurse who does not see the Thabat result, read against the education-adjusted cut-off. The MoCA is the Department of Health and Public Health Centre mandated tool, so Thabat is measured against it, not instead of it.

Clinical

Specialist assessment in neurology or geriatrics, as the IFHAS pathway already sends people below the MoCA cut-off.

Who is verified

Everyone flagged by Thabat or below the MoCA cut-off, plus a random 10% of everyone else, drawn from a list fixed before the first enrolment.

Bias correction

Begg–Greenes weighting corrects sensitivity and specificity for the partly verified sample.

Worked from Buderer (1996): size the people below and above the cut-off for the precision wanted, then divide by the share expected in each group. A low share below the cut-off drives enrolment.

Assumptions

Expected sensitivity 85% (±7%), specificity 85% (±5%), 12% below the MoCA cut-off, 15% lost by 12 months, 95% confidence.

Sensitivity

1.96² × 0.85 × 0.15 / 0.07² = 100 below the cut-off → 100 / 0.12 = 834

Specificity

1.96² × 0.85 × 0.15 / 0.05² = 196 at or above → 196 / 0.88 = 223

Enrolment

max(834, 223) / 0.85 → 1,000 (500 per centre)

Subgroups

Reading sensitivity within ±10 points in one level needs 49 people below the cut-off. Levels with fewer than 30 are reported with their denominator and marked, not read.

The 12% share is an assumption. The first three months of IFHAS Memory Testing data replace it before the final sizing goes to the committee.

• DoH Research Ethics Committee approval and registration on ClinicalTrials.gov before the first person is enrolled.
• Accuracy reported to STARD 2015. The early clinical use of the AI parts reported to DECIDE-AI.
• The six WHO principles for AI in health are mapped in the trust centre. RAI & standards
• An independent safety group (a family physician, a geriatrician and a biostatistician) reviews at 250 and 500 enrolled.
• Every flagged result is reviewed by a clinician before the person sees it. Data stays in the UAE under ADHICS.

Every endpoint is reported for every level, with its denominator. Current subgroup results

1. Pause enrolment if, at the 250-person interim, the lower 95% bound of sensitivity is below 60%.
2. Pause if sensitivity or specificity in any readable subgroup level is 15 points or more below the overall figure.
3. Pause flag release if false referrals pass 40 per 1,000 at any interim. Checks continue, results go to the clinician only.
4. Any serious event linked to disclosure goes to the safety group within 72 hours. Enrolment pauses until it rules.