Educational bias in the Montreal Cognitive Assessment among hemodialysis patients: A domain-level analysis

Cognitive impairment is common in hemodialysis (HD) patients and increasingly informs clinical decisions, including capacity to consent and the planning of dialysis care. Most estimates rely on the Montreal Cognitive Assessment (MoCA), an instrument strongly dependent on educational attainment. In low-literacy HD populations, this dependence may cause systematic over-identification of impairment and distort apparent risk factors such as sex. We examined how educational attainment shapes MoCA-defined impairment in older HD patients and used the sex difference as a natural experiment to expose the bias.

This cross-sectional study enrolled 161 patients aged 50 years or older receiving maintenance HD at three Western-region National Guard centers in Saudi Arabia. Impairment was defined as MoCA < 26. We compared mean MoCA and impairment across educational strata; tested for education-related differential item functioning (DIF) across MoCA domains after matching on overall ability; quantified the apparent sex effect by sequential adjustment, formal mediation and within-stratum comparison; estimated the share of the educational score gap recoverable by the standard one-point correction; and tabulated impairment across screening thresholds. Independent determinants were identified by multivariable logistic regression.

Impairment affected 107/161 patients (66.5%). Mean MoCA was 17.6 in illiterate patients, 24.1 in those who read/write and 25.5 in college-educated patients (illiterate-college difference 7.9 points). The standard one-point correction offset 13% of this difference; 81% of illiterate patients remained below threshold even after a three-point correction. Education-related DIF was present in the four literacy-loaded domains (attention, executive, orientation, language; all p ≤ 0.006) and absent in the three least literacy-dependent domains (delayed recall, abstraction, naming; all p > 0.3), after matching on overall ability. Women had a higher crude impairment prevalence than men (78.8% vs 57.9%; p = 0.010; crude odds ratio 2.70); this fell to 1.61 (95% CI 0.73-3.53) after adjustment for age and education; education mediated 62% of the sex effect on MoCA score, and the sex difference was non-significant within every education stratum. In sensitivity analyses using lower cutoffs, a residual female association became apparent (adjusted odds ratio 2.66 at < 24 and 2.48 at < 23). In multivariable analysis only older age and lower education were independently associated with impairment (area under the curve 0.76-0.78); no dialysis-related or biochemical variable, including iPTH, was associated.

In this low-literacy HD population, MoCA-defined cognitive impairment was governed by educational attainment, with domain-level evidence of literacy bias, and the apparent female excess was largely explained by educational attainment rather than by sex, although a residual female association persisted at lower cutoffs, consistent with incomplete adjustment for literacy. Because cognitive screening informs consent and care decisions, education-blind use of the MoCA in such settings risks systematic misclassification. Education-aware screening-education-specific norms or cutoffs, attention to literacy-neutral domains, and confirmatory assessment-is preferable to raw thresholds.