Attention deficits as a target for AI-based screening of cognitive impairment after myocardial infarction

Background

Both cardiovascular diseases and cognitive impairment (CI) are major public health problems in ageing societies. Growing evidence indicates shared mechanisms between ischaemic heart disease, including myocardial infarction (MI), and CI; however, cognitive assessment is not routinely implemented in cardiology care.

Purpose

To evaluate the prevalence and dynamics of CI in patients hospitalised for MI and to develop a prototype AI-based digital screening tool to classify CI after MI, focusing on attention- the cognitive function most frequently impaired in this patient population.

Methods

A total of 468 patients with MI treated with percutaneous coronary intervention were prospectively assessed during hospitalisation (days 2–3) and at 6-month follow-up. Cognitive function was evaluated using the Mini-Mental State Examination (MMSE) and the Clock Drawing Test (CDT). CI was defined as MMSE <27 points and/or ≥1 CDT error. Clinical and psychological data were collected. Based on these data, an AI-assisted screening tool was developed using automated analysis of CDT drawings combined with selected MMSE items assessing attention and recall.

Results

Previously undiagnosed CI was identified in 37% of patients during hospitalisation and in 25% at follow-up. CI both during MI and at 6 months was present in 8.9–12.5% of patients, CI limited to the peri-infarction period in 16.3–18.1%, and CI emerging at follow-up in 7.7–9.8%, depending on the test used. Attention deficits were present in all cases of CI and closely correlated with changes in global cognitive status; improvement or deterioration in cognition during follow-up was accompanied by corresponding changes in attention in 81% and 100% of patients, respectively.

Compared with patients without CI, those with CI were older (60.4 ± 10.2 vs. 58.3 ± 9.3 years; p = 0.03) and demonstrated more severe myocardial injury, reflected by higher B-type natriuretic peptide levels (p = 0.01) and lower left ventricular ejection fraction (p = 0.05). They also had more advanced coronary atherosclerosis and higher LDL cholesterol levels (both p = 0.05). Depressive symptoms and sleep disturbances were more pronounced in patients with CI.

These findings formed the basis for development of a prototype AI-based screening tool aimed at early identification of CI in patients after MI. The tool integrated two CDT tasks (free placement of numbers on the clock face and setting the clock hands to a specific time, 11:10) with MMSE attention and recall items. The combined feature set achieved the highest performance, with an accuracy of 0.93, precision of 0.97, and AUROC of 0.95.

Conclusion

CI is common after MI and often remains unrecognised. Attention deficits represent the predominant cognitive domain affected. An AI-based screening approach focused on attention may support early identification of post-MI patients with CI in cardiology practice while minimising the burden on medical staff.Proposed AI-based screening tool.