The Integrated Cognitive Assessment showed convergent validity with both MoCA and ACE and offers potential for remote measurement of cognitive performance in Alzheimer disease and mild cognitive impairment.
Dag Aarsland, MD, PhD
Early diagnosis of patients with Alzheimer disease (AD), as well as mild cognitive impairment (MCI), may be detected via the Integrated Cognitive Assessment (ICA), a 5-minute artificial intelligence (AI) computerized test, according to the findings from a recent study published in Frontiers in Psychiatry. Useful for large-scale screening of cognitive impairment and high-frequency monitoring, the ICA is unbiased in terms of language, culture, and education, according to its developer, Cognetivity Neurosciences.1,2
A total of 230 patients were given the ICA, with results demonstrating convergent validity with the Montreal Cognitive Assessment (MoCA; Pearson’s r = 0.58; P <.0001) and Addenbrooke’s Cognitive Examination (ACE; r = 0.62; P <.0001). ICA also showed a smaller correlation with education years (r = 0.17; P = 0.01) than MoCA (r = 0.34; P <.0001) and ACE (r = 0.41; P <00001), both significant correlations. Performance data was generalized from each population to the other, and the ICA AI model detected cognitive impairment with an area under the curve (AUC) of 81% for patients with MCI and 88% for patients with mild AD.1
“The diagnostic accuracy of the ICA and its novel use of explainable AI, combined with the power to generalize across other languages and cultures, make it uniquely suitable for cognitive screening across large and diverse populations,” coauthor Dag Aarsland, MD, PhD, professor of old age psychiatry, King’s College London, said in a statement.2 “And in light of the FDA’s recent approval of the disease-modifying drug aducanumab, the need for a device capable of screening a wide population of at-risk individuals has never been higher.”
Of the 230 patients studied, 95 were healthy volunteers, 80 were patients with MCI, and 55 were patients with mild AD. All participants completed each of the 3 cognitive assessments, with the ICA AI itself showing increased performance level when provided with additional training data.
Healthy participants were included in the study as a control group, 12 of whom assisted in the investigation of practice effect by completing the ICA test remotely. The 1-way ANOVA testing showed no significant practice effects on healthy participants who completed the ICA test remotely 78 times over the course of 96.8 days, indicating the absence of practice effect. When investigating the level of education bias, investigators found no significant statistical difference in education years when using the ICA model, which is also able to be generalized without necessitating the collection of population-specific, normative data.
“Empowered by AI, the ICA has the clear and exciting potential to achieve enhanced performance over time and to enable personalized medicine irrespective of geographic boundaries, as our new paper shows,” Seyed Khaligh-Razavi, PhD, chief scientific officer, Cognetivity, said in a statement.2 “The use of AI in decision making, especially for diagnostic decisions in healthcare, requires a level of explainability from the model which can be used to understand the important factors that lead to its output. This level of explainability is clearly achieved in the ICA. It can give clinicians full confidence in the platform, support improvements in system performance over time, and protect against the serious danger of bias.”
Investigators identified limitations in terms of recruitment, as fewer younger adults with mild AD were recruited, which was attributed to a small population of young mild AD patients. The study also included fewer adults with mild AD who also had higher years of education.
In light of the COVID-19 pandemic, investigators called attention to the need for more partially and fully remote cognitive assessment capability. The ICA can be used as a digital cognitive biomarker for both AD and MCI detection, further able to be used as a monitoring tool in the clinic and potentially remotely, without the loss of testing accuracy. According to investigators, additional validation will be required before remote administration is possible.