The ICA test (also referred to as Cognetivity’s ICA or ‘CognICA’) has been described in detail in previous publications (Kalafatis and Khaligh Razavi, 2019; Kalafatis et al., 2019; Khaligh-Razavi et al., 2019; Kalafatis et al., 2021). The ICA test is a rapid visual categorization task. The test takes advantage of the human brain’s strong reaction to animal stimuli. One hundred natural images (50 of animals and 50 of not containing an animal) of various levels of difficulty are selected and are presented to the participant in rapid succession.

Each image is presented for 100 ms followed by a 20 ms inter-stimulus interval (ISI), followed by a dynamic noise mask (for 250 ms), followed by the subject’s categorization into animal vs. non-animal.

The raw data from the ICA is composed of reaction time and categorization accuracy on the images. This data was used to calculate summary features such as overall accuracy, and speed using the same methodology as described previously (Khaligh-Razavi et al., 2019; Kalafatis et al., 2021).

Accuracy is defined as follows:

Speed is defined based on participant’s response reaction times in trials they responded correctly:

A summary ICA Index, is calculated as follows:

The ICA Index describes the raw test result, incorporating speed and accuracy, the two main elements of the ICA test.

The AI model utilizes inputs from accuracy and speed of responses to the ICA rapid categorization task (with the ICA Index as an input feature), as well as age, and outputs an indication of likelihood of impairment (AI probability) by comparing the test performance and age of a patient to those previously taken by healthy and cognitively impaired individuals. The AI model is able to achieve an improved classification accuracy relative to using any single feature from the ICA test.

A probability threshold value of 0.5 was used to convert the AI probability to the AI prediction of healthy or cognitively impaired (MCI/mild AD).

The ICA AI model used in this study was a binary logistic regression machine learning model which is a supervised linear classifier. The algorithm’s task is to learn a set of weights from a regression model that maps the participant’s ICA test results and demographics to the classification label of healthy or cognitively impaired.

The ICA AI model was exclusively trained on previous datasets, including data from Khaligh-Razavi et al. (2019) and Kalafatis et al. (2021), and no data from the ADePT study was used to re-train or adjust the model.

Health Research Authority and Health and Care Research Wales approval for this study was obtained in February 2020. The study is registered in the ISRCTN Registry (ISRCTN16596456). Approved 27/02/2020, North of Scotland Research Ethics Committee (Summerfield House, 2 Eday Road, Aberdeen, AB15 6RE, UK; +44 (0)1224 558458; nosres@nhs.net), ref: 20/NS/0029.

All participants were recruited among attendees at the National Health Service (NHS) memory assessment services at the point of referral by their GP. Participants were recruited from Devon Partnership NHS Trust, North Bristol NHS Foundation Trust, Oxford Health NHS Foundation Trust, and Sussex Partnership NHS Foundation Trust.

Protocol for the Accelerating Dementia Pathway Technologies (ADePT) Study (Kalafatis et al., 2022) is also published and includes more detailed information about the study protocol and procedures.

The participants who did not have a formal diagnosis of a neurodegenerative disease were triaged as per usual clinical practice and were asked to complete the ICA in parallel with the diagnostic assessment.

The main study inclusion criterion was referral to the memory clinic by a GP. Patients recruited were 55 to 90 years old. Potential participants had to be fully informed of and understand the objectives, procedures, and possible benefits and risks of the study and have the capacity to provide written consent.

Subjects that met the following criteria were excluded from the study cohort:

Participants enrolled in the study were required to attend one visit at a designated memory clinic or remotely at their home (Assessment Visit 1 [AV1]). Participants were asked to complete the ICA. Prior to taking the ICA, participants were requested to view a short training video to assist them in completing the task successfully. After taking the ICA, patients completed the following short questionnaires:

Lastly, a brief medical history of the participants via electronic health care records was obtained, mainly focusing on any cognitive tests that have been taken by the participants.

Participants were given the option to carry out a retest visit (Assessment Visit 2 [AV2]) whereby they were again given the chance to take the ICA test either remotely or face-to-face, complete a usability questionnaire, and respond to inquiries on stimulants, fatigue, and sleep. The overall study pathway for participants is detailed at a high level within Figure 2.

The primary outcome of the study was designed to evaluate the accuracy and efficiency of GP referrals in comparison to the specialist diagnosis of MCI/dementia. Here’s a step-by-step breakdown of how this was achieved:

By comparing GP referrals, ICA assessments, and specialist diagnoses, we aimed to demonstrate the potential of the ICA in reducing unnecessary referrals and improving the efficiency of the dementia care pathway.

For the purposes of these analyses, patients referred to the memory clinic were divided into the following 3 groups, based on their memory clinic outcome: (A) those who receive a diagnosis of MCI or dementia, (B) those who are identified as healthy or receive a diagnosis of a brain or mental disorder other than MCI or dementia, and (C) those who receive an inconclusive diagnosis.

Participants with an inconclusive outcome after the memory clinic assessment were excluded from further analysis.

Participants in group A were counted as correct GP referrals. Participants in group B were counted as unnecessary or incorrect referrals.

Comparison with Specialist Diagnosis of MCI/Dementia: The metrics for GP referrals that were calculated are the following:

Likewise, the following complementary metrics for the ICA were calculated:

In a secondary outcome analysis, we compared with specialist diagnosis of all types of cognitive impairment (those due to MCI, dementia, or other neurological or mental disorders).

Cut-offs were applied to cognitive tests taken in primary care and secondary care, to investigate the predictive accuracy of these tests compared to the clinical diagnosis determined at the memory clinic. Cognitive tests taken in the 6 months prior to the memory clinic assessment were used for the analysis. The results from the cognitive tests are compared to ICA results on the same patients. The cognitive tests used for this analysis were ACE-III, MoCA, MoCA Blind, GPCOG and SCIT. A sufficient number of participants had completed these assessments within 6 months of the memory clinic visit to make the analysis viable.

The test–retest reliability of the ICA was analyzed by the following:

In addition to the above, we also employed the Kappa statistic (κ) to assess the test–retest reliability for the ICA as well as ICA association with the outcome of other cognitive tests (Altman, 1991). This metric provides a measure of agreement between two raters (in this case, two tests) while accounting for the agreement that might occur by chance.

The formula for the Kappa statistic is given by:

Where:

A Kappa value of 1 indicates perfect agreement, while a value of 0 indicates agreement equivalent to chance. Interpretation of the Kappa statistic typically follows these guidelines:

Multiple choice responses from participants were analyzed by calculating the proportion of participants who selected each option. Questions relating to frequency of tablet or mobile phone use were used to assess familiarity with technology, in particular, touch screen devices. The ease of understanding the ICA instructions and level of difficulty of the categorization task was analyzed by calculating the proportion of participants who reported finding each of these steps very easy, easy, moderately difficult, difficult, or very difficult.

Patients with inconclusive outcomes were excluded from our analysis. Other than that we did not have any other missing data regarding the calculations needed for primary and secondary outcome measures.

Clinical Practice Context: Currently, GPs’ clinical judgment is the primary determinant for initiating full evaluations at memory clinics for conditions like dementia. The cognitive tests available in primary care are either rudimentary, time-consuming, or detect cognitive impairment only when the disease has advanced considerably.

Out of 99 participants, 86 participants completed all assessments and questionnaires successfully. Some participants did not complete the ICA (13.13%), the majority of these were patients diagnosed with dementia (69.2%), followed by 7.7% of patients with MCI, 7.7% inconclusive, 7.7% with non-dementia conditions and 7.7% healthy. Of those who did not complete the trial, 11 did not complete the ICA, either due to inability to complete the test, or due to the Researcher’s decision to cancel the test while 1 participant decided to not proceed with the study procedures. Finally there was also 1 participant that was discontinued due to passing away prior to receiving their diagnosis. Table 1 illustrates the recruited participants breakdown.

The distribution of age and education years for all participants is shown in Table 2, and broken down by diagnosis in Table 3.

The following analysis is based on participants who completed the required assessments and questionnaires.

Of the 86 participants who completed the ICA, 5 had inconclusive diagnosis, hence the number of participants under investigation for the primary outcome analysis is 81 (Tables 4–7).

In this analysis the precision of GP referral is compared to specialist diagnosis of MCI, dementia and all other types of non-dementia neurological/mental disorders.

Tables 8, 9 provide a comprehensive breakdown of GP referrals in relation to specialist diagnoses for various cognitive impairments. They detail the total number of patients referred, those diagnosed with specific conditions, and the proportions of necessary and unnecessary referrals. Overall, 81.5% of the GP referrals were necessary, aligning with specialist diagnoses of cognitive impairments, while 18.5% were deemed unnecessary.

Tables 10, 11 provide insights into the performance of the ICA test in classifying cognitive impairments. The ICA AI outcome demonstrated an accuracy of 0.87 [CI¹: 0.81, 0.94] with a commendable AUC of 0.93 [CI: 0.86, 0.99]. Specifically, the sensitivity and specificity in detecting cognitive impairment were 0.89% [CI: 0.82, 0.96] and 0.80 [CI: 0.61, 0.97], respectively. A detailed breakdown of sensitivity and specificity for MCI and dementia, separately, as well as a comparison to the standard of care pen and paper tests, is further elucidated in Table 12. When comparing these metrics to the GP referrals from Table 8, the ICA shows a higher accuracy in correctly identifying patients with cognitive impairments. The ICA would have referred 62 patients, with 89.4% of these referrals being correct, further emphasizing its potential as a reliable screening and referral tool.

Note that in Tables 9, 10 the reported performance metrics for the classification is healthy vs. impaired (i.e., MCI/Dementia/Other). The breakdown of sensitivity and specificity for healthy vs. MCI and healthy vs. dementia is included in Table 12.

There were 24 participants who completed the initial visit and the retest. On average, there were 9.7 days between the test and retest (with an SD of 3 days, ranging from minimum of 5 days to a maximum of 14 days). The correlation between the ICA test retest is shown in Figure 3. The Spearman Rank coefficient for the test–retest ICA Index was calculated to be 0.79 (p < 0.000). While the average ICA Index in the retest group was 3.8 points higher than the initial test (see Bland Altman plot in Figure 4), paired t-test value of p was >0.1, suggesting this was not a statistically significant increase in score on the retest (Table 13). The Spearman Rank coefficient for the test–retest of ICA AI score was found to be 0.81 (p < 0.000).

The 2×2 matrix of ICA prediction on the first and second test is shown in Table 14.

The test-retest prediction agreement for the ICA is 87.5%.

In assessing the consistency of the ICA predictions between the first and second visits, we also utilized the kappa statistic (κ) to measure the agreement while accounting for chance agreement (Altman, 1991). Typically, Kappa values of 0.41 to 0.60 are considered moderate and a kappa of 0.61 to 0.80 represents substantial agreement. The observed proportion of agreement between the two visits was 87.5%. After calculations, the kappa statistic was found to be 0.7427. This value suggests a substantial level of agreement between the ICA predictions on the initial visit and the subsequent visit. This high degree of consistency underscores the reliability of the ICA tool across repeated assessments.”

In this section cut-offs are applied to cognitive tests taken in primary care and secondary care, to investigate the predictive accuracy of these tests compared to the clinical diagnosis determined at the memory clinic. Cognitive tests taken in the 6 months prior to the memory clinic assessment were used for the analysis. The results from the cognitive tests are compared to ICA results on the same patients.

The cognitive tests used for this analysis are ACE-III, MoCA, MoCA Blind, GPCOG and SCIT. A sufficient number of participants had completed these assessments within 6 months of the memory clinic visit to make the analysis viable.

We also calculated the correlation between the ICA Index and cognitive tests taken in memory clinics and GP settings, where there was at most 12 months difference in date between the tests. Results were included where there were at least 10 data points for the correlation calculation. The ICA index had a Pearson correlation of 0.57 (p < 0.000) with ACE-III and 0.77 (p < 0.002) with MoCA, both of which were taken as part of the memory clinic assessment. The ICA had a lower correlation with the more rudimentary assessments taken in the GP setting, e.g., a correlation of 0.37 (p < 0.18) with GPCOG (see Appendix A1 for more details).

The ACE-III was the most common memory clinic assessment, completed by 68 participants. A cut off score of ≥88 was used for ACE-III scores for binary prediction of healthy and impaired. On average the ACE and ICA assessments were conducted 38 days and 24 days from the diagnosis date, respectively.

ACE outcomes for participants, broken down by diagnosis, are shown in Table 15, and compared with ICA outcomes for the same patients. A comparison of these cognitive tests should also consider that the average ICA completion time is ~9 min, compared to approximately 30 min for ACE-III.

Excluding inconclusive cases, the overall accuracy of ACE-III on these participants (n = 66) is 88%. Here we assumed that detecting cognitive impairment in non-dementia cases is a correct classification.

Comparing ICA and ACE predictions, where both the ICA and ACE were completed, results in the matrix shown in Table 16.

The overall percent agreement between the ICA and ACE is 79%. The kappa statistic for ICA vs. ACE is 0.464, indicating moderate agreement. Supplementary Table S3 shows the diagnosis, ICA and ACE scores for the participants where the ICA and ACE had differing predictions.

For MoCA assessments, which were conducted in memory clinics, a cut-off of 26 or greater was used to predict healthy patients. On average the ICA test was taken 17 days apart from the diagnosis date, while the MoCA assessment was taken 58 days from diagnosis. MoCA outcomes for participants, broken down by diagnosis, are shown in Table 17.

Excluding inconclusive cases and non-dementia conditions, the overall accuracy of MoCA on these participants (n = 10) is 70%. Here we assumed that detecting cognitive impairment in non-dementia cases is a correct classification.

Comparing ICA and MoCA predictions, where both the ICA and MoCA were completed, results in the matrix shown in Table 18.

Overall the ICA and MoCA have a percentage agreement of 70%. The kappa statistic for ICA vs. MoCA is 0.444, suggesting a moderate agreement. The cases where the ICA and MoCA prediction were different are shown in Supplementary Table S4.

This analysis was based on MoCA Blind assessments carried out by the memory clinic. A score ≥ 18 for healthy was used as the cut-off for healthy participants (Mahendran et al., 2015). The ICA assessment was 15 days from diagnosis date, while the MoCA Blind assessment was 16 days from diagnosis date. MoCA Blind outcomes for participants, broken down by diagnosis, are shown in Table 19.

The overall accuracy of MoCA Blind for these participants (n = 14) is 93%. Note the small sample size in comparison to ICA and ACE. Here we assumed that detecting cognitive impairment in non-dementia cases is a correct classification.

Comparing ICA and MoCA Blind predictions, where both the ICA and MoCA Blind were completed, results in the matrix shown in Table 20.

The kappa statistic for the comparison between ICA and MoCA Blind is 0, indicating no agreement beyond chance, despite a high observed agreement of 92%. This discrepancy is due to the skewed distribution in the contingency table and the small sample size.

There was only one instance (from 12) in which the assessments differed (Supplementary Table S5). In this case the ICA predicted impairment, in agreement with the clinical diagnosis of MCI, while the MoCA Blind score of 18 was suggestive of a healthy patient.

The following cut offs were used for the GPCOG assessment, based on Brodaty et al. (2004):

The ICA test was completed on average 15 days before diagnosis, while the GPCOG test was completed on average 67 days before diagnosis.

GPCOG outcomes for participants, broken down by diagnosis, are shown in Table 21.

The overall accuracy of GPCOG for these participants (n = 16) is 75%. Note the small sample size in comparison to ICA and ACE. Here we assumed that detecting cognitive impairment in non-dementia cases is a correct classification. We also combined ‘Further Investigate’ with ‘Impaired’ to indicate that GPCOG had identified cognitive impairment.

Comparing ICA and GPCOG predictions, where both the ICA and GPCOG were completed, results in the matrix shown in Table 22.

The observed percent agreement between GPCOG and ICA is 0.50 (excluding cases labeled as ‘Further Investigate’). The kappa statistic for the comparison between ICA and GPCOG is 0.193, suggesting only a slight agreement beyond chance.

For the 6CIT assessment, carried out at GPs, a cut off score of 8 or greater was used to indicate cognitive impairment, based on Brooke and Bullock (1999). For the participants who completed the 6CIT and ICA, the ICA was completed on average 31 days before diagnosis, while the 6CIT test was completed on average 103 days before diagnosis.

6CIT outcomes for participants, broken down by diagnosis, are shown in Table 23.

Excluding inconclusive cases, the overall accuracy of 6CIT for these participants (n = 13) is 69%.

Comparing ICA and 6CIT predictions, where both the ICA and 6CIT were completed, results in the matrix shown in Table 24.

The observed percent agreement between 6CIT and ICA is 0.54. The kappa statistic for the comparison between ICA and 6CIT is −0.05, suggesting that there is no agreement between ICA and 6CIT beyond what would be expected by chance.

Table 12 provides a comprehensive comparison of the ICA tool with various standard care cognitive tests in terms of their sensitivity and specificity for detecting cognitive impairment in MCI and Dementia. The table showcases the performance metrics of the ICA alongside several widely recognized cognitive tests, offering a clear perspective on how the ICA stands in relation to these standard assessments.

For MCI, the ICA tool demonstrates a sensitivity of 83% and a specificity of 80%. For Dementia, its sensitivity rises to 93%, while maintaining a specificity of 80%. These figures are juxtaposed with the metrics from standard care tests, such as the MMSE, GPCOG, 6CIT, and others. The table also provides the proportion of patients receiving each of these standard tests, offering insights into their prevalence in clinical settings.

The weighted row at the bottom of the table gives a consolidated view, factoring in the proportion of patients for each test, of the average sensitivity and specificity across all standard care tests. This provides a benchmark against which CognICA’s performance can be evaluated.

In comparison to the weighted average of standard care tests, CognICA demonstrates superior sensitivity for both MCI and Dementia, while maintaining comparable specificity.

An update to the CognICA application introduced a new test flow, which automated the pre-test demo/trial image process. More specifically, the pre-test image process was automated while a training video was introduced to facilitate remote assessments. These changes did not affect the core of the test.

This resulted in an increased ICA completion rate among the groups with impaired diagnoses (Table 25). For example for dementia patients the completion rate increased from 78 to 89%.

A further benefit of the updated test flow was a reduction in the number of trial images completed by patients. The completion rate of the ICA could also have been influenced by the new training video. This was introduced partway through the study, and is discussed in more detail further on. However, among dementia patients, there was an increased rate of completion (from 80 to 85%) when the video was watched compared to when it was not watched.

On average, from device hand over to completion of the test the ICA took ~9 min to complete in this setting. This is based on the researcher’s approximation of the time taken, and is not a precise measurement. The mean time taken for ICA completion was reduced from visit 1 to visit 2, from 9 min to 7 min (Supplementary Table S9) face-to-face administrations were on average 1 min shorter than remote administrations of the test (Supplementary Table S10); and across diagnoses, healthy individuals completed the test 3 min faster than cases with a diagnosis of dementia (Supplementary Table S11).

The t-test of visit 1 to visit 2 time, p = 0.03. This suggests that as expected the second test takes less time to complete compared to the first test. The t-test of face to face vs. remote test time taken p = 0.16, suggesting there is no significant difference in time taken across the two scenarios. While the meantime taken of all tests is less than 10 min, all cases where the time taken was >15 min occurred for dementia patients.

Of all the participants who completed the questionnaire (n = 99), about 41% had no prior experience with tablets, while ~35% used tablets on a daily or frequent basis, the rest used tablets only a few times a week (%10) or otherwise very rarely (~14%).

Healthy participants were more likely to be frequent users of tablets compared to MCI and Dementia patients. It is important to consider here that the sample size for healthy patients is smaller (n = 15) than MCI patients (n = 19), and Dementia patients (n = 54) (Supplementary Figure S1).

There is a higher percentage of regular smartphone usage (~%45) –compared to tablet–, however a similar pattern exists suggesting that those with higher levels of cognitive impairment use smartphone devices on a less regular basis (Supplementary Figure S2).

Of the 22 participants who attempted the ICA remotely for their first visit, only one chose not to continue and did not complete the ICA. The majority were able to operate the ICA remotely without difficulty. 55% of participants found it very easy or easy to operate the device from receiving the device to starting the instructions. 19% rated the experience with medium difficulty.

Those with some previous tablet experience found it easier to operate the device compared to those who had never used tablets previously.

In the retest (7 participants conducted a remote retest), none of the participants indicated a difficult/very difficult experience in operating the ICA remotely.

Most participants (~68%) found it very easy or easy to understand the ICA instructions. Those with cognitive impairment (particularly dementia patients) were more likely to find understanding the ICA test instructions difficult or very difficult (Supplementary Figure S3).

Those with some tablet experience were more likely to find understanding the test instructions very easy or easy, and less likely to find it difficult. However a confounding factor is that healthy patients are more likely to use tablet devices.

Looking just at dementia patients, we find that within this subgroup the same pattern of easier understanding of instructions with increased tablet use exists.

On retest visits, in comparison to the first visit, a higher proportion of participants found the test instructions very easy or easy (80%), with a lower percentage indicating difficulty.

Most participants had no or little anxiety in taking the ICA. On self-reported anxiety levels, MCI and dementia participants were more likely to report more anxiety compared to other groups. 10% of MCI and dementia cases said they became very anxious, compared to none (0%) in other groups.

All participants who undertook a retest reported either the same or lower level of anxiety on the retest. There were no reports of increased anxiety on the retest.

Overall most participants found the ICA test challenging, with less than 20% identifying the test as very easy or easy. This is to be expected as the test is designed to be challenging, there is no ceiling effect, and it is rare even for young healthy individuals to get a full score on the test.

However, after the retest the perception of participants changed, with the most common response being ‘medium’ difficulty (Supplementary Figure S4).

Approximately 55% of those who did a retest indicated the second test was as difficult as their first, while 46% indicated they found the second test easier.

In the majority of cases participants were able to carry out the test without distraction, whether in face to face visits (~85%) or remotely (~70%). Participants were more likely to report being ‘a little distracted’ in remote tests compared to face to face tests (21% compared to 8%).

There is variation in how participants held the iPad during the test. The most common (and recommended) position was held in both hands, however ~22% of participants placed the iPad on a stand, and ~ 32% placed the iPad flat on the table.

Participants with cognitive impairment were less likely to hold the device in both hands, and more likely to place the tablet flat on the table.

However, with some tablet use, participants are more likely to hold the device in both hands. Supplementary Figure S5 in the supplement shows this for dementia patients.

A significant majority of participants (~84%) expressed willingness to take the ICA again in the future. The percentage was slightly lower for those with cognitive impairment (~78%).

The percentage of those willing to take the ICA again increased to over 90% for those who also completed the retest.

The primary outcome of the study compared GP referral with specialist diagnosis of MCI/dementia. Of those the GP referred to memory clinics, 78% were necessary referrals, with ~22% unnecessary referrals, or patients who should have been referred to other services as they had disorders other than MCI/dementia. In the same population, the ICA was able to correctly identify cognitive impairment in ~90% of patients, with approximately 9% of patients being false negatives. From the subset of unnecessary GP referrals, the ICA classified ~72% of those as not having cognitive impairment, suggesting that these unnecessary referrals may not have been made if the ICA was in use.

The proportion of necessary GP referral in our work is higher than what has been found in other studies like Creavin et al. (2021), where 56% was reported. It is likely that the GPs’ referrals assessed for the purpose of this study had a special interest in Dementia diagnosis, thus the higher rate of accurate referrals. The sample size here is less than 100 participants from four sites in the UK, therefore GP referral accuracy from larger studies should also be considered. However the GP referral precision obtained from this study aligns with the values obtained from Sussex audit report for 2020–2021, which indicated there was no diagnosis of cognitive impairment in only 8% of cases from those who presented at the memory clinic.

Due to the challenges in recruiting patients from various GP sites, a main limitation of this study is that only those who were referred by GPs have been investigated. Therefore, the results from this study cannot be used to determine the false negative rate, i.e., those who did have cognitive impairment but were not referred by their GP. We seek to address this in future studies where the ICA will also be deployed in primary care.

In response to the COVID-19 pandemic, and the move toward remote assessments, the study protocol was amended to allow remote administration of the ICA. Twenty-two participants attempted the ICA remotely, with only one participant not wishing to continue. Most patients found the remote operation of CognICA very easy or easy, however, those with no prior tablet experience were more likely to have difficulty. In the majority of cases participants were able to carry out remote tests without distraction, however participants were more likely to report slight distraction in remote tests compared to face to face tests.

Overall the usability results show that most participants were able to understand the ICA instructions and complete the test, even though they found the rapid visual categorization task itself challenging. Patients’ usability experience was found to be related to previous familiarity with phones and tablets; those with some prior experience found it easier to understand the instructions and to operate the device. They were also more likely to hold the device in both hands rather than placing it flat on the table. There were also differences in perception among patients based on their clinical diagnosis. As expected those with MCI, dementia or a non-dementia condition were more likely to have difficulties, and to feel slight or mild anxiety in taking the ICA.

Those who took a retest reported an improved overall usability experience. They found the instructions easier to understand, experienced less or the same level of anxiety, found the test easier or the same as before and found operating the device in remote assessments easier than the first visit. Both after the first and second visits, a significant majority of patients indicated they would be willing to take the ICA again in the future.

Some patients did not manage to complete the test, in particular those with dementia or more severe cognitive impairment. An update to the ICA software which allowed participants to seamlessly progress from the pre-test demo and trial images to the main test, as well as the introduction of a new instruction video helped to increase the completion rate. For dementia patients the increase was from 78% in CognICA 1.2 to 89% in CognICA 1.3. It is important to note that ICA is particularly designed to detect and monitor subtle cognitive impairments or changes to cognitive performance in earlier stages of the disease, where there is an unmet need in clinical practice.

The Integrated Cognitive Assessment (ICA) emerges as a pivotal tool in the realm of primary care, addressing several challenges inherent in the current referral process for cognitive impairment.

The ICA’s objective measure of cognitive performance offers a potential solution to refine the referral process from primary care to specialized memory clinics. By accurately identifying those with cognitive impairment, the ICA can help ensure that those who truly require specialized care are referred, reducing unnecessary referrals, as well as reducing undiagnosed or late-diagnosis of dementia.

The economic strain on healthcare systems, particularly the NHS, is significant. By reducing the number of unwarranted referrals, the ICA not only alleviates the financial burden but also optimizes resource allocation, ensuring that specialized care is available to those who genuinely need it.

Early detection is paramount in managing cognitive impairment. The ICA facilitates this by offering a swift, non-intrusive assessment that can be easily integrated into primary care settings. This ensures that patients receive timely interventions, leading to better outcomes and an enhanced quality of life. This timely diagnosis can further lead to significant cost savings for the NHS, as highlighted by Shore et al. (2022). This becomes even more pertinent with the advent of new disease-modifying therapies for Alzheimer’s disease on the horizon.

The real-world data from the ADePT study, in conjunction with previous clinical studies of the ICA (Khaligh-Razavi et al., 2019 and Kalafatis et al., 2021), provides compelling evidence for the efficacy of the ICA as a screening tool. This robust evidence base underscores the potential of the ICA to revolutionize the dementia care pathway, starting at the primary care level.

In light of these considerations, the adoption of the ICA in primary care settings presents a forward-thinking approach to dementia care, aligning with the broader objectives of enhancing patient outcomes, optimizing resource utilization, and ensuring cost-effectiveness.