Principal ResultsThis study investigated the effectiveness of a measure of subjective cognitive decline, ECog, to detect CI in two cohorts: web-based in BHR and in-clinic in ADNI. The first finding was that the ECog demonstrated moderate accuracy in distinguishing individuals with CI and CU in both BHR and ADNI. The second finding was that the web-based ECog in BHR showed a comparable ability to detect CI, similar to the in-clinic ECog scores in ADNI. The third finding was that the ECog12, the short form of the ECog, was as effective as the full-length ECog for detecting CI in the BHR. These results indicate that ECog scores, whether administered web-based or in-clinic, and in both short-form and full-length versions, are useful in identifying individuals clinically diagnosed with CI.
The first finding of this study is that the web-based ECog demonstrated a moderate association with CI. Both online Self- and SP-ECog scores effectively distinguished individuals clinically diagnosed with MCI and CU. This result aligns with previous research that has shown a strong association between web-based SP-ECog and self-reported diagnoses of MCI or AD [7]. Unlike prior research that relied solely on either remotely collected data [7,26] or in-clinic data [6,8,9], our analysis included individuals with both web-based ECog data and clinical evaluations, allowing for a more rigorous assessment of the relationship between web-based subjective cognitive decline and objectively diagnosed CI. This finding highlights the potential clinical utility of online subjective cognitive decline measures, suggesting that they may reflect underlying cognitive impairment as determined by expert clinical evaluation. Previous in-clinic studies have reported that informant-reported ECog scores are better at predicting CI from CU than self-reported scores [6,8,9]. A prior study focusing on older adults with CI found that web-based SP-ECog could distinguish AD dementia from MCI, while Self-ECog could not [27]. The ability of both Self- and SP-ECog to identify CI in this study might be due to the fact that most participants in BHR had mild cognitive symptoms [28], which makes it less likely that they are unaware of their cognitive status [29]. It is also noteworthy that establishing web-based ECog cut point scores with good sensitivities, specificities, and accuracies is useful as a screening tool to detect CI in the older adult population. Our findings demonstrate the utility of web-based ECog for detecting the clinical diagnosis of CI, particularly in mildly impaired populations.
The second finding is that web-based ECog and in-clinic ECog produced comparable results in discriminating CI from CU. Previous studies have found that online ECog scores closely correspond with in-clinic ECog scores [30]. Our study newly revealed that web-based ECog is as effective as in-clinic ECog for detecting CI from CU, as demonstrated by the AUC comparison, which showed no significant difference between the two measures. This finding suggests that remotely collected ECog can serve as a viable alternative to in-clinic ECog for cognitive screening, expanding accessibility to cognitive assessments beyond traditional clinical settings. The trend toward overall higher discriminative AUC values, although not significant, in ADNI compared to BHR might be attributed to the nature of web-based scales, including response quality, lower engagement due to convenience, and potential technical issues [31-33]. From a clinical perspective, however, the lack of statistical significance in the AUC comparison suggests that both web-based and in-clinic ECog can be considered equally effective for CI screening, with the web-based version offering the added benefits of convenience and accessibility, particularly in populations with limited access to in-person assessments.
The third finding is that Self-ECog12 performed comparably to full-length Self-ECog in distinguishing CI from CU in a remote setting. ECog12, derived from the original 39-item ECog, maintains good psychometric properties for various cognitive domains, including memory, visuospatial function, language, and executive function [8]. Previous studies have shown that ECog12 is as effective as full-length ECog in identifying the risk of MCI [6,9]. The fact that web-based ECog12 shows a similar association with the diagnosis of CI versus CU as the 39-item ECog supports its use in web-based clinical studies like BHR. Recent large-scale neuroscience studies have started collecting online data from questionnaires about cognitive function, leveraging extensive and longitudinal collections [34-36]. ADNI 4, the current phase of ADNI, also recently implemented web-based ECog12 in a novel online screening approach to aid in selecting participants for plasma AD biomarker analysis and in-clinic evaluation [12,21]. However, SP-ECog12 showed a significantly lower probability of detecting CI from CU compared to SP-ECog in this study, although both SP-ECog and SP-ECog12 showed moderate AUCs. We believe this discrepancy may be due to the brevity and the reduced amount of information collected by study partners in the web-based ECog12. This might be further supported by the fact that study partners, who may not know the participants well or see them frequently, can still take part in the BHR study [7]. Nevertheless, the web-based ECog12 remains a useful and time-efficient tool for identifying individuals at risk of MCI in remote settings.
Recently, advances in online technology have allowed the use of web-based cognitive tests to identify CI with convenience and scalability [37,38]. Self-administered web-based digital neuropsychological tests are an effective method for identifying CI, offering benefits such as accessibility, efficiency, and diagnostic accuracy for dementia [39]. Many well-known in-clinic tests, including the Montreal Cognitive Assessment, Cogstate Brief Battery, and Cambridge Neuropsychological Test Automated Battery (CANTAB, Cambridge Cognition Ltd), have been adapted for online use [40-43]. Although neuropsychological tests objectively measure cognitive abilities through rigorous standardized tasks, ECog, which is a brief measure of subjective cognitive decline, captures an individual’s or their informant’s perception of cognitive and functional changes over the past 10 years in real life. This can provide early indicators of cognitive decline that objective tests might miss [5,29]. As ECog and ECog12 can be easily administered in a remote setting with validity for identifying CI from CU, integrating both objective and subjective measures of cognitive decline in web-based studies could enhance early detection of MCI who are potential candidates for disease-modifying therapy and comprehensive understanding of cognitive health.
LimitationsThis study has several limitations. First, the BHR eVAL study and ADNI consist predominantly of participants with over 16 years of education, 85% of whom are non-Latinx White. This lack of demographic diversity likely limits the generalizability of the findings and introduces bias related to differences in race, ethnicity, cognitive reserve, and socioeconomic status—factors that can influence the diagnosis, prognosis, and treatment response in dementia. African American and Latinx older adults are up to 1.5 to 2 times more likely to develop AD and AD-related dementias compared to White older adults [44]. These disparities may stem from differences in medical risk factors for AD (eg, depression, diabetes, obesity, and hypertension), biological factors (eg, genetics, epigenetics, or proteomic markers), factors contributing to cognitive reserve (eg, education and occupation), and nonbiological or social factors (eg, economic stability, safety, housing, and environment) [12]. Additionally, BHR has limitations, including selection bias, low data integrity, informant variability, and high dropout rates. Participants are more likely to have higher digital proficiency and increased awareness of cognitive health issues, further skewing the representativeness of the sample. On the other hand, web-based studies offer the advantage of scalability, enabling participation from individuals who might otherwise be unable to join due to geographic or time constraints. However, challenges such as data inaccuracy, missing information, and the inability to verify whether participants completed assessments independently or with assistance remain. Another important limitation is the difficulty in managing participant dropout over time. Prior studies have also shown that non-White or Latinx individuals, as well as those with lower educational attainment, are more likely to discontinue participation in follow-up assessments in web-based settings [45]. These patterns highlight the need for more inclusive recruitment strategies and efforts to retain diverse populations in future studies to ensure findings are more broadly applicable.
The second limitation is that this study was not originally designed to compare web-based and in-clinic populations. Instead, it compared these populations retrospectively, meaning that the recruitment process was not perfectly controlled from the outset, and there may be participation self-selection biases. Third, the cross-sectional design of this study limits the interpretation of the results due to the lack of longitudinal cause-and-effect relationships between ECog scores and clinical diagnoses. Fourth, the lack of correction for multiple comparisons is an additional limitation of the study.