Abstract
Subjective memory is useful to detect cognitive deficits not apparent on objective tests. This research evaluates whether memory rated by an interviewer predicts incident dementia. Health and Retirement Study participants without cognitive impairment at baseline whose memory was rated by the interviewer were analyzed (N = 12,749). Worse interviewer-rated memory was associated with higher risk of incident dementia over 15 years (HR = 1.40, 95% CI = 1.27–1.54). The association was evident even among participants in the top quartile of objective memory performance (HR = 1.71, 95% CI = 1.26–2.32). Interviewer-rated memory may be a low-cost supplement to neuropsychological evaluation to predict dementia risk beyond self-reports and objective cognitive testing.
Keywords
Poor episodic memory identified through objective tests [1], self-report [2], or knowledgeable observers [3] is a consistent predictor of incident dementia. This association is not surprising given that memory deficits are a core component of Alzheimer’s disease and related dementias [4, 5]. Strikingly, both self-reported and observer-rated memory function predict incident dementia even when objective memory tests do not indicate memory deficits [2, 3]. Knowledgeable observers in particular have been useful for identifying cognitive changes that might have gone unnoticed by the individual [6] or undetected on standard memory tests [7]. Less is known about whether observers less well acquainted with the individual also pick up signs of poor memory that are predictive of dementia over a long follow-up. The present study used interviewer observations from the Health and Retirement Study (HRS) to test whether a simple interviewer-rated item on participant memory during the interview was associated with developing incident dementia over an up to 15-year follow-up. We further address whether this observer rating is predictive among participants with the best objective memory performance and compare it to self-rated memory.
METHODS
Participants and procedure
The HRS is an ongoing longitudinal study of adults aged 50 and older in the United States and their spouse, regardless of age [8]. Participants are interviewed about their health, financial situation, and well-being every two years. Participants provided informed consent prior to each interview. The 2006 assessment was chosen as baseline for the current study because it was the first to have a relatively balanced administration of interviews across face-to-face and phone modalities (instead of primarily one or the other). Cognitive measures were administered at every two-year assessment through 2020. Participants were selected into the analytical sample if they scored within the normal range of cognitive function (see below) at baseline, had an interviewer-rated memory score, and at least one follow-up assessment of cognition (N = 12,749, Mage = 65.70, SD = 10.07; range = 25-98). See Supplementary Figure 1 for flow chart of participant inclusion. IRB approval was not needed for this analysis because it was based on de-identified, publicly available data.
Measures
Interviewer-rated markers of cognitive function
Interviewers were staff (i.e., trained research assistants) who worked for the HRS. Following the 2–3-hour interview, interviewers rated the item, “How much difficulty did the respondent have remembering things that you asked (him/her) about?” on a scale from 1 (no difficulty) to 5 (could not do at all). Other items relevant to cognition were considered: distracted, attentiveness, effort, vocabulary, and question repetition (see Table 1 note for specific items and response scales). Interviewers administered the entire protocol and were not blinded to other measures.
Descriptive Statistics for Study Variables in the Full Sample and by Dementia Status at Follow-up
Descriptive Statistics for Study Variables in the Full Sample and by Dementia Status at Follow-up
N=12,749; n = 11,805 no dementia; n = 944 dementia. Self-rated memory was rated on a scale from 1 (excellent) to (poor). Episodic memory is the sum of immediate and delayed recall. Interviewer-rated memory was measured on a scale from 1 (no difficulty) to 5 (could not do at all). Interviewer-rated distraction was measured on a scale from 1 (never) to 3 (often). Interviewer-rated attentiveness was rated on a scale from 1 (not at all attentive) to 3 (very attentive) and reverse-scored in the direction of inattentiveness. Interviewer-rated effort was rated on a scale from 1 (a lot) to 4 (none). Interviewer-rated vocabulary was rated on a scale from 1 (below average) to 3 (above average) and reverse-scored in the direction of worse vocabulary. Interviewer-rated repeat question often was rated on a scale from 1 (seldom) to 3 (often). *p < 0.05 difference between participants who did and did not develop dementia over follow-up. **p < 0.001 difference between participants who did and did not develop dementia over follow-up.
Cognitive status was measured with the modified Telephone Interview for Cognitive Status (TICSm) [9]. The TICSm is a well-validated measure [10, 11], with a cut-off value for dementia validated by comparisons with clinical diagnoses in the Aging, Demographics, and Memory Study [9], and used to track trends in dementia prevalence over time [12]. The TICSm is the sum of three tasks: immediate and delayed memory of 10 words (20 points), serial 7 subtraction (5 points), and backward counting (2 points). From possible 27 points, cutoffs have been identified and validated [9, 12] for dementia (≤6), cognitive impairment not dementia (7–11), and non-impaired cognitive function (≥12). Participants in the non-impaired range at baseline were selected for the analytic sample.
Covariates
Sociodemographic covariates were age in years, sex (0 = male, 1 = female), race (1 = Black and 1 = otherwise identified both compared to 0 = white), Hispanic ethnicity (0 = no, 1 = yes), and education in years. Mode of interview (0 = face-to-face, 1 = telephone) and baseline cognitive status (continuous TICSm score) were also included as covariates. Supplemental analysis controlled for depressive symptoms, measured as the sum of 8 items from the Center for Epidemiological Studies Depression scale and self-rated memory, measured with the item, “How would you rate your memory at the present time?” and subjective hearing, measured with the item, “Is your hearing excellent, very good, good, fair, or poor?” both rated from 1 (excellent) to 5 (poor).
Analytic approach
Cox regression was used to test whether interviewer-rated memory was associated with risk of incident dementia. Time was measured in years from baseline and coded as time-to-incidence for participants who developed dementia. Cases were censored at the last available cognitive assessment for participants who did not develop dementia. Interviewer-rated memory was entered as a predictor of incident dementia over the up to 15 years of follow-up, controlling for sociodemographic covariates, interview mode, and continuous TICSm score. A supplemental analysis further controlled for depressive symptoms and self-rated memory. We tested whether the association was moderated by age, sex, race, ethnicity, education, interview mode, or baseline cognitive function (total TICSm score at baseline). A sensitivity analysis restricted the analytic sample to participants who scored in the top quartile of memory performance (sum of immediate and delayed recall≥13) to test whether this item was predictive among participants with the top objective memory performance in the sample (this model was limited to the participants in the top quartile of memory but did not include the memory score as a predictor because of the overlap with the total TICSm score). A second sensitivity analysis restricted the sample to participants who rated their memory as good or excellent (n = 4,043). These analyses were, in part, to test whether the interviewer made the memory rating based on the objective memory test or participant reported memory, respectively. A third sensitivity analysis split the sample by time to dementia (≤5 years, 6–10 years, and 11–15 years) to examine whether the association was only apparent close to dementia onset or whether interviewer-rated memory had long-term predictive power. We repeated this set of analyses on the other interviewer-rated items relevant to cognition to evaluate whether the association was apparent with other items or specific to interviewer-rated memory.
RESULTS
Descriptive statistics are in Table 1. Over the up to 15-year follow-up and 126,707 person-years, 7.4% (n = 944) of the sample who scored in the non-impaired range of cognition at baseline developed incident dementia. Results of the Cox regression are in Table 2. Interviewer-rated memory was associated with incident dementia (HR = 1.40, 95% CI = 1.27–1.54, p < 0.001): Every 1-point worse interviewer-rated memory was associated with 40% increased risk of developing dementia over the follow-up. The association was similar when depressive symptoms, self-reported memory, and subjective hearing were included as covariates. None of the interactions was significant, which indicated the association was similar across age, sex, race, ethnicity, education, interview mode, and cognitive function. Most notably, the sensitivity analysis indicated that the association was apparent and stronger when the sample was limited to participants with the best objective memory performance concurrent with the interviewer ratings; depressive symptoms, subjective memory and subjective hearing were not significant predictors in this restricted sample (Table 2). The second sensitivity analysis indicated that the association also remained when the sample was restricted to participants with better self-perceived memory (HR = 1.51, 95% CI = 1.21–1.87, p < 0.001). The third sensitivity analysis also indicated that although the association was weaker the longer the time to dementia, it was significant in each follow-up time-period (≤5 years: HR = 1.63, 95% CI = 1.37–1.94; n = 255 incident dementia; 6–10 years: HR = 1.40, 95% CI = 1.21–1.61; n = 447 incident dementia; 11–15 years: HR = 1.26, 95% CI = 1.02–1.56; n = 242 incident dementia). Supplementary Table 1 reports specificity, sensitivity, positive predictive value, and negative predictive value at each point of the response scale for interviewer-rated memory compared to none.
Cox Regression Predicting Dementia from Interviewer-Rated Memory
Cox Regression Predicting Dementia from Interviewer-Rated Memory
N = 12,749. HR, Hazard Ratio; CI, Confidence Interval. The restricted sample is the sample restricted to participants who performed in the top 25% on the baseline objective memory test (N = 3,544).
Among the other cognition-related ratings made by the interviewer, inattentiveness, a worse vocabulary, and often asking to repeat questions were associated with a greater risk of incident dementia over the follow-up in the first model and when controlling for depressive symptoms, self-rated memory, and subjective hearing (Supplementary Table 2). None of these other ratings, however, was associated with dementia risk when the sample was limited to participants in the top 25% of objective memory performance concurrent with the interview. Interestingly, when split by time to dementia, inattentiveness was a better longer-term than shorter-term predictor of dementia, whereas a worse vocabulary and often asking to repeat questions were better shorter-term than longer-term predictors.
This research indicates that a simple rating of memory from an interviewer was a significant predictor of incident dementia over up to 15 years and was the only interviewer-rated item to remain significant when the sample was limited to participants with the best objective memory performance at baseline. Interviewer-rated memory was a good marker of future dementia among the most cognitively healthy.
These findings make at least two contributions. First, the results support growing evidence for the importance of subjective ratings of memory, particularly by observers, and it extends the association to observers less acquainted with the target. Subjective cognitive decline (SCD), in the absence of an objective impairment, for example, predicts incident dementia [13], as does cognitive function rated by knowledgeable observers, such as a spouse or other family member [14]. This study indicates that a simple evaluation by an interviewer may help detect a deficit predictive of possible dementia that is not captured in either objective cognitive performance or self-reported memory.
Second, this research indicates a simple interviewer rating has long-term power to predict who may develop dementia and may thus be useful as an early marker of impairment. Notably, interviewer-rated memory predicted dementia risk even among participants who scored in the top quartile of memory function at baseline. Such performance on an objective memory test typically suggests that the individual has good cognitive function and is not at risk of impairment. The SCD literature highlights the limitation of objective measures [15]. The present research suggests that memory deficits predictive of dementia can be detected from interviews and are more predictive over long periods of time than self-reported memory.
The association was similar across age, sex, race, ethnicity, and education, and, of note, by interview mode. It may have been expected that the association would be stronger in the face-to-face interview because the interviewer could have picked up on nonverbal cues not available over the phone. The association, however, was similar across interview mode, which indicated that the interviewer detected memory deficits through the interaction rather than through visual cues.
Strengths of this research include the large sample, interviewer-rated observations, and long follow-up. Limitations include the use of a performance-based measure of dementia instead of a clinical diagnosis and lack of knowledgeable informant ratings of memory (e.g., from spouses) to compare the predictive power of interviewer-rated memory. Future research should replicate this association with a clinical diagnosis in addition to performance-based identification of dementia. In addition, it was not possible to evaluate inter-rater reliability of interviewer-rated memory since only one interviewer rated each participant. Future research will need to better evaluate how to measure interviewer-rated memory to develop the most psychometrically robust measures. The current research was also not able to account for participant’s language preference. The associations were the same accounting for whether the interview was administered in English or Spanish (based on participant preference) but the interview was not offered in other languages. Future research would also benefit from comparing interviewer-rated memory to ratings made by knowledgeable informants (e.g., spouses) to determine whether interviewer-rated memory provides additional information on risk beyond ratings by knowledgeable observers. Still, this research indicates that memory performance rated by a relatively naïve interviewer (albeit with knowledge of aging because on their employment with HRS) predicts incident dementia, above and beyond objective memory performance and self-rated memory, and especially among those with the best objective memory performance. If replicated, subjective ratings by clinical staff (e.g., individuals conducting neuropsychological testing) could provide valuable information in predicting risk of impairment beyond standardized cognitive performance tests. Utility of these ratings will be informed by future research that identifies the subjective criteria interviewers base their ratings on and identifying which cues are most predictive of dementia. A better understanding of such cues may be useful to integrate into dementia surveillance to identify those most at risk not identified through neuropsychological testing and provide information on how neuropathology manifests in symptoms that may be detectable in conversations prior to clinical symptoms.
Footnotes
ACKNOWLEDGMENTS
We gratefully acknowledge the Health and Retirement Study whose public data made this work possible. The Health and Retirement Study is sponsored by the National Institute on Aging (NIA-U01AG009740) and conducted by the University of Michigan.
FUNDING
Research reported in this publication was supported by the National Institute on Aging of the National Institutes of Health under Award Number R01AG053297 and R01AG068093. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
CONFLICT OF INTEREST
The authors have no conflict of interest to report. Dr. Aschwanden and Dr. Terracciano are Editorial Board Members of this journal but were not involved in the peer-review process nor had access to any information regarding its peer-review.
