Participants
The INSIGHT-PreAD cohort has been described previously [14]. Briefly, 318 participants (and 318 study partners) were included in the study. They were French individuals between 70 and 85 years of age, with normal scores on Mini-Mental State Examination (MMSE ≥ 27), Clinical Dementia Rating scale (CDR = 0), and Free and Cued Selective Rating Test (FCSRT total recall score ≥ 41), who reported cognitive complaints at the study baseline (the subject answered yes to both of the following questions: Do you complain about your memory? Is this a regular complaint that has lasted for more than 6 months?). They had no evidence of monogenic AD mutations or other neurological disorders. The study partner had to be a person close to the subject, aware of potential recent changes in health and cognition. Each participant gave his/her informed consent, and Paris VI ethical committee approved the study protocol.
INSIGHT-PreAD is a prospective ongoing cohort study. When we performed the statistical analyses, the participants were undergoing their M54 or M60 visit. Only data up to M36 were analyzed in this article (7 timepoints), because the data of subsequent visits had not yet been fully collected and/or checked. However, at the time of the analyses,14 participants had been labeled as “decliners” (that is, exhibit at least two of the following changes over two consecutive evaluations 6 months apart: CDR increasing from 0 to 0.5 and/or an MMSE below 26 and/or a FCSRT total score below 40). The INSIGHT-PreAD protocol stipulated that participants would stop the follow-up as soon as they were classified as “decliners.” Three subjects were classified as “decliners” at the 24-month visit, three subjects at 36 months, one at 42 months, one at 54 months, and eight subjects at 60 months.
Cognitive measures
We investigated the impact of baseline ACD on the evolution of the cognitive scores that are supposed to be most relevant to the study of ACD in early-stage AD.
All INSIGHT-PreAD participants performed cognitive screening tests every 6 months and underwent a comprehensive neuropsychological evaluation every 12 months.
On the one hand, impaired awareness is associated with a suboptimal online self-monitoring, error detection and disinhibition [15,16,17]. On the other hand, memory disorders prevent correct comparisons between current and past performance [18]. Therefore, we included (1) the Trail Making Test (TMT) B-A score, the Lexical and Semantic Fluency and the Frontal Assessment Battery (FAB), as measures of executive functioning; (2) the free recall and total recall scores from the FCSRT, an episodic memory test sensitive to hippocampal damage; and (3) the Mini-Mental State Examination (MMSE) as a measure of global cognitive functioning.
Determination of the Awareness of Cognitive Decline Index (ACDI)
The procedure for identifying ACDI is reported in our previous publication [10].
In summary, the subjects and their study partners filled out two similar versions of the Healthy Aging Brain Care Monitor (HABC-M [19, 20]). This is a questionnaire asking how often, during the last 2 weeks, the participant has encountered certain difficulties in his/her everyday life. The questions are the same in the participant version and the study partner version and only asked slightly differently. For example, the first question in the participant version is “Over the past two weeks, how often did you have problems with judgment or decision making?”. In the informant version it is “Over the past two weeks, how often did your loved one have problems with judgment or decision making?”
Answers range from 0 (never) to 3 (very often). Since we aimed at studying awareness of cognitive decline, we only considered the HABC-M cognitive score, which is the sum of 6 items, and ranges from 0 to 18. The ACDI was determined by subtracting the HABC-M cognitive score obtained by the informant from that obtained by the subject. The ACDI ranged from − 18 to 18, where higher scores indicated heightened ACD (patient’s report > informant’s report), and lower scores, low ACD. The ACDI was computed at each visit (every 6 months) and treated as a continuous variable.
Brain imaging
In the present study, we included baseline neuroimaging markers of AD.
All participants performed amyloid-PET imaging using 18F-AV-45 (18F-florbetapir; Amyvid™, Avid Radiopharmaceuticals) as a tracer. The standardized uptake value (18F-AV-45-SUV) was calculated in target regions (i.e., left and right precuneus, anterior cingulum, posterior cingulum, parietal, temporal, and orbitofrontal cortex) with the CATI pipeline (Centre d’Acquisition et de Traitement d’Images, https://cati-neuroimaging.com), and normalized to the cerebellum and pons, resulting in a SUV ratio (SUVr). In the present study, we considered the amyloid load as a continuous variable (mean 18F-AV-45 SUVr of the aforementioned regions) and as a dichotomic variable. To this end, the SUVr positivity threshold was 0.79, which was analogous to the threshold found using a method validated by Gael Chetelat in the IMAP study.
We also examined cortical glucose metabolism using fluorodeoxyglucose (FDG) PET. A metabolism index was calculated by averaging the FDG-SUVs of four bilateral regions of interest, whose metabolic changes are considered to be a “signature” of AD [21]: posterior cingulate cortex, inferior parietal lobule, precuneus, and inferior temporal gyrus. The pons was used as a reference region. The FDG-SUV has been included as a continuous variable.
More details about imaging data acquisition are available in previous works [14].
Statistical analysis
For the first objective, we performed a Latent Class Linear Mixed Model (LCLMM [22]) to investigate heterogeneous trajectories of ACD, since they are expected in a cohort of memory-complainers. The LCLMM first identifies G classes of subjects who share a similar trend of evolution of the ACDI and then compares the classes. In order to find the adequate and clinically relevant number of classes G, we computed the model from one to three classes and selected the one which minimized the Bayesian Information Criterion (BIC). The mean of posterior probabilities and the percentage of posterior probabilities higher than 0.7 were computed. The evolution of the ACDI was modeled by the interaction between classes and visits. Using a multinomial logistic model, the baseline characteristics of each class (i.e., amyloid load, glucose metabolism, age, gender, and educational level) were compared to the class with the largest number of subjects. Normality of residuals and random effects as well as heteroskedasticity were checked visually. For this analysis, subjects with at least two timepoints of ACDI and with no missing baseline data (amyloid load, metabolism, age, gender, and education) were included.
We also performed generalized linear mixed effects models (GLMM) to evaluate the effect of the ACDI at baseline on changes in cognitive scores (objective 2) and the effect of longitudinal changes in ACD on the evolution of cognitive scores (objective 3). Link function was chosen for the underlying data generation mechanism with logit for binomial data and identity for continuous data. One GLMM was performed for each score, and then the Benjamini-Hochberg method was used to correct for multiple comparisons. For the second objective, we entered the following baseline variables as fixed effects: ACDI, age, gender, and educational level, visit, and the interaction between visit and ACDI, to test the impact of ACDI at baseline on changes in cognitive scores. All two-way interactions between these effects were tested independently and were added in the final model if significant. The participant was added as a random effect. For the third objective, we entered the following baseline variables as fixed effects: class, age, gender, and educational level and visit. All two-way interactions between these effects were tested independently and were added in the final model if significant. The participant was added as a random effect. Type II likelihood ratio tests were used to test each fixed effect and interaction. Cohen’s f2 were calculated, using the marginal R2 [23], for each effect to estimate their size. For this analysis, we only included subjects with at least two timepoints for each cognitive score, and with no missing data for ACDI at baseline or class, age at baseline, gender, and education. Baseline characteristics were compared between subjects included and excluded in the analysis using the χ2 test for categorical variables and Student’s t test for continuous variables.
A p value < 0.05 was considered significant. Statistical analyses were performed using R3.6.1. The packages lme4 (1.1-21) and LCMM (1.8.1) were used to perform LMM and LCLMM, respectively.