In the present study, using two separate cohorts, we modelled disease progression from preclinical AD to AD dementia and determined whether APOE ε4 status and sex affected progression across the entire AD spectrum. Our main findings were as follows. Our novel disease progression model indicated that it would take 7.8 years for preclinical AD to progress to MCI due to AD and 15.2 years to progress to AD dementia based on median ADAS-cog 13 scores. APOE ε4 carriers and women had worse cognitive trajectories across the entire AD spectrum. Across all sex and APOE ε4 combinations, female APOE ε4 carriers had the fastest cognitive decline. Taken together, our findings provide a further understanding of AD progression across the disease spectrum, and they will help to design individualized therapeutic and preventive strategies to ameliorate cognitive decline.
We modelled the AD disease progression course using two different cohorts and estimated that it took almost 15 years for preclinical AD to progress to AD dementia. In a recent article [15], 14.5% of individuals with preclinical AD developed incident MCI due to AD within a 3.7 year (mean) follow-up period, and 3.2% developed AD dementia within 4.2 years of follow-up [15]. Additionally, studies have found that 32.7% [15] and 70.0% [16] of individuals with MCI due to AD developed AD dementia within 3.2 and 3.6 years of follow-up, respectively [15, 16]. However, 2–4 years of follow-up may not be sufficient to estimate the entire course of disease progression. These previous findings, thus, mainly characterize fast decliners in each disease stage. However, our estimated course is consistent with indirect evidence provided in previous studies [6, 17], according to which the temporal lag between Aβ deposition and the clinical syndrome of AD dementia was a decade [6]. In a meta-analysis, age-related increases in amyloid positivity on PET in participants with normal cognition paralleled age-specific, AD-type dementia prevalence estimates with an intervening period of about 20 years [17, 18]. Another study estimated that it took 19.2 years for 11C-PiB levels observed in healthy controls with a 1.5 SUVR threshold to reach the mean SUVR of AD (2.3) [19]. Our finding that it would take more than 15 years for preclinical AD to progress to AD dementia suggests that appropriate interventions are needed to prevent preclinical AD from progressing to AD dementia.
In the present study, the estimated time from the preclinical to prodromal stage (7.8 years) was similar to that from the prodromal to dementia stage (7.4 years). Initially, we expected that the preclinical phase might be longer than the prodromal phase. Our finding might have been related to our definition of the prodromal phase using the early stage of MCI. If we define MCI due to AD as LMCI, the estimated time from preclinical AD to LMCI (8.9 years) would be longer than that from MCI due to AD to AD dementia (6.3 years). Alternatively, the study design—in particular, whether a study includes volunteer or clinic-based participants—might affect time-to-event estimates. For example, studies may overestimate the progression rate in the presymptomatic phase because the included participants might have more concerns about their cognition. Our disease progression model could therefore be used to estimate the current and future state of preclinical AD patients in a prevention trial.
Another main finding is that APOE ε4 and sex had distinct effects on the progression course across the AD continuum. Our finding that APOE ε4 aggravated cognitive decline across the entire AD spectrum regardless of sex is partially consistent with previous studies. While APOE ε4 is a well-known risk factor for AD dementia in the preclinical or prodromal stage [20], it has been debated whether APOE ε4 predicts a worse prognosis [21, 22]. A previous study by our group revealed that APOE ε4 predicted more rapid hippocampal and cortical atrophy in dementia with AD [21]. However, other studies have suggested that AD patients with APOE ε4 had a lower global amyloid burden than matched APOE ε4 non-carriers [22–24]. This discrepancy might be due to differences in the study populations (patients who progressed to AD dementia over time in the current study sample compared to patients who had already progressed to AD dementia in previous studies).
A more noteworthy finding that female APOE ε4 carriers showed more prominent cognitive decline than did male APOE ε4 carriers across the AD spectrum [25, 26]. Our findings are consistent with a previous study [25], which showed that women with higher Aβ levels had a faster cognitive decline than men and that women with preclinical AD who were APOE ε4 carriers declined faster than their men counterparts. However, the previous findings were not statistically significant after correction for multiple comparisons [25]. Our findings further suggest that female APOE ε4 carriers had a steeper cognitive decline than did male APOE ε4 carriers throughout the entire AD spectrum. Therefore, developing a progression model stratified by these factors will help to select cohorts for AD clinical trials.
Several possible explanations may account for the combined effects of sex and APOE [27–30]. A potential mechanism could be that oestradiol promotes synaptic sprouting in response to injury through an APOE-dependent mechanism [27]. Additionally, oestrogen might promote neural function under normal conditions, but exacerbate dysfunction when network activity is disrupted [28]. Alternatively, a previous study showed that the APOE ε4-by-sex interaction on cerebrospinal fluid (CSF) tau levels was significant, suggesting that the increased APOE-related risk in women may be associated with tau pathology [29]. In a recent multicohort study [30], women showed a stronger association between APOE and CSF tau levels than did men, particularly among amyloid-positive individuals, suggesting that APOE may modulate the risk of downstream neurodegeneration in a sex-specific manner, particularly in the presence of amyloidosis.
The ADNI is a well-organized, longitudinal cohort that serves as an excellent resource to investigate the disease course of AD. This study, however, has several limitations. We only included participants who were amyloid-positive by PET. This leaves open the possibility that some patients had another primary pathological diagnosis. Although participants clinically diagnosed with frontotemporal dementia or dementia with Lewy bodies and who had moderate to severe white matter hyperintensity were excluded from the ADNI dataset, we did not consider the effects of other neurodegenerative pathologies, including cerebrovascular disease, α-synuclein, transactive response DNA-binding protein, argyrophilic grain pathology, and hippocampal sclerosis, on the progression model. Importantly, amyloid positivity might only be a contributing or incidental factor in some patients with dementia. This argument is mitigated to some degree by the fact that we included participants who progressed from MCI due to AD to AD dementia. Additionally, we found that the ADAS-cog 13 scores in some participants with CN and MCI improved over time. Although we controlled for LEs, we did not completely exclude the possibility that LEs might affect the disease progression to some degree.
Nevertheless, ADAS-cog 13 is the standard tool used in many clinical trials to assess AD, which makes our results more interpretable across studies than if we had used another instrument. Finally, our progression rate from NC to MCI (29.1%) was higher than has been observed in community-recruited older adults. For example, a greater risk of progression from NC to MCI was observed in clinically-recruited older adults (30% per year) than in community-recruited older adults (5% per year) [31]. The ADNI used identical recruitment mechanisms to those of typical trials, including advertising and recruitment from memory clinics. Although our data might not be representative of the general population, the recruitment and subject baseline characteristics were similar to those of a typical AD clinical trial.