Toward the right treatment at the right time: Modeling the trajectory of cognitive decline to identify the earliest age of change in people with Alzheimer's disease

Abstract Introduction Age is the greatest risk factor for Alzheimer's disease (AD). A limitation of randomized control trials in AD is a lack of specificity in the age ranges of participants who are enrolled in studies of disease‐modifying therapies. We aimed to apply Emax (i.e., maximum effect) modeling as a novel approach to identity ideal treatment windows. Methods Emax curves were fitted to longitudinal cognitive data of 101 participants with AD and 1392 healthy controls. We included the Mini‐Mental State Examination (MMSE) and tests of verbal fluency and executive functioning. Results In people with AD, the earliest decline in the MMSE could be detected in the 67–71 age band while verbal fluency declined from the 41–45 age band. In healthy controls, changes in cognition showed a later trajectory of decline. Discussion Emax modeling could be used to design more efficient trials which has implications for randomized control trials targeting the earlier stages of AD.

method of recruitment is to enroll participants with a diagnosis of mild AD or mild cognitive impairment (MCI; for DMT) who are within a typically large age range.A systematic review of 165 RCTs for AD treatments found that the mean age of participants was 73.6 years (SD = 8.2), 12 while 90% of RCTs for AD define a range from 55 to 85 years old, 20% of RCTs allow 50 to55 years old, and 15% allow for 85to 90-year-old participants.
In order to identify the optimal age to intervene earlier in the lifespan, approaches capable of pinpointing the earliest inflection point in the transition from healthy to pathological aging are required.While fluid and biomarkers are used for participant identification in RCTs, 13 tests of cognition should provide a complementary approach for monitoring disease progression 14 and to improve precision of recruitment as they represent abilities used daily which impact an individual's functioning.Some have also argued that for a treatment to be called disease-modifying, it should show a beneficial effect on cognition 15 ; therefore, sensitive measures are required to evaluate treatments 16,17 at different stages of the disease process.The most commonly used methods employed in RCTs to date have been based on changes observed in the very late stages of the disorder.One approach to identify the earliest point of cognitive decline is to model the trajectory of change in cognition by age by repurposing statistical approaches used in pharmacology research, namely Emax (i.e., maximum effect) models that are commonly used in modeling dose-response. 18evious work applying Emax modeling in Down syndrome, which is considered a genetic variant of AD 19 has identified the trajectory of cognitive decline in the preclinical and prodromal stages of AD.These findings have shown that changes in cognition can be detected as early as 20 years before the average age of diagnosis.We have demonstrated that such results can be used to assist with decisions on the optimal age to intervene and used to calculate sample size estimation for AD prevention trials. 14While these results have assisted trail design for the first RCT of a DMT to treat AD in Down syndrome, 20 it is currently unclear whether this approach can be applied to people with sporadic AD, where there may be more heterogeneity in terms of age of onset and progression.

Objective
The aim of this study was to validate Emax modeling as a novel method to retrospectively identify the earliest age of change in cognition of people who would ultimately develop sporadic AD and compare these individuals to healthy controls without AD.

Study design and setting
We used data from a prospective longitudinal cohort study of adults aged 65 years old and older from Montpellier, France. 21Baseline assessments were completed between 1999 and 2002 with follow-up

Participants
Our sample consisted of participants taken from a larger sample of 2259 individuals aged between 65 and 90 years who were selected by random sampling from the electoral rolls of the Montpellier district as part of the Enquête Sante ´Psychologique-Risques, Incidence et Traitement (ESPRIT) study of late-life neuropsychiatric disorders. 21e to variability in follow-up time between assessments (2.8 years to 4.45 years), we sampled 1493 participants who had T2 data which was at least 3 years to a maximum of 4 years from their baseline assessment in order to reduce the impact of time between assessments on performance.
Baseline prevalent cases of all-cause dementia were excluded from further analysis in order to focus on new cases of AD during the study time.The remaining participants were categorized into two groups based on whether they were ever diagnosed with possible and probable AD while enrolled in the study (n = 101) or did not receive a diagnosis of any subtype of dementia (n = 1392, the healthy control group).We chose to analyze data from people who had a confirmed AD diagnosis during their enrolment in the study as this allowed us to work backward to identify when these individuals began to exhibit cognitive decline from their baseline assessment, with the certainty that they would eventually develop AD.Dementia diagnosis was based on the DSM-IV criteria. 22We included a healthy control group to validate the results found in people with AD.

Ethics
The study protocol was approved by the Ethical Committee of the University-Hospital of Bicêtre (France) and written informed consent was obtained from each participant.

Measures
We measured verbal fluency, executive functioning, and global cog- (MMSE) 25 was used as a global measure of cognitive functioning.

Statistical analysis
Following previous methods, 14 we fitted Emax models to cognitive data to examine their trajectory of change by increasing age.To identify the earliest age-bands of changes in performance for each test of cognition we explored dose-response relationships between performance change over a 3-to 4-year period with increasing "doses" of age.In the Emax models, we assumed a sigmoidal relationship between performance decline and age in years.This constrained a baseline level of cognitive stability, followed by a constant period of decline that eventually asymptotes.To explore change in performance on a yearly basis, mean proportional change between timepoints was calculated across participants in 5-year smooth moving-average baseline age bands, starting at age 65 and subsequently incrementing by 1 year.females and like the dementia group, participants were primarily of high school level education.They were in their early seventies at baseline and mid-seventies at T2.Approximately 18% of healthy controls

As presented in
were apolipoprotein E (APOE) ε4 carriers.

Earliest age-band of changes in cognition for people with AD and healthy controls
For performance on both the MMSE and verbal fluency (Isaacs Set Test), the Emax curves for people with AD (Figure 1) showed a steeper and earlier decline compared to the healthy control group (Figure 2).

Sample size estimations for a hypothetical RCT people with AD
The EC 1 age-band for verbal fluency was 41-45 years old, which is substantially younger than our data would allow to calculate a sample F I G U R E 1 E max curves for people with AD.

DISCUSSION
We found that in participants who ultimately develop AD, the earliest changes general cognitive functioning and verbal fluency can be detected considerably earlier than the mean age of AD diagnosis in our sample (mean = 82 years old, SD = 5.56).Moreover, mean age of AD diagnosis was similar to AD onset previously reported in the literature (80 years old). 27Importantly, we demonstrated a later trajectory of change in the large healthy population of older adults who did not develop AD.
Previous authors have suggested one of the reasons RCTs in AD have been largely unsuccessful is because of poor trial design. 28There has been substantial work on methods to improve RCT design including research on participants' priorities to increase willingness to participate in RCTs, 29 suggestions to use Bayesian adaptive RCT designs, 30 and using polygenic hazard scores to stratify participants. 31However, limited consideration has been given to the age of participants when they enter into an RCT and the accepted convention is to enroll partic-ipants over 50, 60, or 65 years to 85 years old depending on the target population, study aims or choice of outcome measure.Age is the most salient risk factors for AD, with the possibility of a diagnosis doubling every 5 years after age 60 years. 32By 85 years and older, a diagnosis of AD is nearly 14 times higher in this age group compared to people aged 65-69 years old. 33Yet these individuals are all collapsed in a single group for RCTs.We used age of participants in age-bands to model how they performed on measures of cognition with increasing age.Our results suggest that Emax modeling could contribute and benefit RCT development by guiding the optimal recruitment period to maximize any potential benefits of treatments which could prevent or delay the development of AD.
An important consideration for an effective treatment is to ensure that it is administered to the right participants at the right time.RCTs which are more selective in the age range of participants and recruit participants when they exhibit the earliest signs of decline could potentially improve the success of DMT by intervening earlier in the disease stage.This approach could also help to provide better data related to treatment efficacy, 30 which is currently lacking in the literature.Several authors have made recommendations for earlier detection to be utilized in RCTs to optimize intervention windows. 11,34These data provide validation of Emax modeling as a viable approach to identify the earliest inflection of change in cognition to intervene earlier in the AD disease course.
As previously shown, combining trajectory modeling with sample size calculations is important to identify appropriate measures which can be used as cognitive endpoints in RCTs. 14The results from the present study emphasize this consideration.While the measure of global cognitive functioning (MMSE) was able to detect the earliest changes in cognition up to 15 years before the mean age of AD diagnosis within our sample, the decline in performance was negligible and it would require a sample of 23,096 participants.The MMSE is commonly used in RCTs and was used as a cognitive endpoint in studies of DMT such as aducanumab, 35 but our results suggest that while it may be sensitive to early changes, the sample size calculations demonstrated that it may not be particularly useful as a cognitive endpoint due to the large numbers needed to detect a change in performance between control and treatment groups.
The results from the verbal task suggest that this cognitive ability shows the earliest decline in the 4th decade of life in people who eventually develop AD.Tests of verbal fluency and visuo-spatial memory are known to be sensitive to early AD related changes with decline trajectories being evident in the decade preceding diagnosis. 36reover, our previous work with Emax modeling and verbal fluency has demonstrated that it is sensitive to the earliest AD-related change in adults with Down syndrome. 14Another study in adults with Down syndrome showed that verbal fluency performance was negatively correlated with the established AD biomarkers, neurofilament light, and glial fibrillary acidic protein. 37However, the EC 1 age-band (1% of the maximum effect of age on performance) for the healthy control group was also a similar age band (43-47 years old) as the AD group (41-45 years old).Verbal fluency is a test which is highly sensitive to changes in multiple brain regions 38,39 ; therefore, the early changes and early age bands identified could reflect general age-related changes instead of AD related changes in these participants.These results should be examined further and future work using Emax modeling could investigate the trajectory of change using a lifespan approach by incorporating participants across a broader age range including those in midlife and/or younger to examine the robustness of the current findings.

Strengths and limitations
Utilizing data from people who ultimately developed AD, we have shown that there is benefit to refining the age of recruitment in identifying potential participants, since trajectory modelling can pinpoint the earliest decline in cognitive abilities, even in people with sporadic AD, which is considerably more heterogenous 40 than genetic variants which exhibit a more predictable onset. 41We included only participants who had an AD diagnosis and excluded participants with other dementias making our findings more applicable to current RCTs which are primarily conducted in people with AD.Moreover, our healthy controls and AD participants were sampled from community dwelling adults increasing the generalizability of their performance on the measures of cognition.Our results also provide replication and validation data for previous work in Down syndrome 14 but in a larger sample size from the general population.Finally, we utilized a novel and innovative approach to address a neglected aspect of RCT design which could improve treatment outcomes.
There are several limitations which should be addressed.First, the cognitive tests used in the present study may not be the preeminent choice as cognitive endpoints for an AD prevention RCT, and more sensitive tests could be examined in future work.We also did not include measures of functional abilities which are important to consider as outcome in RCTs. 42AD is a heterogeneous condition 43 and performance on cognitive tests can vary.Moreover, our sample size was limited to 101 AD participants.While we made efforts to account for this heterogeneity by calculating performance in 5-year smooth moving-average baseline age bands, the variability in performance found in AD could have impacted our results.Furthermore, we only considered age as a risk factor, but there are many other variables that can increase a person's risk of AD including APOE. 44Future studies would benefit from considering the impact of these variables using a larger sample within the age ranges exhibiting the earliest changes in cognition to understand how these may contribute to risk of cognitive decline due to AD.

CONCLUSION
These results suggest that trajectory modeling could be used to enrich and improve RCT design by providing data on the ideal age range to intervene.We have shown that sensitivity to change depends on the primary outcome as well as the age of participants, so authors designing RCTs need to ensure the population and primary outcome are consistent with the research question.Combining this approach with markers of risk such as APOE e4 or those with biomarker evidence of potential early AD (e.g., amyloid PET) could help to design more efficient RCTs, particularly for those targeting the stages of disease before an AD diagnosis.
nitive functioning using tests which are sensitive to early cognitive changes in MCI and preclinical studies.The Isaacs Set Test 23 was used to measure verbal fluency, the Trail Making Test, Part B 24 assessed frontal executive functioning and the Mini-Mental State Examination Verbal fluency showed the earliest EC 1 values in people with AD, demonstrating change in the 41-45 age-band, and between the ages of 43 and 47 years for healthy controls (Table2).This was followed by global cognitive functioning (MMSE) which showed the earliest decline from ages 67 to 71 years for people with AD, but much later in healthy controls with decline only starting in the 87-91 age band.The Emax model for executive functioning (Trail Making Test, Part B) failed to converge for people with AD.For healthy controls, it showed a late EC 1 value, with the earliest changes in performance being detected between the ages of 85 and 89 years.
U R E 2 E max curves for healthy controls.

Table 1
Demographics of people with AD and healthy controls.
, the AD sample consisted of predominantly females and participants were mostly educated to high school level TA B L E 1 a Dementia group, n = 98; healthy controls, n = 1375.(36%).The mean age at enrolment was approximately 75 years old and on average participants were in their late 70s at T2.They were diagnosed with AD in their early 80s and over a third of the sample were APOE ε4 carriers.The healthy control group comprised of nearly 60% Age-bands of 1%, 5%, and 10% of the maximum effect of age on performance for people with AD and healthy controls TA B L E 2a Not EC 1 value.Calculated using independent samples one-sided t-tests (α = 0.05) and 80% power.Abbreviation: MMSE, Mini-Mental State Examination.sizeforahypothetical RCT.However, as Table3 demonstrates, the 77-81 age-band showed an achievable sample size for an RCT with a total sample size of 768 (Cohen's d = −0.18)with a treatment which was 35% effective and this test as the cognitive endpoint.With an AD treatment which was 75% effective, an RCT would require total of 168 participants (Cohen's d = −0.39).Using the EC 1 age-band of 67-71 years old for global cognitive functioning, an RCT with a treatment which was 35% effective would require a total of 106,042 participants (Cohen's d = −0.03)and 23,096 (Cohen's d = −0.03)if treatment was 75% effective.