What works and what does not work in Alzheimer’s disease? From interventions on risk factors to anti‐amyloid trials

Alzheimer's disease (AD) is a progressive neurodegenerative disorder with no approved disease‐modifying therapy (DMT). In this review, we summarize the various past approaches taken in an attempt to find treatments capable of altering the long‐term course for individuals with AD, including: translating epidemiological observations into potential treatment options; seeking a single‐treatment approach across the continuum of AD severity; utilizing biomarkers for assessing target engagement; using biomarkers as early surrogates of clinical efficacy; and enriching study populations to demonstrate adequate placebo decline during the limited duration of clinical trials. Although targeting the amyloid‐β (Aβ) pathway has been central to the search for an effective DMT, to date, trials of anti‐Aβ monoclonal antibodies have failed to consistently demonstrate significant clinical efficacy. Key learnings from these anti‐Aβ trials, as well as the trials that came before them, have shifted the focus within clinical development programs to identifying target populations thought most likely to benefit from treatments (i.e., individuals at an earlier stage of disease). Other learnings include strategies to increase the likelihood of showing measurable improvements within the clinical trial setting by better predicting decline in placebo participants, as well as developing measures to quantify the needed treatment exposure (e.g., higher doses). Given the complexity associated with AD pathology and progression, treatments targeting non‐amyloid AD pathologies in combination with anti‐amyloid therapies may offer an alternative for the successful development of DMTs.


| INTRODUC TI ON
Alzheimer's disease (AD) is a progressive neurodegenerative disorder with a complex underlying pathology (Serrano-Pozo, Frosch, Masliah, & Hyman, 2011). The progression of AD occurs along a continuum, beginning with asymptomatic neuropathological changes that gradually lead to memory loss as well as other cognitive, functional, and behavioral impairments (Burns & Iliffe, 2009;Rafii & Aisen, 2015). These impairments result in the gradual loss of independence, with many patients eventually becoming unable to perform routine activities of daily living and, ultimately, may lead to premature death (Apostolova, 2016). Globally, the economic cost of dementia was estimated to be $818 billion in 2015, a 35.4% increase in cost compared with the $604 billion estimate from 2010 (Wimo et al., 2017).
Despite the devastating consequences and the overall burden of AD, an effective disease-modifying therapy (DMT) capable of stopping or slowing the progression of the clinical symptoms and the underlying pathology, has not been established for this disease. Current therapies available in clinical practice are limited to symptomatic treatment and do not specifically slow underlying neuronal damage or, therefore, alter the course of disease progression (Grossberg, Tong, Burke, & Tariot, 2019;Yiannopoulou & Papageorgiou, 2013).
In this article, we aim to review some of the approaches tried, as well as lessons learned, in the search for a DMT capable of effectively stopping or slowing the decline of individuals with AD. The specific approaches that we focus on include: (1) translating epidemiological observations into potential treatment options; (2) seeking a single-treatment approach across the continuum of AD severity; (3) utilizing biomarkers for assessing target engagement; (4) using biomarkers as early surrogates of clinical efficacy; and (5) enriching study populations to demonstrate adequate placebo decline during the limited duration of clinical trials.
The approach of targeting individual, potentially modifiable risk factors in individuals with AD dementia has been evaluated in interventional studies (Table 1). However, it is important to keep in mind that an observed association with AD does not necessarily equate to causality, which may explain in part why the majority of these interventions have failed to demonstrate a significant effect on disease progression so far. For example, modifying metabolic/cardiovascular risk factors did not lead to a significant slowing of cognitive decline (AD2000 Collaborative Group, 2008Aisen et al., 2000Aisen et al., , 2003Aisen et al., , 2008Feldman et al., 2010;Gold et al., 2010;Harrington et al., 2011;Kwok et al., 2011;Reines et al., 2004;van Rossum et al., 2012;Sano et al., 2011; SPRINT MIND Investigators for the SPRINT Research Group et al., 2019;Stein, Scherer, Ladd, & Harrison, 2011;Sun, Lu, Chien, Chen, & Chen, 2007;Van Gool, Weinstein, Scheltens, & Walstra, 2001); similar results were observed with estrogen replacement therapy (Table 1) (Mulnard et al., 2000;Rigaud et al., 2003).
In addition, there is evidence that supports the notion that regular physical exercise may serve as an option for preventing cognitive decline and dementia. Although some observational studies have demonstrated robust associations of physical activity with both delayed onset (Larson et al., 2006) and a reduced risk of dementia F I G U R E 1 Important risk factors for neurodegeneration. References: 1. (Nieoullon, 2011). 2. (McKenzie et al., 2017. 3. (Pike, 2017). 4. (Christensen & Pike, 2015). 5. (Barnes & Yaffe, 2011). 6. (Sommer et al., 2017). 7. (Kivipelto & Solomon, 2006). 8. ) (Andel et al., 2008;Buchman et al., 2012;Laurin, Verreault, Lindsay, MacPherson, & Rockwood, 2001), other observational studies have failed to show an effect on cognitive performance (Makizako et al., 2015;Smith et al., 2011). For trials evaluating the efficacy of exercise intervention programs, methodological heterogeneity across studies may have an influence on whether a benefit is observed and can also limit the conclusions that are drawn from the analyses (Farina, Rusted, & Tabet, 2014). While observational studies have inherent limitations that are difficult to overcome, they can inform interventional studies and remain a key area of AD research (Haeger, Costa, Schulz, & Reetz, 2019).
One reason for the lack of success associated with interventional studies that targeted potentially modifiable risk factors may be the presence of unaccounted confounding factors (Andrews, Marcora, & Goate, 2019). Thus, targeting a potentially modifiable risk factor without controlling for all potential confounders could render the interventional study more likely to be unsuccessful. Furthermore, the type of intervention may significantly influence clinical efficacy.
For example, exercise interventions vary greatly in their methodology; some may include aerobic or resistance exercises exclusively, whereas others combine exercise interventions with cognitive stimulation or with nutritional supplements (Farina et al., 2014;Haeger et al., 2019), resulting in a differential effect on cognition.
In addition, the duration of the intervention itself, regardless of the approach, may also limit our ability to detect clinical efficacy; intervention durations even > 1 year (Aisen et al., 2008;Feldman et al., 2010;Sano et al., 2011;Van Gool et al., 2001) may not provide an adequate exposure time to result in a measurable clinical benefit.
Although targeting individual risk factors has not yielded clear success for individuals already symptomatic with AD, epidemiological evidence suggests that a multidomain approach may benefit individuals at risk for AD (Ngandu et al., 2015). This concept was examined in the Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability (FINGER) (ClinicalTrials. gov, 2016 [NCT01041989]; Ngandu et al., 2015). The FINGER study was a 2-year, randomized controlled multidomain interventional study to prevent cognitive decline in at-risk elderly people from the general population. The study enrolled 1,260 participants, who were recruited from population-based national surveys. Participants were aged 60-77 years, with Cardiovascular Risk Factors, Aging, and Incidence of Dementia (CAIDE) Risk Score ≥ 6 and cognition measures that were at mean or slightly lower than expected for age evidenced by their performance during Consortium to Establish a Registry for Alzheimer's Disease screening. Eligible participants were randomized to either a multidomain intervention group (n = 631), which included nutritional counseling, physical activity, cognitive training, and vascular risk factor monitoring; or to a control group (n = 629) that received general health advice. Findings on the primary outcome, assessed by the Neuropsychological Test Battery (NTB), showed a between-group difference favoring the intervention group, which had a 25% greater improvement in the NTB total score at 24 months compared with control treatment (Ngandu et al., 2015).
Analyses of the NTB domain scores demonstrated improvement in executive functioning (83% higher) and processing speed (150% higher) in the intervention versus control groups (Ngandu et al., 2015). Moreover, health-related quality of life declined in the control group, but improved in the intervention group; general health and physical function at both 12 and 24 months were statistically significantly better in the intervention group (Strandberg et al., 2017). Overall, these results suggest that intervention based on epidemiological observations can lead to a viable multidomain interventional strategy for preserving aspects of cognition in identified at-risk individuals (Ngandu et al., 2015;Strandberg et al., 2017). However, they do not show that such benefits translate into a reduced risk for all-cause dementia, or dementia because of AD.

F I G U R E 2
Relative risk of modifiable risk factors in Alzheimer's disease. Reference: (Barnes & Yaffe, 2011) To summarize, addressing even strong risk factors for the development of AD in individuals who are already symptomatic has not been beneficial to date. The current multimodal interventions for elderly persons at risk for AD have demonstrated an impact on the type of age-associated cognitive decline observed with normal aging in the absence of dementia, but they have yet to prove an effect relevant to AD or other forms of dementia.

| S EEK ING A S ING LE-TRE ATMENT APPROACH ACROSS THE CONTIN U UM OF AD S E VERIT Y
AD progression occurs along a continuum, beginning as an asymptomatic preclinical stage and eventually resulting in cognitive and functional impairments as well as, ultimately, premature mortality (Apostolova, 2016). Based on this continuum, is it feasible for a drug to demonstrate efficacy during all stages of the disease? Early investigations using primarily neurotransmitter-based therapies have suggested that it is not. In Table 2, we have summarized the effect sizes of some AD drug therapies that have been evaluated to date.
Currently, the only approved therapeutic options are symptomatic treatments such as the N-methyl-D-aspartate receptor antagonist, memantine, and acetylcholinesterase inhibitors; both therapies have demonstrated temporary slowing of cognitive decline alone or in combination compared with placebo (Howard et al., 2012;Parsons, Danysz, Dekundy, & Pulte, 2013;Raina et al., 2008). Interestingly, memantine has shown a greater effect size on cognition and functional measures in more severe stages of AD (Schneider, Dagerman, Higgins, & McShane, 2011;Winblad, Jones, Wirth, Stoffler, & Möbius, 2007), whereas cholinesterase inhibitors have shown efficacy in a broader population, including individuals with mild, moderate, and severe AD (Table 2) (Blanco-Silvente et al., 2017;Winblad et al., 2009). In addition to their use as monotherapies, the combined use of memantine and acetylcholinesterase inhibitors has been shown to reduce clinical worsening in moderate-to-severe AD compared with those receiving the cholinesterase inhibitor, donepezil, alone (Atri et al., 2013), although this additive effect was no longer statistically significant compared with placebo following 1 year of treatment in a different study with TA B L E 1 Interventions targeting potentially modifiable risk factors in individuals with AD

Failed Agents Results Summary
NSAIDs and other antiinflammatory agents

Rofecoxib
No significant difference on ADAS-Cog a or CDR-SB versus placebo at 1 year in mild-to-moderate AD (Aisen et al., 2003;Reines et al., 2004) Naproxen No significant difference on ADAS-Cog a or CDR-SB versus placebo at 1 year in mild-to-moderate AD (Aisen et al., 2003) Prednisone No difference on ADAS-Cog a or CDR-SB versus placebo at 1 year in probable AD (Aisen et al., 2000) Hydroxychloroquine No difference on ADAS-Cog a versus placebo at 18 months in early AD (minimal or mild AD) (Van Gool et al., 2001) Aspirin Higher mean MMSE and lower mean basic ADLS score versus no aspirin at 3 years in mild-to-moderate AD (AD2000 Collaborative Group, 2008 Estrogen replacement therapy Estrogen No improvement on Clinical Global Impression of Change (CGIC), MMSE, or CDR at 1-year in mild-to-moderate AD (Mulnard et al., 2000) No significant changes on ADAS-Cog, MMSE, or CGC-plus versus placebo at 28 weeks in mild-to-moderately severe AD (Rigaud et al., 2003) Insulin control Rosiglitazone No difference on ADAS-Cog a versus placebo at week 24 in mild-tomoderate probable AD (Gold et al., 2010) No difference on ADAS-Cog a or CDR-SB at week 48 in mild-to-moderate AD (Harrington et al., 2011) Vitamin D Vitamin D Supplementation No benefit on ADAS-Cog a versus low-dose vitamin D in mild-to-moderate AD (Stein et al., 2011) Statins Simvastatin No effect on ADAS-Cog a versus placebo at 18 months in mild-to-moderate AD (Sano et al., 2011) Atorvastatin No significant difference on ADAS-Cog11 or ADCS-CGIC versus placebo at week 72 in mild-to-moderate probable AD (Feldman et al., 2010) Homocysteine B 6 + B 12 + Folic Acid No significant difference on ADAS-Cog11 or ADL function versus placebo at week 26 in mild-to-moderate AD (Sun et al., 2007) No beneficial effect on ADAS-Cog a or CDR-SB over 18 months versus placebo in mild-to-moderate AD (Aisen et al., 2008) Methylcobalamin + Folic Acid No significant difference on Mattis dementia rating scale (MDRS) at 24 months in mild-to-moderate AD (Kwok et al., 2011) different methodology and outcome measures (Howard et al., 2012).

Furthermore, discontinuation of cholinesterase inhibitors in patients
with AD or non-AD dementia has been associated with behavioral worsening (Daiello et al., 2009) (Siemers et al., 2016). Results of these secondary analyses served as the basis for EXPEDITION 3, a phase III placebo-controlled, double-blind study in participants with mild AD who were amyloid-positive by florbetapir positron emission tomography (PET) or by CSF Aβ(1-42) measurements (Honig et al., 2018). Unfortunately, even in the selected population with mild AD, solanezumab showed no slowing of cognitive decline on the primary outcome measure of ADAS-Cog14 (Honig et al., 2018  in studies with mid-domain binding antibodies (solanezumab and crenezumab), with ARIA-H ranging from 4.9%-13.1% (Cummings, Cohen, et al., 2018;Doody et al., 2014). Phase II studies of the primarily conformational (i.e., protofibril)-binding BAN2401 antibody reported ARIA-E rates of ≤10% (ARIA-H not disclosed) . Overall, ARIA-E remains a safety concern predominantly during trials evaluating N-terminus anti-amyloid monoclonal antibody therapies.
In summary, learnings from the clinical development of anti-Aβ monoclonal antibodies helped to identify target populations that could most likely benefit from timely and sufficient exposure to anti-Aβ monotherapy. Since anti-Aβ monoclonal antibodies failed to demonstrate clinical benefit in individuals with mild-to-moderate AD, at which stage the underlying pathological changes may already escalate in an amyloid-independent fashion, targeting Aβ species in these participants may occur too late to achieve clinical benefits (Pimplikar, Nixon, Robakis, Shen, & Tsai, 2010). The lack of clinical benefit in these individuals suggests that amyloid is not the only factor involved in the pathogenesis of AD (Pimplikar et al., 2010). The pathology may include both amyloid-dependent and -independent processes (i.e., amyloid may initiate the pathophysiology in AD which eventually becomes amyloid-independent) (Hyman, 2011). Thus, anti-amyloid therapies may not carry the same clinical benefit for patients later in their disease course and, because of this hypothesis, the focus shifted to evaluating individuals at an earlier stage of disease, for example, in early (i.e., prodromal-to-mild) AD, where the neuropathological changes associated with AD may be less advanced and more dependent on the presence of amyloid (Sperling, Mormino, & Johnson, 2014). Given the continuum of AD progression and the complexity associated with AD pathology, it is highly likely that different treatments targeting various stages of the disease may be needed; a therapy demonstrating efficacy in one stage does not guarantee that the same therapy will work in another stage.
Furthermore, sufficient continuous exposure to higher doses of an anti-Aβ therapy is another key learning, as demonstrated by the recent aducanumab results (Biogen, 2019). The high rate of failure associated with AD drug development has suggested that to have the best chance for treatment success, the right therapy at the right dose needs to be delivered to the right patient at the right time in the disease process, for the right duration (Cummings, Feldman, & Scheltens, 2019). However, whether this is the case because certain drugs only work at certain stages, or whether this reflects the ability of our current outcome measures and trial designs to demonstrate small treatment effects, remains to be seen.

| UTILIZING B I OMARK ER S FOR A SS E SS ING TARG E T ENG AG EMENT
As a result of key learnings over time, the field has evolved from one where any drug was expected to be applicable to any stage of AD, to one in which patients are targeted based upon stage of disease. The understanding of biological markers and their use in AD has therefore assumed new importance. An area of AD research in which biomarkers have demonstrated value is in the assessment of target molecule engagement. In this section, we will describe studies that evaluated target engagement of anti-amyloid therapies using imaging (e.g., PET), CSF and/or blood-based biomarkers.
In phase II testing of bapineuzumab, a reduction in PET amyloid protein load was observed (Rinne et al., 2010). 10 mg/kg donanemab was administered every 2 weeks .
In the phase Ib PRIME study, aducanumab treatment was associated with statistically significant dose-dependent reductions in Aβ across brain regions, except for the pons and subcortical white matter (two areas in which Aβ plaques would not be expected to accumulate) (Sevigny et al., 2016 (Yang et al., 2018). Studies of solanezumab (phase I and II) revealed evidence of target engagement by dose-dependent increases in CSF total Aβ (Farlow et al., 2012;Siemers et al., 2010). Results from the SCarlet RoAD study suggested a doseand time-dependent effect of gantenerumab on brain amyloid load as measured by SUVR on amyloid PET and also on a number of CSF markers that are thought to be disease-relevant, such as tTau, pTau, and neurogranin (Ostrowitzki et al., 2017).
Studies have also revealed evidence of target engagement by dose-dependent increases of total Aβ in plasma. Phase I and II studies of solanezumab demonstrated substantial dose-dependent increases in Aβ in plasma as well as in CSF (Farlow et al., 2012;Siemers et al., 2010). BAN2401, a humanized IgG1 monoclonal antibody which selectively binds and clears Aβ protofibrils (van Dyck, 2018), was investigated in a single-and multiple-ascending dose study (Logovinsky et al., 2016). Small dose-dependent increases in plasma Aβ(1-40) within a few hours after the first dose of BAN2401, and after the final dose, were reported. Plasma Aβ(1-40) levels declined over time with the fall in serum concentration of BAN2401 (Logovinsky et al., 2016). In addition, following administration of crenezumab, total plasma Aβ40 and Aβ42 have been shown to increase significantly, demonstrating target engagement in the periphery (Lin et al., 2018).
Lastly, although including imaging (e.g., PET), CSF and/or bloodbased biomarkers in clinical trials can be useful for providing a better understanding of the changes in underlying AD pathology in response to treatment (Blennow, 2017), there are still considerable challenges in establishing the appropriate use of biomarkers in both AD drug development and clinical practice. It is important to note that demonstrating target engagement through biomarkers does not guarantee success in later stages of drug development . Evidence of target engagement shown by a specific or the "right" biomarker, however, helps to demonstrate biological activity that may translate into clinical efficacy .
Biomarkers, therefore, may eventually provide surrogate outcomes in clinical trials of AD, if demonstrated to be predictive of clinical outcomes .

| US ING B I OMARKER S A S E ARLY SURROG ATE S OF CLINI C AL EFFI C AC Y IN PHA S E II TRIAL S
Despite their success in demonstrating target engagement, with helping to identify patients in the very early stage of the disease, and for shaping clinical trial programs, biomarkers studied to date have not become definitive surrogates of clinical efficacy. The majority of studies to date suggest that biomarker changes alone in phase II fail to predict clinical efficacy in phase III. For example, in the phase II clinical trials evaluating the gamma-secretase inhibitor, semagacestat, there was a significant reduction in plasma Aβ(1-40) concentrations (Fleisher et al., 2008); however, no significant reduction in CSF Aβ levels and no group differences in cognitive or functional measures were observed (Fleisher et al., 2008). Phase III semagacestat findings also demonstrated no significant change on the ADAS-Cog11 and there was worsening of several clinical outcomes, including ADCS-ADL, CDR-SB, Neuropsychiatric Inventory (NPI), and MMSE (Doody et al., 2013).
Anti-Aβ immunotherapy with solanezumab revealed a dose-dependent change in plasma and CSF Aβ in phase II; however, no changes in cognitive scores occurred (Farlow et al., 2012;Siemers et al., 2010). Although the biomarker data in phase II demonstrated dose-dependent changes, in phase III testing, there was no significant change on several clinical outcomes, including ADAS-Cog11, ADCS-ADL, CDR-SB, NPI, and MMSE, following treatment with solanezumab (Doody et al., 2014). In the recently updated analysis from the EMERGE and ENGAGE trials, the subset of participants with higher exposure to aducanumab had reduced amyloid PET and performed better on the CDR-SB versus placebo (Biogen, 2019).
Amyloid biomarkers were also used to advance the BACE1 inhibitor, verubecestat (MK-8931), from phase I/II, in which results demonstrated a 90% reduction in CSF Aβ (Forman et al., 2013), to the phase II/III trial (EPOCH) in participants with mild-to-moderate AD (Egan et al., 2018); later, Merck began the phase III APECS trial in participants with prodromal AD/MCI (Egan et al., 2019). EPOCH and APECS futility analysis results both demonstrated that verubecestat did not improve cognitive and functional decline (Egan et al., 2018(Egan et al., , 2019; participants with prodromal AD had worse cognitive decline than those treated with placebo on the CDR-SB, ADAS-Cog13, and ADCS-ADL-MCI outcome measures (Egan et al., 2019 ClinicalTrials.gov, 2019b [NCT03036280]); however, these trials were terminated early because of an unfavorable risk/benefit profile (Eisai, 2019).
Findings from the phase II trials evaluating the anti-Aβ immunotherapy, bapineuzumab, demonstrated no effect on Aβ(1-42) or tTau as measured by sandwich enzyme-linked immunosorbent assay Significant reductions in CSF tTau and pTau were reported in the phase III SCarlet RoAD study, but this study did not show clinical efficacy (Ostrowitzki et al., 2017). Starting at the 105 mg dose and uptitrating to the 225 mg dose in the phase III Marguerite RoAD study was associated with a significantly greater percentage reduction in tTau and pTau versus placebo . Similarly, CSF pTau levels were reduced by 13% following treatment with BAN2401 in the phase II study in early AD (Molinuevo et al., 2019). To date, reduction in CSF pTau has only been associated with improved global outcomes in the subset of participants with higher aducanumab exposure in the updated EMERGE and ENGAGE trial analyses (Biogen, 2019).
Based on the examples described earlier, the discordance between biomarkers and clinical efficacy may be because of the over-generalization of biomarkers that simply reflect the presence of pathology to the hope that they would predict or reflect cognitive or functional benefits. First, there is no reason to assume that change in a biomarker of the underlying pathology is correlated with clinical cognitive benefits, as the established neurodegeneration would not be expected to disappear (Figure 3). Second, patient heterogeneity is an important consideration; people with the same level of neurodegeneration could have different responses to treatment based upon other factors, such as the presence of additional non-AD proteinopathies (Robinson et al., 2018) and comorbid cerebrovascular disease, being an APOE ε4 carrier, and possibly because of unidentified genes that either amplify neurodegeneration or provide neuroprotection (Jack et al., 2010

| ENRI CHING S TUDY P OPUL ATI ON S TO DEMON S TR ATE ADEQUATE PL ACEBO DECLINE DURING THE LIMITED DUR ATI ON OF CLINI C AL TRIAL S
During AD clinical trials, participants in placebo arms worsen; however, they may decline slowly and demonstrate large and increasing variability during follow-up periods ( Figure 4) (Schneider & Sano, 2009). Yet the ability to demonstrate drug-placebo differences depends upon predictable decline in the placebo group. In an analysis of nine trials with available follow-up data, the mean changes and standard deviations on different versions of the ADAS-Cog indicate that approximately 25% of participants do not worsen by more than 1 point over 18 months (Schneider & Sano, 2009). This observation may explain why modest drug effect cannot reliably be recognized during limited clinical trial periods. More importantly, it raises the question, "How can detecting efficacy be improved?". Aside from relying on stronger drug effects, strategies to address lack of placebo decline may include having a larger sample size, or recruitment strategies that focus on specific participants who are not in an advanced stage of the disease but are more likely to progress during the study period without treatment.
In active disease, the pace of progression can be predicted by the preceding rate of deterioration (Capitani, Cazzaniga, Francescani, & Spinnler, 2004). For example, the Functional Assessment Staging Test (FAST) procedure characterizes seven stages in the course of AD from normal aging to severe dementia, and progression through future FAST stages can be statistically predicted based upon progression through earlier stages (Thalhauser & Komarova, 2012).
Thus, at the initial clinic visit, an assessment of the patient can predict subsequent longitudinal performance on multiple cognitive and functional measures over time . Using this approach, slow and intermediate progressors have been shown to diverge from the fast progressors on the ADAS-Cog over time ( Figure 5) . This predictive relationship also holds true for global performance, ADL measures, and even mortality .
In addition, the clinical trial evaluating the γ-secretase inhibitor avagacestat in prodromal AD or MCI provides evidence for the potential power to select participants who are more likely to progress quickly. In the randomized prodromal AD cohort, participants had clinical symptoms of cognitive impairment but not dementia, ratio ≥ 0.39), whereas participants in the observational cohort met MCI criteria but were CSF biomarker-negative (Coric et al., 2015).
Results demonstrated no significant treatment differences; but at 2 years, progression to dementia was more frequent in the prodromal AD cohort (30.7%) than in the observational cohort (6.5%) (Coric et al., 2015).
Analyses of baseline covariates revealed that CDR-SB, Functional Activities Questionnaire, and hippocampal volume were the three main factors in predicting progression type; fast progressors had a baseline CDR-SB ≥ 2, Functional Activities Questionnaire ≥ 4, and hippocampal volume less than the median (i.e., Delor criteria). On the basis of these prognostic factors, 81% of MCI participants could correctly be assigned to the slow-or fast-progressing subpopulations and 77% of MCI-to-AD conversions could be predicted (Delor et al., 2013). Automated Battery (CANTAB); however, it should be noted there was no trend observed for CDR-SB (Ostrowitzki et al., 2017;Retout et al., 2015).
In the crenezumab trials, inclusion criteria in the phase II ABBY and BLAZE studies required participants to have a CDR-SB score ≥ 0.5 and MMSE of 18-26 points (Cummings, Cohen, et al., 2018;Salloway et al., 2018). Biomarkers were evaluated in the BLAZE study  and biomarker inclusion criteria were also incorporated into the phase III CREAD study protocol . Moreover, CREAD participants were required to have a CDR Global Score of 0.5 or 1 and MMSE ≥ 22, as well as a Free and Cued Selective Reminding Test (FCSRT) free recall of ≤ 27 and cueing index ≤ 0.67 , in addition to amyloid pathology criteria: PET scan positive for cerebral amyloid-β and  FCSRT helps to identify participants with an elevated risk of developing AD dementia (Grober, Veroff, & Lipton, 2018), thus increasing the potential of enriching a clinical trial population with individuals with early AD who are likely to progress during the study. Since the ability of a DMT to demonstrate efficacy versus placebo is based partly on the rate of decline, or progression, observed within the placebo group, the trajectory of the placebo group helps to determine the treatment difference at the end of a clinical trial (Cummings, Ritter, & Zhong, 2018). Although the CREAD trials were stopped early for low likelihood of meeting the primary endpoint, inadequate progression in CDR-SB was not a contributing factor in either the prodromal or mild AD subgroups (Sink et al. Data presented at CTAD 2019). In addition, almost half of participants screened for CREAD failed early in the screening process because of not meeting the FCSRT inclusion criteria; thus, the cueing index may have helped to identify a population likely to decline and with higher rates of progression (Sink et al. Data presented at CTAD 2019). Incorporating inclusion criteria such as FCSRT helps to identify a trial population likely to decline and with higher rates of progression, which may improve the power of clinical trials to detect efficacy and may allow for more efficient trials in early AD. However, a potential caveat associated with enriching for progressors may be that these individuals are potentially also less likely to respond to therapy, especially with anti-Aβ monotherapy.
An additional strategy to improve efficacy findings by way of enriching study populations may be to exclude older patients from clinical studies. Age-related decrements across multiple cognitive domains, including memory, working memory/executive functions, regardless of AD status, are well established (Mormino & Papp, 2018). Furthermore, elderly patients with AD typically present with multiple comorbid neuropathological abnormalities unrelated to amyloid that further contribute to cognitive loss (Kawas et al., 2015;White et al., 2016). In addition, with increased age, the likelihood of multiple co-pathologies being present also increases, which may in turn contribute to the severity of dementia (Kawas et al., 2015). Together, the age-related cognitive decline and increased likelihood of comorbidities in older AD patients may mask any treatment-related efficacy signals, particularly within the short duration of traditional phase III studies.

| SUMMARY
In summary, targeting the potentially modifiable risk factors for AD does not appear to benefit individuals with manifest AD; however, targeting multiple AD risk factors with a multidomain intervention (e.g., a combination of diet, exercise, and lifestyle change) in asymptomatic, at-risk populations, may provide cognitive benefit to people at risk for AD and could delay the onset of dementia. Confirmation of this hypothesis with respect to dementia has yet to be evaluated.
Based on the continuum of AD progression, it is unlikely that the same drug or DMT will benefit all stages of disease; thus, selecting the "right" trial participant to use the "right" DMT at the "right" time along the continuum, and for the "right" duration, may be of importance in trials evaluating therapies for AD. Furthermore, biomarkers have demonstrated great value in assessing target engagement in different clinical trial populations and have helped to make it possible to identify individuals who are on the pathway to development of AD in the prodromal stage. Although the relationship between biomarkers and clinical efficacy is still under investigation, the updated analyses from the phase III EMERGE and ENGAGE aducanumab studies have generated evidence to examine the link between biomarker changes and clinical efficacy, although these data have not yet been presented or published. Lastly, applying clinical participant selection criteria to bolster signal detection in clinical trials can be predictably accomplished. Most of these advances have become apparent through learnings in AD clinical trials that did not meet their clinical efficacy goals, but nonetheless advanced the field.

| CON CLUS IONS
In conclusion, a lot has been learned over the years about what does and does not work in selecting treatments for evaluation, selecting participants for clinical trials, and measuring efficacy; but we are still not where we need to be in preventing and treating AD. While targeting amyloid accumulation in symptomatic stages of AD may not provide the full level of disease modification that is ultimately needed, it is likely to make a lasting difference in the progression of disease. The field will advance, not all at once, but through successful trials of stage-dependent treatments, probably delivering with small effect sizes as a beginning, which will then stimulate the level of investment needed to develop even more definitive approaches.
The treatment of AD will likely require combinations of therapies and brain protection strategies at every stage along the continuum of the disease.

ACK N OWLED G M ENTS
Medical writing support was provided by Angela Morris, PhD, and Helen Singleton, PhD, of Health Interactions, Inc., and funded by F.