Individualized clinical management of patients at risk for Alzheimer's dementia

1 Introduction

Late-life Alzheimer's disease (AD) develops over an extended preclinical period [1]-[4]. Considering over 46 million people in the United States alone have preclinical AD, this predementia period offers a unique opportunity for early intervention to address modifiable risk [5].

Given the paucity of effective AD treatments, prevention or delay of dementia is essential. Furthermore, AD drug trials may have been more successful if initiated earlier in the disease course [6]. It is therefore important to evaluate the effectiveness of AD interventions across the disease spectrum, especially in at-risk individuals before clinically evident decline.

Population-attributable risk models estimate that risk factor modification (e.g., hypertension, insulin resistance, physical inactivity, hearing loss, depression) may prevent up to one-third of AD cases [7], [8]. These targetable risk factors may influence AD pathological pathways (e.g., glucose hypometabolism, inflammation, oxidative stress, amyloid burden, trophic factors) [8], [9]. The Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability (FINGER) study was the first large long-term randomized controlled trial (RCT) showing multidomain interventions (nutrition/physical activity/cognitive training) can maintain cognitive function and reduce the risk of cognitive impairment among at-risk older adults from the general population [10], [11]. Other RCTs applying lifestyle modifications have demonstrated similar effects in mild cognitive impairment (MCI) participants and adults at-risk for cognitive decline [12], [13]. However, encouraging data from RCTs require translation to clinical practice, including verification of how patient compliance (or “dose response”) affects outcomes [14].

Considering the heterogeneity of AD pathology, the application of precision medicine allows for interventions that can be targeted for individual patients [12], [15]. The National Institutes of Health defines precision medicine as “an emerging approach for disease treatment and prevention that takes into account individual variability in genes, environment, and lifestyle for each person”. [16]. An overall structure of how precision medicine may be achieved in the future will be through convergence of technological advances (e.g., big data, genomic sequencing, “-omics” technologies, systems biology, integrated disease modeling) as it is hypothesized that deconstructing the disease into multiple subsets that exist within a heterogeneous population, and tailoring therapies accordingly, may be preferentially effective based on individual biological make-up (protein-protein interactions, epigenetic modifications, metabolic pathways) [17], [18]. A term that has been used to adapt this approach, using currently available clinical assessments in everyday practice [19], is clinical precision medicine, where medical history (e.g., lifestyle patterns, life-course events), physical/neurological examination, anthropometrics, commercially available blood biomarkers (including genetics), and cognitive assessments inform a multimodal management plan [20], [21]. Patients are followed up longitudinally to evaluate the effectiveness of, and further refine, personally tailored interventions. In 2013, an Alzheimer's Prevention Clinic (APC) was established in New York, with research collaboration in Puerto Rico [21], [22]. APC's mission is to mitigate late-life AD dementia risk by applying individualized clinical management strategies toward primary, secondary, and tertiary AD prevention while simultaneously studying its comparative effectiveness (Supplementary Fig. 1) [23].

In this proof-of-concept trial, we investigated effects of multidomain evidence-based individually tailored interventions on cognition, AD/cardiovascular risk scores, and AD-risk biomarkers in real-world clinical practice [22], [24].

2 Methods

2.2 Procedures

Participants underwent a comprehensive screening evaluation: detailed clinical history, physical examination, anthropometrics, blood biomarkers, apolipoprotein E (APOE) genotyping, and cognitive assessment (Supplementary Table 1 and detailed in prior publication) [22]. Additional assessments were ordered in symptomatic patients (incorporating American Academy of Neurology Guidelines [25]), when indicated. Amyloid positron emission tomography (PET) or cerebrospinal fluid biomarkers were used to confirm/exclude AD pathology in participants with a clinical diagnosis of MCI or early mild dementia. Participants were diagnosed as normal cognition, subjective cognitive decline, preclinical AD, MCI due to AD, or early mild AD dementia incorporating the 2011 National Institutes of Health and the Alzheimer's Association diagnostic criteria (Appendix C) [[1], [22],[26], [27]].

Enrolled participants were given individualized, multidomain intervention recommendations informed by clinical and biomarker data (methods previously described) [22], and received a mean of 21 recommendations by a neurologist or family nurse practitioner (Fig. 1). Categories of recommendations included patient education/genetic counseling, pharmacological approaches (medications/vitamins/supplements), nonpharmacological approaches (customized recommendations for exercise, nutrition, vascular risk, sleep, cognitive engagement/training, stress, general medical care), and others based on methods previously published [22]. Longitudinal follow-up occurred every 6 months with continual refinement of interventions for each participant.

On follow-up, each participant was assessed as “compliant” or “not compliant” with each individual recommendation. A compliance score was calculated as a percentage of recommendations adhered to on a scale of 1-10 (1 represents 0-10% of recommendations, etc.) as independently assessed by two clinicians based on patient report at the visit and patient Likert scale ratings. Clinicians then assigned an overall compliance score by consensus before review of any follow-up data. Higher-compliance participants were prespecified as following >60% of all recommendations given, versus lower-compliance participants (≤60%) [28].

As an example of the application of the previously published method of an individualized clinical approach, a perimenopausal 59-year-old woman (apolipoprotein E4 [APOE ε3/ε4] heterozygote) without subjective cognitive complaints and a past medical history of untreated “borderline” hypertension (~140s/80s), hyperlipidemia and abdominal obesity, elevated waist-to-hip ratio (.93), elevated visceral body fat, insulin resistance, elevated homocysteine, and normal (albeit suboptimal) memory function received 25 individualized recommendations [22]. These included patient education about potential risks/benefits of long-term hormone replacement therapy, genetic counseling, referral to a preventative cardiologist for blood pressure control (goal 120s/70s) and consideration of a coronary calcium scan for cardiovascular risk stratification, exercise counseling including a targeted amount/type of aerobic-versus-resistance training (geared for body-fat reduction), nutrition advice centered on Mediterranean-style diet (emphasis on fatty fish and extra-virgin olive oil consumption to address elevated LDL and low HDL-cholesterol), while limiting high-glycemic foods (considering insulin resistance) and optimizing B-complex (B12/folate/B6) vitamin intake (considering elevated homocysteine) and cocoa flavanols (considering insulin resistance, elevated blood pressure, and lower-than-expected memory performance), as well as several other detailed recommendations such as sleep hygiene, cognitive engagement/training strategies, stress management, ongoing care with primary care physician (Fig. 1), and information on AD prevention clinical trials which she may soon qualify for based on age/genotype [22]. An introductory course on AD prevention (10 lessons, 2+ hours of interactive-multimedia content) that has been shown to increase knowledge and willingness to participate in AD prevention clinical trials is also recommended via the online learning portal AlzU.org [29]. On follow-up, she was given a compliance score of 8 based on clinical consensus, and was thus classified as a higher-compliance Prevention participant (based on following 71-80% of the 25 recommendations).

Adverse events were recorded during each follow-up, with the treating clinician asking all participants whether they experienced any side effects/harm related to assigned interventions. Trial registered at ClinicalTrials.gov (NCT03687710).

2.3 Outcomes

The primary outcome was change in performance on the modified Alzheimer's Prevention Initiative Cognitive Composite (m-APCC) from baseline to 18 months [30]. Statistical comparisons were performed between higher- and lower-compliance groups within each diagnostic classification and against matched historical control cohorts: National Alzheimer's Coordinating Center (NACC) and Rush University Memory and Aging Project (Rush) (Fig. 2).

The original APCC was empirically determined to document progression of preclinical cognitive decline related to AD progression, and was selected due to its concurrent use in two AD prevention clinical trials (Alzheimer's Prevention Initiative Generation Program, Autosomal-Dominant AD Trial) [31], [32]. Similar to other trials [33], [34], we refined the APCC based on the selection of tests administered (Supplementary Table 2 and prior publication of neuropsychological measures used in our clinic) [24]. Tests comprising the m-APCC were selected to represent the same cognitive domains as those used in the APCC [24].

Secondary outcomes included changes on a composite of neuropsychological tests associated with nonpathological cognitive aging (CogAging, Appendix D), two AD risk scales (Australian National University–AD Risk Index [ANU-ADRI], Cardiovascular Risk Factors, Aging and Incidence of Dementia [CAIDE]), two cardiovascular risk scores (American College of Cardiology/American Heart Association [ACC/AHA], Multi-Ethnic Study of Atherosclerosis [MESA]), and risk biomarkers (Supplementary Table 1) [35]-[38].

See Supplementary Table 1, 9–12/Appendix E for exploratory outcomes/results.

3 Results

3.3 MMRM for primary outcome—m-APCC

3.3.1 Compliance by diagnosis group (Prevention vs. Treatment) interaction

In Prevention participants, higher- and lower-compliance groups showed significant improvements by 4.6 (95% CI = 3.09–6.19, P < .0001) and 4.5 (2.24–6.84, P = .0002) points on the m-APCC, respectively. There was no difference between these groups (−2.79 to −2.61, P = .9488) (Fig. 3).

In Early Treatment participants, the higher-compliance group increased by 4.8 points but this was not significant (−1.06 to −10.67, P = .1073). The lower-compliance Early Treatment group had significant worsening by 6.0 points (−10.83, −1.20, P = .0148). The difference between these groups (10.8 points) was significant (4.67–16.97, P = .0007).

3.3.2 Historical comparison for the primary outcome

The higher-compliance Prevention group improved more than NACC by 3.1 (1.14–5.06, P = .0012) and Rush by 4.9 (2.55–7.25, P < .0001). The lower-compliance Prevention group improved more than NACC by 2.9 (0.74–5.06, P = .0088) and Rush by 4.0 (1.26–6.74, P = .0055) (Table 2).

The higher-compliance Early Treatment group improved more than NACC by 10.3 (5.99–14.61, P < .0001) and Rush by 5.3 (0.20–10.40, P = .0428). Lower-compliance Early Treatment did not differ from NACC (P = .9820) or Rush (P = .1115).

See Supplementary Table 6 for additional analyses matching our amyloid-confirmed participants to enriched NACC participants who were APOE ε4 carriers.

See Supplementary Fig. 4 for additional details.

4 Discussion

To our knowledge, this is the first empirical trial in a clinical setting indicating that individualized AD risk factor management may improve cognitive function which may be related to AD pathology. In addition, secondary analyses demonstrated that multidomain tailored interventions may reduce calculated AD and cardiovascular risk scores across a broad range of ages and diagnostic classifications, and may potentially have a cognitive-aging–modifying effect on nonpathological age-related cognitive decline. Within the Early Treatment group, cognitive improvements were seen only in the higher-compliance group, suggesting that close adherence to the interventions is needed to derive benefit within the context of definitive AD pathology. However, cognitive improvements were seen in both the higher- and lower-compliance Prevention participants, with both compliance groups demonstrating improvements compared with historical cohorts. Furthermore, our population was easily and quickly recruited from a real-world clinical setting and the interventions were well-tolerated, adding to the translational value for practice (Appendix J). In addition, while socioeconomic factors will differ among varied cohorts, intervention-related costs negatively impacted adherence in only 7.1% of participants (Appendix K).

Because m-APCC measures AD-related cognitive change, improvements may have resulted from targeting risk factors that lead to AD pathogenesis; however, direct evidence of changes in these pathways was not obtained. Additional evidence from longitudinal volumetric magnetic resonance imaging, fludeoxyglucose-PET and amyloid-PET would demonstrate whether observed improvements were related to disease modification. Neuroimaging data are forthcoming from a brain imaging substudy (n = 135) begun in 2018 [9]. Furthermore, longitudinal measurement of potential key AD-related biomarkers, such as those related to neuroinflammation and synaptic dysfunction, may be incorporated into future studies to investigate the direct effects on AD pathology.

Cognitive aging composites indicated that the estimated delay of cognitive decline may be approximately three years in Prevention participants and two years in the higher-compliance Early Treatment group. Improvements in cognitive decline related to nonpathological cognitive aging may potentially be linked to reducing vascular dementia risk and/or targeting other factors (e.g., synaptic plasticity, alterations in neuronal structure, dysfunction of neuronal networks) [43]. However, owing to lack of cognitive aging biomarkers, biological factors related to this potential response were not measured and are thus unclear. Treatment effects observed using both cognitive composite measures may suggest that treatment response is broad. Therefore, addressing risk factors that collectively impact overall health may assist in mitigating age-related cognitive decline, along with other potential health benefits stemming from treating comorbidities (e.g., cardiovascular risk).

We observed improvements in several AD-risk biomarkers. In Prevention participants, improvements were found in HDL-C, hs-CRP, adiponectin, and 25-hydroxy-vitamin D. In Early Treatment participants, fibrinogen, homocysteine, HDL-C, hs-CRP, adiponectin, and Lp(a) improved. However, none of these correlated with improvement in cognition. Unexpectedly, a worsening of cystatin C of 0.1 mg/L corresponded to an improvement in CogAging. One possible explanation for lack of correlations between biomarkers and cognition is that such relationships likely involve multiple biomarker changes that may vary by person, as well as varied baseline values within a broad spectrum of reference ranges. A Bayesian hierarchical analysis also did not identify an individual biomarker or category of biomarkers that was primarily associated with observed cognitive changes. See Appendix H/I for discussion.

While further study incorporating a host of biomarkers pre- versus post-intervention may help to inform causality, we observed changes in efficacy outcomes such as serum biomarkers, anthropometrics, and risk scales (Supplementary Fig. 6). These changes, along with the comparison of compliance groups, correction of multiple covariates, and matched historical comparisons, may suggest that these findings were potentially driven by the prescribed interventions.

Traditionally, treatment trials have attempted to isolate one effect at a time for single interventions, but the complexity of AD may require targeting multiple mechanisms simultaneously to affect disease progression. Our initial evidence of broad effects across risk scales, and measurements of cognition and biomarkers changing in expected directions, suggest this approach warrants further rigorous study.

Our study has several limitations. Our key limitation is the lack of a concurrent, randomized control group. Two considerations led to this design. A true control group may not have been possible because well-informed participants actively enrolled in an AD risk reduction study may seek out and make lifestyle and/or other behavioral changes that impact outcomes. In addition, given the setting of a real-world clinical practice where patients seek AD risk reduction care for modifiable risk factors in a clinic outside of a traditional solely research environment, it would not be feasible to withhold treatment from a nonintervention randomized control group.

The disadvantage of an uncontrolled study is that it is unclear whether observed effects are due to baseline characteristics of participants or other aspects of general study conduct unrelated to treatment. In an attempt to mitigate these effects, we corrected the model for baseline predictors of compliance by including them as covariates to better ensure the improved outcomes were not primarily due to baseline characteristics. We also used historical comparison cohorts with similar demographics and matched each participant based on age and baseline m-APCC. We used historical comparisons to also help account for study procedure effect, such as practice effects. Compared with these matched historical comparisons, participants demonstrated greater improvements at similar time-points. Because these historical comparison groups were not part of any intervention, a response associated with intervention expectations may potentially explain part of the cognitive benefit that was observed in our study. However, the 18-month duration is longer than is usually expected for this type of effect. Furthermore, improvements found in laboratory biomarkers and AD and CV risk scales are less likely to be influenced by placebo effects [44]. Future studies which include randomized nonintervention groups would allow for more definitive conclusions.

Because few NACC participants and no Rush participants had available amyloid biomarkers, we were unable to match on confirmed AD pathophysiology. After updating the matching algorithm to include APOE ε4 positivity as an enrichment strategy for NACC participants, our cohort continued to show cognitive benefit. While the lack of amyloid biomarkers is an important limitation, we would have expected the rate of cognitive decline in historical subjects without amyloid-confirmation to be slower than a matched population with amyloid confirmation, resulting in a more conservative estimate of the intervention effect. Unexpectedly, when enriching for APOE ε4, we observed less decline in the enriched population.

Another limitation stems from the study environment of a real-world clinical practice and the challenges of rating compliance. There is a paucity of evidence on how to use compliance in comparative effectiveness studies as an outcome to differentiate treatment effects. While some studies have defined high compliance as following two-thirds of prescribed recommendations, we selected 60% for two reasons [45], [46]. An initial motivation was that a cut-point of 60% led to a roughly even number of patients in higher- and lower-compliance categorizations when care was previously provided in the clinic (from 2013 to 2014 before initiating the comparative effectiveness study). Furthermore, we applied categorizations from a prior study quantifying compliance into 4 groups: noncompliant (compliant to treatment schedule less than 20% of the time), low (20% to 59% of the time), moderate (60% to 79% of the time), and high (≥80% of the time) compliance [28]. Based on this study framework, we divided our participants into higher (≥60%) and lower (<60%) compliance groups. See Appendix K/Supplementary Table 13 for additional information.

It is important to note that because lower compliance is often related to disease severity, statistical corrections for baseline m-APCC, HbA1c, homocysteine, AHA/ACC, and age were applied to decrease the possibility of bias because of these potential confounders. Furthermore, the separation of diagnosis and compliance groups was critical due to a strong compliance by diagnosis interaction effect.

While our sample size was modest and further stratification led to relatively small diagnostic and compliance groups, observed effects seen were of a large enough magnitude to still be detectable. Continued recruitment across additional sites globally (n = 1000 planned) will allow for confirmation of these proof-of-concept results and more detailed analysis of patient subgroups (e.g., age, ethnicity), biomarkers, and intervention approaches. Expanded recruitment may enable deeper understanding of precision effects and more definitive conclusions, and allow assessment of the impact of medical comorbidities and concomitant medications.

Practice effects due to repeated cognitive test exposure are another potential concern. To mitigate this, we administered alternate test forms at each time-point and required that participants complete simulated at-home cognitive assessments before baseline. This also primed participants to testing conditions/procedures and mitigated test anxiety in an effort to reduce practice effects. In addition, practice effects on cognitive measures tend to occur at briefer test-retest intervals than those involved in this study, and the comparison with historical cohorts who took related measures repeatedly demonstrated improvement beyond what can reasonably be explained by practice effects [47].

While the m-APCC was our primary outcome, there is no gold standard for which cognitive measures should be used (and how often), and the degree of benefit which should be expected [48]. Cognitively normal patients at risk may have a lower ceiling for benefit as they do not yet manifest cognitive decline. As such, assessment scales cannot be easily repeated from prior treatment trials, and novel composite measures sensitive to predementia decline may hold promise [24], [48]. Because the study was conducted in the real-world clinical setting and one of the treating clinicians left the practice because of geographical relocation, 24 participants (58.6%) were lost to follow-up for this reason. Future studies should consider safeguards to account for similar factors that can substantially influence discontinuation rate. However, because the major contributing factor to discontinuation would not be expected to be related to response to treatment, it may be less likely that loss of these patients introduces bias in our results.

Furthermore, while patients who seek risk reduction care tend to be highly motivated, this approach may not be as effective in patient populations with lower motivation. Factors related to compliance are detailed in Appendix K. Also, despite the study's translational value, long-term effects are unknown. Longitudinal assessments are ongoing. In addition, while the median age in our cohort was 61 years, and the mean age was 60 years, our cohort included a broad age-range due to younger, middle-aged, and older patient demand in a real-world clinical setting. Nevertheless, most participants (~75%) were aged 50 years or older. Age was included as a continuous linear covariate in the primary model, and as such, all estimated changes were for an average aged person (60 years old). The Prevention group had 0.1 points less improvement per year of age, and the Early Treatment group had 0.2 points less improvement per year of age. As such, an older population may demonstrate less improvement in cognition and, similar to AD drug trials, this intervention may be more effective in younger and/or less impaired populations. Future studies are warranted to more deeply understand age effects of this intervention.

Recognizing that more validation is necessary, we offer this framework as a potential approach for patient care while further clarifying its effectiveness (see Supplementary Fig. 7 to visualize levels of personalization). Given the magnitude of disease, significant morbidity of late-life dementia, and growing interest in applying preventative neurology to clinical care, it is important to report these findings as larger studies are developed and while our own sample size grows. Overall, these results help extend prior RCT/observational findings into a clinic setting where individualized lifestyle modifications produced measurable benefits.

5 Conclusion

We envision a day when individualized AD risk factor management may be applied for care to tens of millions of patients at risk for AD dementia. From a practical clinical perspective, there is ample evidence to support recommending established lifestyle changes known to benefit overall brain health. However, important challenges remain for researchers, clinicians, patients, and health policy decision makers on how best to evaluate—both objectively and ethically—any new information, findings and knowledge that promote and/or maintain brain health.

We have previously proposed that the field must pursue four strategic objectives: (1) more rigorous study of the comparative effectiveness of individualized risk management; (2) establish a consortium of clinician researchers who can apply and continually refine this framework for AD preventive care; (3) support the design of a large multisite international study to validate clinical effectiveness; and (4) advocate for public and private funding to move health services research into the realm of precision medicine clinical trials [22].

Although international consortiums and research networks, such as World-Wide FINGER, United States Study to Protect Brain Health Through Lifestyle Intervention to Reduce Risk (U.S. POINTER), the European Prevention of Alzheimer's Dementia initiative, and others, continue to drive awareness and research into the effectiveness of lifestyle changes on brain health and dementia risk reduction, there remains a critical need to establish global standards/harmonization around the science of comparative effectiveness research. It is our hope that international funding agencies and foundations will consider adoption of comparative effectivenes research as a prioritized strategic funding opportunity for 2020 and the new decade.

Supplementary data

Supplementary data related to this article can be found at https://doi.org/10.1016/j.jalz.2019.08.198.