Predicting Symptom Onset in Sporadic Alzheimer’s Disease: “How Long Do I Have?”

“HOW LONG DO I HAVE?”—A CRITICAL QUESTION

By the time individuals develop symptoms of Alzheimer’s disease (AD), they are already at an advanced stage of a disease that has silently progressed for over a decade or more. When individuals first present with mild forgetfulness due to AD, their brains are full of amyloid plaques, tau tangles, and extensive death of neurons has already occurred [1, 2]. Numerous attempts to slow the disease after symptom onset have either failed entirely or resulted in uncertain clinical benefit [3]. It seems that to effectively slow AD, interventions should be started very early, before extensive neuronal death has occurred, likely before the onset of the mildest symptoms of disease.

The Anti-Amyloid Treatment in Asymptomatic Alzheimer’s Disease (A4) study, which has completed enrollment of participants and is currently ongoing, was designed to do exactly this—to treat individuals with early AD brain changes while they are cognitively normal to slow or prevent the onset of AD symptoms [4]. To identify cognitively normal older individuals with the amyloid plaques characteristic of AD, an amyloid PET scan was performed, which was read as either positive or negative. Those who met inclusion criteria and were amyloid PET positive were randomized to either an anti-amyloid treatment (Solanezumab) or placebo. The study design involved disclosure of amyloid PET status to cognitively normal individuals and included aims to examine how the disclosure of these results affected individuals.

As a dementia specialist at one of the many sites participating in the A4 study, I disclosed amyloid PET status to individuals during the enrollment phase of the trial. Many of those interested in the study had seen the effects of AD dementia firsthand via an affected family member, and most worried about their own risk of developing AD dementia. Although I regularly give bad news in the form of dementia diagnoses to patients as part of their clinical care, this disclosure was different—the individuals receiving the news were cognitively normal. The A4 study provided training and a protocol for how to approach this very sensitive topic [5]. The key message I relayed to individuals with a positive amyloid PET scan was that they had elevated brain amyloid, but this did not necessarily mean they would experience cognitive impairment or develop AD dementia. For individuals with a negative amyloid PET scan, the message was that they did not have elevated brain amyloid at this time, but at some point in the future they could develop higher levels of amyloid and potentially develop AD dementia. Most individuals seemed to easily understand these messages and could repeat back the key points. They asked good questions such as, “How sure are you that I am positive (negative)?” and “How positive (negative) am I?” Unfortunately, the dichotomous nature of the amyloid PET scan result made these questions difficult to answer [6].

Interestingly, the most consistent question I received from individuals who were amyloid PET positive was some version of, “How long do I have?” These individuals knew they had amyloid plaques and were at higher risk of developing AD dementia but found increased risk an abstract and unsatisfying concept. They wanted to know when they personally would develop symptoms. This makes sense—developing AD symptoms in two years is a completely different matter to contemplate than developing symptoms in fifteen years. These individuals understood that the timing of AD symptom onset could completely change the meaning of their positive results. This question also is critical to clinical trials, especially trials like A4. Treatments can only be demonstrated to be effective if those in the placebo group develop cognitive decline over the study period. If most participants in the placebo group remain cognitively normal over the ∼3–5 years of the trial, a much larger cohort is needed to see potential beneficial effects of the treatment, increasing the time and cost of the trial. Thus, answering the question of “How long do I have?” is not only important to individuals, but key to finding treatments that slow onset of AD symptoms.

USING THE AMYLOID CLOCK TO PREDICT SYMPTOM ONSET

Unlike in clinical trials or in the clinic, where amyloid PET scans are read as positive or negative, in research studies the amyloid PET signal is precisely measured on a continuous scale. AD researchers have known for almost ten years that brain amyloid burden, as measured by amyloid PET, increases relatively consistently during the preclinical phase of AD when amyloid can be detected but symptoms have not yet manifest [7, 8]. In fact, amyloid accumulates so consistently that it can be used like a clock. The clock starts at a low level of amyloid accumulation, at a tipping point after which the clock ticks consistently until around the time of symptom onset [9]. Amyloid accumulation as measured by amyloid PET behaves somewhat like tree rings, with a new layer of amyloid being added each year. By scanning individuals multiple times over many years, several groups have developed methods to estimate the typical increase in amyloid PET signal each year [7, 9–11]. The primary conclusion from this work is that amyloid plaques accumulate in the brain for 10–20 years before the onset of AD dementia symptoms.

Given that the amyloid clock ticks consistently during the preclinical phase of AD, can we use the amyloid clock to estimate when a cognitively normal individual is likely to develop symptoms of AD dementia? Recent work suggests that the answer is “yes,” that it is possible to estimate when an individual will develop AD symptoms. Using different mathematical approaches, cohorts, and amyloid PET methods, findings from several groups have converged: the amyloid clock can be used to estimate risk for cognitive decline and progression to AD dementia [12–14]. Further, the age that an individual reaches a tipping point in amyloid accumulation can be estimated using the amyloid clock, and this age is highly correlated with the age of AD symptom onset [13]. Multiple groups are also examining how best to use fluid biomarkers to create clocks for AD. Although this is an active area of research that is still evolving, it does seem that it will be possible to estimate the age of AD symptom onset with a degree of accuracy that will be useful to clinical trials.

In a seeming contradiction to the amyloid clock concept, the quantity of amyloid plaques in the brain does not correspond well to dementia severity, which has been a major criticism of the amyloid cascade hypothesis of AD [15–17]. Amyloid plaques are necessary but not sufficient to cause AD dementia, as illustrated by the many cognitively normal older individuals who have brains full of amyloid plaques. In fact, the quantity of tau tangles appears to be better correlated with symptomatic AD than amyloid plaque burden [18, 19]. However, the amyloid cascade hypothesis proposes that amyloid initiates the cascade, not that amyloid is the proximate cause of dementia. Careful scrutiny of the amyloid clock shows that amyloid accumulates non-linearly, and that it stops keeping time around the time of symptom onset [13]. In fact, around the time of AD symptom onset, amyloid accumulation reaches a plateau and even declines in some individuals [9, 13]. Further, older individuals are more likely to develop AD symptoms at lower levels of amyloid burden—the association between amyloid burden and symptoms is not constant and depends on age [13, 14]. It is also important to consider that AD symptom onset and dementia progression are related but distinct concepts. Therefore, while amyloid burden does not have a strong linear correlation with dementia severity in autopsy series, this does not contradict the notion that amyloid initiates a cascade of events that culminates many years later in AD dementia. With help from the amyloid clock, the initiation of the amyloid cascade can be linked to the onset of dementia symptoms, but this connection fades after symptom onset.

APPROPRIATE USE OF ESTIMATES OF SYMPTOM ONSET

As algorithms to predict AD symptom onset are developed, it will be important to consider their appropriate uses. At this early stage, showing that multiple methods in different cohorts reach similar conclusions establishes the feasibility of models to predict AD symptom onset [12–14]. It will also be important to test selected methods/algorithms across cohorts to determine whether models are likely to be generalizable. All models have error, and the error of the models should be quantified. In autosomal dominant AD, the estimated age of symptom onset typically falls within ∼5 years of actual onset, and even with this level of error the estimate improves the efficiency of clinical trials [20, 21]. Even if a model to predict AD symptom onset had systematic errors (e.g., the actual onset was typically two years later than the estimated onset), using the estimated age of AD symptom onset would likely still facilitate identification of high risk individuals, improving clinical trial efficiency. It is also important to consider that some small number of individuals will be outliers who develop symptoms much earlier or much later than expected based on the model. With a research study or clinical trial that enrolls hundreds of participants, the effects of outliers on the overall results will be outweighed by the rest of the cohort. Given that the potential pitfalls of estimating symptom onset can be relatively well managed in an observational research study or clinical trial, these settings seem appropriate venues for testing the accuracy and utility of models that estimate age at symptom onset.

However, what about the individual who wants to know, “How long do I have?” These individuals often want to know for purposes of long-term planning, and/or preparing themselves or their families emotionally for the onset of AD dementia. Would it be appropriate to disclose to an individual their estimated age of AD symptom onset? Notably, the AD research community has been moving towards disclosing more results to research participants, partly because some research participants have asked for their personal research results [22]. However, providing information that is well validated (e.g., amyloid PET status) seems much more reasonable than providing predictive estimates that have not yet been validated. Before considering disclosure of an estimated age of AD symptom onset, it would be important to have validated the model in multiple cohorts, including cohorts representative of the individual wanting results. For example, some AD biomarker levels may vary by self-identified race and/or other factors [23], and models created in one group (e.g., non-Hispanic Whites) may not be generalizable to other groups [24]. Further, genetic factors, medical comorbidities, and social determinants of health may influence the age of AD symptom onset, and the effects of these factors must be examined in models of AD symptom onset [24, 25]. There must be high confidence in models of symptom onset before disclosure to individuals can occur.

If and when an individual’s estimated age of AD symptom onset is disclosed, it would be of paramount importance to emphasize the uncertainty of the estimate. In general, more precise estimates are likely to be more useful and comprehensible. For example, if a 70-year-old individual were told they are likely to develop AD symptoms at age 80 years, plus or minus five years, this range (75 to 85 years) is so broad as to not be very informative. In contrast, if the same 70-year-old individual were told they are likely to develop symptoms at age 80 years, plus or minus 3 years, this somewhat narrower range (77 to 83) may allow a degree of planning to occur. To emphasize the uncertainty of the estimate, it may be preferable to disclose the age range of estimated AD symptom onset (e.g., 75–85 or 77–83 years), and that some individuals develop symptoms before and after these ages.

The positive and negative effects of disclosing the estimated age of AD symptom onset to a cognitively normal individual needs to be assessed. Studies have shown that cognitively normal individuals typically cope well with disclosure of information that affects their risk for developing AD dementia, such as APOE genotype [26]. Individuals also cope well after learning their amyloid PET status [27]. However, compared to the abstract concepts of elevated risk or elevated brain amyloid, which are associated with unclear consequences, an individualized estimate of the age at AD symptom onset is a relatively concrete concept that predicts an event in that person’s life. Because it provides some specificity, an estimate of the age at AD symptom onset may be more likely to impact decision making. Some individuals may use this estimate in a constructive manner, such as integrating this information into their financial management plan and/or major life decisions such as when to retire or move. It may also increase motivation for healthy lifestyle habits like regular exercise. However, there could be adverse consequences, such as an individual near their estimated age of AD symptom onset who becomes overly critical of normal memory lapses, or unnecessarily curtails more demanding activities. Anxiety and depression about a predicted onset of AD symptoms may by itself create cognitive impairment. Disclosing increased risk for AD dementia has not been clearly shown to have major adverse consequences to individuals [26, 27], at least over the short-term, but it is unknown if a specific estimate of the age at AD symptom onset would have more profound effects.

Estimating the age of AD symptom onset appears to be feasible, as evidenced by converging findings from several approaches and cohorts. It is currently unclear how precise these estimates will become, or if a model can be developed that accurately estimates the age at AD symptom onset for a large proportion of the older population. However, it seems likely that these models will enable improved identification of cognitively normal individuals at high risk of developing AD symptoms within a given time interval, which would increase the efficiency of clinical trials. Currently, models are under development and disclosing the estimated age of AD symptom onset to individuals is not recommended until the models are well validated. As the models are refined, we must also learn more about the implications of disclosing estimates of the age of AD symptom onset to individuals who want to know, “How long do I have?”

DISCLOSURE STATEMENT

The author’s disclosure is available online (https://www.j-alz.com/manuscript-disclosures/21-5722r1).