Can Biomarker Data Predict Individual Risk for Alzheimer’s?

Scientists know that Alzheimer’s biomarkers such as amyloid plaques predict dementia, but most prior studies examined aggregate risk across groups of people. For an individual, biomarkers are still difficult to interpret because an abnormal result does not mean the disease is imminent. A new prognostic tool may add some clarity. In the October 16 JAMA Neurology, researchers led by Ingrid van Maurik at VU University Medical Center, Amsterdam, described statistical models that combine biomarker and demographic data to calculate how likely it is that a person with mild cognitive impairment will progress to AD or some other dementia. A model like this could help doctors make sense of biomarker results for their patients, van Maurik suggested. “It would provide more specific information than do group statistics, and help improve communication between clinicians and patients,” she told Alzforum.

  • A new statistical model estimates a person’s risk of AD.
  • The model requires standardized biomarker data.
  • It could be used to render a person’s prognosis.

Commentators said the approach was the next logical step in biomarker use. “This study points toward what I believe will be the future use of biomarkers,” said Henrik Zetterberg at the University of Gothenburg, Sweden. He noted that such a tool would not have been possible five years ago. Now, because of the increasing standardization of biomarkers, individual risk prediction might be within reach, he said.

Overall, people with elevated brain amyloid tend to develop Alzheimer’s dementia within a few years, especially if they also have tau tangles (Sep 2013 news; Apr 2016 news; Jun 2017 news). However, while this holds true in general, short-term outcomes vary greatly from person to person, with some people harboring plaques but staying mentally sharp for several years (Dec 2014 conference news). Among people who already have mild cognitive impairment, only some have amyloid in their brains and progress to dementia; how fast that occurs depends on other factors, such as cognitive reserve or vascular disease (May 2015 news; Feb 2017 news; Aug 2017 news). 

Van Maurik and colleagues wondered if they could develop a predictive model for a person’s progression risk. They analyzed data from 525 people with MCI and an average age of 67 who participated in the Alzheimer’s Biomarkers in Daily Practice (ABIDE) study based in Amsterdam (Aug 2017 conference news; de Wilde et al., 2017). The majority of the participants underwent structural MRI scans at baseline, as well as lumbar punctures to collect cerebrospinal fluid. Over an average follow-up of 2.4 years, about half the cohort progressed to a clinical diagnosis of dementia. For 201 participants, the diagnosis was AD, while 52 developed other forms of dementia.

The researchers constructed prognostic models that retrospectively calculated the risk associated with each of seven variables over both a one-year and a three-year span. They assessed for hippocampal volume, whole-brain volume, CSF Aβ42, CSF tau, age, sex, and MMSE score. When they considered only demographic data, older women with worse MMSE scores had the highest risk of progression to Alzheimer’s dementia, while younger men with better MMSEs had the least, as might be expected. Adding volumetric MRI data shifted these risk estimates, but the effect depended on age. A low brain volume raised risk for younger people, but not for older ones, perhaps because the brain shrinks as part of the normal aging process. On the other hand, a relative lack of brain atrophy predicted preserved mental function at all ages.

CSF markers had a more potent effect on Alzheimer’s dementia risk than did MRI. Abnormal Aβ42 and tau levels jacked up the odds at all ages, although the effect was greatest in younger people. Conversely, having normal CSF biomarkers slashed the odds of developing AD. For example, in this model, a hypothetical 75-year-old woman with an MMSE of 24 runs a 76 percent chance of progressing to Alzheimer’s dementia within three years based on demographic data only, but if her CSF markers are normal, her risk falls to 9 percent.

To create a prognostic tool, the authors combined all the factor-based predictions into a single statistical model. It takes a particular person’s data and calculates his or her percentage risk of progressing to Alzheimer’s dementia within one or three years, along with a 95 percent confidence interval. The model accepts MRI scan data as volumetric measurements or visual reads, the latter being more commonly used in the clinic, van Maurik noted. The model is available in spreadsheet format from the authors upon request.

Van Maurik and colleagues constructed a second model using dementia of any type as the endpoint. In it, hippocampal volume and CSF Aβ42 were less predictive of progression than they were for AD, while CSF tau, whole-brain volume, and MMSE scores dominated.

The authors evaluated the accuracy of these models using a Harrell C statistic, which compares all possible pairs of patients in a cohort to test whether people with the higher risk estimate actually progress more quickly. The Harrell C score for the AD model was 0.77, and for the all-cause dementia model, 0.70. These numbers indicate a robust model, van Maurik told Alzforum. The Harrell C statistic ranges from 0.5 to 1, where 0.7 or higher indicates a good model and 1 means the model predicts outcomes perfectly. In addition, the authors validated the findings in the ADNI-2 cohort, calculating similar Harrell C scores for the models’ performances in that population.

Commenters found the data persuasive. “A strength of the current study is the cross-validation procedure conducted in the same cohort and the external validation in the ADNI cohort,” Kejal Kantarci at the Mayo Clinic in Rochester, Minnesota, wrote to Alzforum (see full comment below).

At the same time, Kantarci and others wondered if adding other variables might improve the model. Kantarci suggested looking at the effects of ApoE genotype. Andrew Budson of Boston University noted that patients typically take cognitive tests beyond just the MMSE. “I wonder how much diagnosis would improve if additional cognitive testing was put into the model,” Budson wrote. PET scans for amyloid, glucose metabolism, and tau are notably absent, though the authors are now adding amyloid PET. In some clinics, especially in the United States, PET scans are more common than lumbar punctures. Because amyloid PET and CSF Aβ usually agree, future models could use whichever type of data was available, van Maurik said.

The model only works with biomarker data collected using the same methods as in the ABIDE study: Innotest ELISAs for the CSF measurements, and specific MRI software, namely FSL FIRST and SIENAX, for calculating volumetrics. Van Maurik is analyzing data from other cohorts that was gathered by different methods, such as automated CSF assays, to make the model more broadly applicable (Apr 2017 conference news). She also wants to turn it into an app clinicians can use on their smartphones or laptops. Zetterberg noted that cardiologists already use apps that run algorithms based on variables such as blood pressure and cholesterol to predict risk for cardiovascular disease.

AD clinicians said they need such an application. “It is still very hard to tell if a 65-year-old person with a family history of Alzheimer’s who complains of memory problems has the disease or not,” Zetterberg noted. Budson said that currently, he can only tell patients with accelerated brain atrophy that such shrinkage is more common in Alzheimer’s patients than normal agers. “Now, I may be able to give patients and families a different answer,” he wrote to Alzforum. Routine clinical use of the model may still be a few years away, however, contingent on further testing and validation.

Van Maurik noted that overall, people with MCI have a 50/50 risk of progressing to dementia. The current model can provide a better answer for about 80 percent of those people, returning a risk estimate that is either much lower or higher than 50 percent and giving patients a clearer idea of their future health outlook. She believes the approach could help select participants for clinical trials as well. “What the model does really well is filter out patients who won’t progress,” she said.—Madolyn Bowman Rogers

Comments

  1. This is a well-designed study to evaluate the risk of AD dementia and dementia in general in MCI patients, using biomarkers. The cohort of MCI patients was recruited from memory clinics and followed up to three years. Time-dependent analyses reveal the multimodal biomarker model for the prognosis of MCI, which may be utilized in clinical settings.

    A strength of the study is the cross-validation procedure conducted in the same cohort and the external validation in the ADNI cohort. Even though there are differences between the two cohorts, with the current cohort being memory-clinic based, and ADNI designed as a clinical trial-like cohort, the models were still quite robust in the ADNI cohort. 

    During the mean follow-up period of 2.4 years, 51.8 percent of the MCI patients remained stable. Longer follow-up of these MCI patients in this stable cohort will likely yield more converters, which may influence the models. Also, it would be interesting to study the effects of APOE e4 status on the models. Although APOE genotyping is not routinely used in the clinic, it is a well-known risk factor for AD dementia and other dementias, and not hard to obtain. 

    View all comments by Kejal Kantarci

  2. For over two decades, biomarker technology has been used in Alzheimer’s research. The ABIDE study is the first major project to explore the adoption of biomarker technology in clinical care of patients with mild cognitive impairment, and is therefore a major step forward for the field. Although the methodology has been validated in a separate, well-characterized cohort of subjects, the major hurdles going forward will be first, determining whether the predictive model generalizes from larger care settings to small practices, where the patient populations are heterogeneous, and second, determining whether the assays can be sufficiently standardized to make the model useful in such settings.

    View all comments by Jeffrey Petrella

  3. In my clinical practice I always go over the MRI scan with my patients, showing them the atrophy in the hippocampus, parietal lobes, or other regions. “Does that mean I have Alzheimer’s disease?” they ask. I then go on to explain to them that if you take 100 people with Alzheimer’s and 100 people aging normally, there will be more atrophy in the group with Alzheimer’s than the group aging normally, but that you cannot predict anything for sure in an individual patient. Now, I’ll be able to give patients and families a different answer. Van Maurik and colleagues have done a study to try to use atrophy on MRI scans, and CSF Aβ and tau, to predict the likelihood that a given patient with mild cognitive impairment (MCI) will develop dementia due to Alzheimer’s or another disorder, one and three years in the future.

    On the one hand, the article is trying to be helpful for front-line clinicians. But on the other hand, the main paper describes brain volumes calculated with programs that are not standardly available. The supplemental tables do list the results calculated with visual estimates, which is what clinicians would use. Additionally, “validated semiquantitative visual rating scales” are referenced but otherwise not described, so anyone who wants to use these scales with tables or calculators would need to go and pull the papers themselves.

    One thing that seems quite odd, at least to this clinician, is the thought that someone would subject patients to a lumbar puncture for CSF and only do an MMSE as their neuropsychological testing. Yet that appears to be the only cognitive testing that the patients undergo (or is, at least, included in the model). One can hope that if patients underwent at least a modest amount of additional cognitive testing the prediction from the clinical data alone would be vastly improved. Given this fact, I do wonder how much the biomarkers would improve diagnosis if additional cognitive testing was put into the model.

    The authors created a calculator they say will be given freely upon request for academic use. I have received the calculator from the corresponding author but have not yet tried it, so I cannot comment on its usefulness or ease of use. Overall, the paper does represent a step forward in translating purely academic research on the relationship between atrophy, Aβ, tau, MCI, and the development of dementia into a clinical realm.

    View all comments by Andrew E. Budson
    • User Profile Image Samantha Burnham
      Commonwealth Scientific and Industrial Research Organisation
    • Posted:

    This is an interesting study by van Maurik and colleagues that proposes an algorithm for calculating an individual's risk of progressing from MCI to an AD diagnosis in one or three years. The models highlighted a negative predictive value for MRI and CSF markers, reiterating their use to screen out individuals from clinical trials and intervention studies. Interestingly, gender appeared to have little impact on the risk of progression to AD. Whilst this represents an interesting method for ascertaining the risk of progression to AD, caution must be taken: The models are based on a relatively small number of individuals (575), with a relatively short follow-up (mean of 2.4 years).

    View all comments by Samantha Burnham

Make a Comment

To make a comment you must login or register.

References

News Citations

  1. Paper Alert: Preclinical Alzheimer’s Stages Predict Progression
  2. CSF Tau Rivals Aβ for Predicting Cognitive Decline
  3. At Risk, or Already Alzheimer’s? Elevated Aβ Predicts Cognitive Decline
  4. Large Studies Agree: Brain Amyloid Accelerates Cognitive Decline
  5. Cognitive Reserve—More Evidence It Prevents Neurodegeneration
  6. Being Bilingual Buffers Against Alzheimer’s by Improving Connectivity
  7. Vascular Problems in 40s, 50s Beget Dementia Down the Road
  8. In Clinical Use, Amyloid Scans Change Two-Thirds of Treatment Plans
  9. Are CSF Assays Finally Ready for Prime Time?

Paper Citations

  1. de Wilde A, van Maurik IS, Kunneman M, Bouwman F, Zwan M, Willemse EA, Biessels GJ, Minkman M, Pel R, Schoonenboom NS, Smets EM, Wattjes MP, Barkhof F, Stephens A, van Lier EJ, Batrla-Utermann R, Scheltens P, Teunissen CE, van Berckel BN, van der Flier WM. Alzheimer's biomarkers in daily practice (ABIDE) project: Rationale and design. Alzheimers Dement (Amst). 2017;6:143-151. Epub 2017 Jan 23 PubMed.

External Citations

  1. Harrell C score

Further Reading

Primary Papers

  1. van Maurik IS, Zwan MD, Tijms BM, Bouwman FH, Teunissen CE, Scheltens P, Wattjes MP, Barkhof F, Berkhof J, van der Flier WM, Alzheimer’s Disease Neuroimaging Initiative. Interpreting Biomarker Results in Individual Patients With Mild Cognitive Impairment in the Alzheimer's Biomarkers in Daily Practice (ABIDE) Project. JAMA Neurol. 2017 Dec 1;74(12):1481-1491. PubMed.

Annotate

To make an annotation you must Login or Register.