Statistics from Altmetric.com
Existing models fail to account for the complexity and multidimensionality of new data and the growing scientific understanding of neurodegeneration.
With the acceptance that the disease processes leading to dementia begin at least as early as mid-life, there is now a major drive to understand and model this period of ‘disease before dementia’ for a variety of reasons. First, to gather a better understanding of the interplay between disease and risk or resilience factors that have been identified epidemiologically (eg, what is the mechanism by which exercise protects the brain) is necessary. Second, to develop algorithms for use in clinical practice to forecast changes in brain health that may lead to dementia that will underpin the design of tailored risk modification interventions and ‘personalised prevention plans’. And third, with an improved knowledge of the genesis and sequence of neuropathological changes in the predementia phase, there will be an expansion of and assurance in the new targets for pharmacological interventions solely or in combination to perturb the most upstream, and therefore perhaps modifiable stage of the disease.
The review by Hou et al 1 ‘models for predicting risk of dementia’ highlights in a detailed and authoritative way how recent work has tried to address the second objective of risk models, that is, the development of an algorithm to identify individuals at higher risk of dementia. The authors made several important conclusions including the acceptable predictive ability of existing models, and the fact that only a handful of the models have been externally validated. The review further evidences the limitations of the historical approach on risk modelling to date. Fundamentally, risk modelling has been based on what was known about the disease at the time the research was conducted that is, the research questions in the original papers were ‘does X predict future dementia?’ rather than ‘what predicts future dementia?’ The latter question can only now really be addressed with advances in complex statistical modelling (Proust-Lima et al,2 2009) and data sciences (ie, machine learning (Pellegrini et al,3 2018)) and on developments which are providing vast amounts of well-phenotyped high quality data from individual (eg, the European Prevention of Alzheimer’s Dementia (EPAD) project (Ritchie et al,4 2016)) and collective infrastructures facilitating access to datasets (eg, via the Collaboration on Alzheimer’s Prevention).
A further challenge in the disease modelling space is the fact that models need to predict an outcome—but what if that outcome is in itself a fuzzy concept for example, a dementia syndrome or mild cognitive impairment? These purely phenomenological concepts are so loose in their tethering to a disease state that any modelling, regardless of their inclusion or not of biomarkers (Ritchie et al,5 2014), will likely lead to inconsistent and spurious findings. To address these challenges, models should focus on biological outcomes (eg, a % increase in amyloid burden in mid-life) and should not be restricted to cross-sectional analyses of a few a priori chosen factors. Rather, improved dementia risk prediction models should incorporate data on (i) fixed and modifiable risk factors, (ii) early evidence of disease through biomarker analysis, (iii) evidence of disease signalled via multiple clinical/cognitive changes and (iv) a dynamic perspective of risk factors and outcomes. This approach to disease modelling—the ‘four-factor modelling approach’ is being undertaken in the EPAD and other aligned projects, and in doing so, will provide crucial and reliable evidence to help achieve the three objectives of modelling. Clarity on why we are modelling dementia risk and what we mean by this has not been hampered historically because of previous lack of insights into preclinical disease and not having the right data available. It is now critical to help advance the field to truly understand disease before dementia, forecast the likelihood of neurodegeneration (onset and progression), and ultimately have the empirical knowledge to hand to inform and deliver prevention.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient consent Not required.
Provenance and peer review Commissioned; internally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.