Robust and interpretable AI-guided marker for early dementia prediction in real-world clinical settings

Lee, L. Y., et al. (2024).

EClinicalMedicine, 102725.

https://doi.org/10.1016/j.eclinm.2024.102725

Background

Predicting dementia early has major implications for clinical management and patient outcomes. Yet, we still lack sensitive tools for stratifying patients early, resulting in patients being undiagnosed or wrongly diagnosed. Despite rapid expansion in machine learning models for dementia prediction, limited model interpretability and generalizability impede translation to the clinic.

Methods

We build a robust and interpretable predictive prognostic model (PPM) and validate its clinical utility using real-world, routinely-collected, non-invasive, and low-cost (cognitive tests, structural MRI) patient data. To enhance scalability and generalizability to the clinic, we: 1) train the PPM with clinically-relevant predictors (cognitive tests, grey matter atrophy) that are common across research and clinical cohorts, 2) test PPM predictions with independent multicenter real-world data from memory clinics across countries (UK, Singapore).

Interpretation

Our results provide evidence for a robust and explainable clinical AI-guided marker for early dementia prediction that is validated against longitudinal, multicenter patient data across countries, and has strong potential for adoption in clinical practice.

A pdf of the research can be found here.

Here is a summary and some thoughts:

Cambridge scientists have developed an AI tool capable of predicting with high accuracy whether individuals with early signs of dementia will remain stable or develop Alzheimer’s disease. This tool utilizes non-invasive, low-cost patient data such as cognitive tests and MRI scans to make its predictions, showing greater sensitivity than current diagnostic methods. The algorithm was able to correctly identify 82% of individuals who would develop Alzheimer’s and 81% of those who wouldn’t, surpassing standard clinical markers. This advancement could reduce the reliance on invasive and costly diagnostic tests and allow for early interventions, potentially improving treatment outcomes.

The machine learning model stratifies patients into three groups: those whose symptoms remain stable, those who progress slowly to Alzheimer’s, and those who progress rapidly. This stratification could help clinicians tailor treatments and closely monitor high-risk individuals. Validated with real-world data from memory clinics in the UK and Singapore, the tool demonstrates its applicability in clinical settings. The researchers aim to extend this model to other forms of dementia and incorporate additional data types, with the ultimate goal of providing precise diagnostic and treatment pathways, thereby accelerating the discovery of new treatments for dementia.

A Just Standard: The Ethical Management of Incidental Findings in Brain Imaging Research

Graham, M., Hallowell, N., & Savulescu, J. (2021). 

Journal of Law, Medicine & Ethics, 49(2), 269-281. 

doi:10.1017/jme.2021.38

Abstract

Neuroimaging research regularly yields “incidental findings”: observations of potential clinical significance in healthy volunteers or patients, but which are unrelated to the purpose or variables of the study.

From the Conclusion

Appealing to considerations of distributive justice provides an answer for these difficult cases. Data about a patient’s brain that may be generated by a neuroimaging scan in a research context is not something a healthy participant is entitled to as a matter of basic care. Accordingly, a researcher has no obligation to generate this information by performing additional scans (i.e., to “look” for incidental findings). Similarly, if a researcher discovers an incidental finding of unknown or uncertain clinical significance, they are not required to refer a participant for follow-up. Screening for brain abnormalities is not a requirement of basic care, and the burdens of follow-up on the health system (given the potential benefits) are inconsistent with distributive justice. This approach thus avoids the problem of trying to determine whether disclosure (is likely to) promote autonomy or benefit the patient. Rather, it requires researchers to ensure that participants are not deprived of anything to which they are entitled as a matter of distributive justice. This includes all of the protections to which participants in research are normally entitled, as well as the disclosure of clinically significant incidental findings.

The research is here.