Machine Learning Prediction of Minor Depression Diagnostic Based on Specific Serologic Pro-Inflammatory Cytokines Variation in Non-Demented Geriatric Subjects

Abstract

Geriatric depression is a major source of disability and reduced quality of life. Chronic inflammation plays an important role in depression pathophysiology and is often associated with ageing, as well. Our research objective is to evaluate how a trained machine learning algorithm may predict minor depression based on activation of specific pro-inflammatory cytokines, in geriatric subjects with mild cognitive impairment. From ADNI1 data, 38 subjects with a clear history of depression, were carefully matched with similar subjects without any history of depression and generated a “training group”. 252 subjects without any history of depression were included in the “test group”. A classic C4.5 decision tree algorithm used a supervised training approach to detect a pattern of cytokines change in relation with previous depression. 92 subjects from the “test group” were predicted to have a form of minor depression. The prediction was checked using clinical and biological independent markers. Age and a significant hippocampal volume reduction were observed in the positive sub- group.

In conclusion, a trained machine learning algorithm is able to predict depression in aged subjects with mild cognitive impairment based on specific cytokines variation analysis. As the prospective aspects of the study were limited and the number of subjects was low, no causal relationship between the cytokines’ changes and the evolution of minor depression in geriatric subjects can be drawn. Another limitation of our study is coming from the design of our study (secondary use of data that originally focused on cognition, with limited information concerning geriatric mood changes).

Note:
Funding declaration: Our study had no funding support to report.

Conflict of Interests: Our study had no conflict of interests to report.

Ethical Approval: ADNI data is open for public access. It is totally anonymized. The de-identification process is performed by data curators. Our study used only non-identifiable data following the Canadian ethics research provisions for secondary data use studies (TCPs 2(2018)). The data investigation research was performed in accordance with declaration of Helsinki principles.

Keywords: Keywords: Geriatric minor depression, C4.5 decision tree prediction, mild cognitive impairment, chronic inflammation, geriatric depression scale score, neutrophils to lymphocytes ratio