Editorial: Advancing biostatistics and informatics applications in mental health research

Authors:
Open Access
Publisher website
Google Scholar
Abstract
In the history of biostatistics application in research, researchers typically start with a question or hypothesis and propose a data model. In contrast, recently emerged machine learning methods, often start with the information already available and propose hypothesis using data as the starting point. These represent the "twocultures" paradigms of statistical modelling described by Breiman (2001) as data modelling and algorithmic modelling. The conventional data modelling approach perceives data generated from an underlying probabilistic structure and focuses on estimation, interpretation, and inference. The algorithmic modelling assumes that data may not arise from an easily described specified model. It uses an algorithm to predict the response and is evaluated by measuring predictive accuracy.When the informatics era came, we have a modern problem and the excitement of growing data in terms of its dimensionalities and volume. The growth of informatics has intensified the importance of both modelling approaches. Researchers now routinely work with high-dimensional and multimodal data, including genomics, neuroimaging, wearable sensors, electronic health records, administrative databases, and natural language text. While traditional statistical methods remain critical for causal inference and evaluation in clinical and epidemiological studies, algorithmic approaches provide powerful tools for uncovering complex relationships within increasingly large and interconnected datasets. The two-culture principle suggests that we should search for the model that provides the best solution-data modelling, algorithmic modelling, or both.The current volume of the topic "Advancing Biostatistics and Informatics Applications in Mental health Research" brings together twelve articles of five themes that collectively demonstrate the expanding role of advanced analytics in mental health research.Several studies demonstrate the value of multivariate approaches for characterizing complex mental health phenomena.Chen utilized network structure learning, an algorithmic modelling method, to explore the interrelations between major depression symptoms and ambivalence over emotional expression (AEQ) and compared the symptom networks between male and female college students. They identified two clusters of comorbid symptoms in the depression-AEQ networks, discovered the bridging symptoms between two clusters and examined the gender difference in network and centrality. Notably, the study combined network methods with statistical validation procedures, illustrating productive integration of algorithmic and data modelling approaches.Youvalari and colleagues provided a profiling analysis of resilience and demonstrated the potential of machine learning to enrich psychological Modelling. Grounded in the five-dimensional (5D) resilience framework and used unsupervised algorithmic modelling, the authors identified hidden psychological patterns clustered into 4 archetypes (Reactive Idealists, Fragile Strivers, Hidden Reactors, Stable Withdrawers). An important innovation of the study is to combine a deep autoencoder for feature extraction, with Gaussian Mixture Model clustering; and provided narrative description of archetype profiles.Lee and colleagues proposed digital phenotyping by integrating wearable Fitbit data, SOMDAY application, neuroimaging data (fNIRS), and genomic information. The periodical nature and activities of sleep was used for the multivariate pattern analysis and identification for digital phenotypes. The analytical framework conceptualized integrating the genetic vulnerability (polygenic risk scores), molecular state (gene expression), and physiological outputs (biomarkers) in the context of environmental and behavioral modulating factors captured through digital phenotyping.The New Zealand study by Zeng and colleagues similarly demonstrated the value of analytical triangulation. By combining data modelling using quantile regression, path analysis, with algorithmic Bayesian network modelling within complex survey data, the authors identified consistent associations between mental wellbeing and factors such as perceived health, life satisfaction, cultural identity acceptance, and trust in educational institutions. The study also highlights methodological advances in calibrating complex survey designs through replicate-weight approaches.Advances in natural language processing (NLP) and large language models (LLMs) are increasingly influencing mental health research and clinical practice. Goga and colleagues conducted texts analysis using social media discussions and semantic techniques. A supporting objective of the research was to create a natural language processing (NLP) tool for classifying depression to obtain the corpus necessary for ontology creation. The innovations were automation of ontology generation from unstructured text and compared them across depressed and nondepressed, religious and non-religious groups, in concept, hierarchical and relation similarities.Niederlohmann E. proposed clinical translation methods for classifying psychological session cues using a perceived-assess-dose-safeguard decision matrix and a conflict-square algorithm (PAD-S/CSA model). The study proposed a quantitative translation from less structured narrative clinical diagnosis notes to clear formulated documentation; to assist psychotherapists to provide next treatment move and tolerance threshold for the treatment. It generates a separated process notation to assist clinical reasoning and decision-making based on safety, tolerance, shame, avoidance, and functional recovery for SMI and multimorbidity.Complementing this work, Filos, Regio M.A. presented a decision-analytic evaluation framework in forensic mental health practice, using large language models (LLMs), and decision curve analysis (DCA). Rather than focusing solely on predictive accuracy, the framework emphasizes evidence integrity,...

This publication has 1 reference indexed in Scilit: