Probabilistic linguistic fuzzy cognitive maps: applications to the critical factors affecting the health of rural older adults

Jian Wu, Yucheng Chen, Zengwen Wang, Guoheng HuBMC Medical Informatics Decis. Mak 2022 · DOI

TL;DR

This paper introduces a novel modeling approach, Probabilistic Linguistic Fuzzy Cognitive Maps (PLFCMs), to analyze the complex factors influencing the health of rural older adults in China. By integrating probabilistic linguistic term sets with fuzzy cognitive maps, the method effectively handles uncertainty and expert opinions, revealing that education and occupation are the most critical factors affecting health in this demographic.

What Problem Does This Paper Address?

As China faces a rapidly aging population, particularly in rural areas, understanding and improving the health of older adults is a critical challenge. The factors influencing their health are numerous, intricate, and interdependent, making it difficult to pinpoint key drivers for targeted interventions. Existing methods often struggle to capture the full complexity of these relationships, the uncertainty inherent in expert judgments, and the probabilistic nature of linguistic evaluations. This paper aims to address this by developing a more robust framework for analyzing these complex causal relationships and identifying the most influential factors.

Methods

The core of this paper is the introduction of a Probabilistic Linguistic Fuzzy Cognitive Map (PLFCM) model. This novel approach extends traditional Fuzzy Cognitive Maps (FCMs) by incorporating Probabilistic Linguistic Term Sets (PLTSs). PLTSs allow experts to express their opinions not just with linguistic terms (e.g., "good," "bad") but also with associated probabilities, reflecting a more nuanced and realistic representation of uncertainty.

The PLFCM model represents factors as nodes and their causal relationships as directed edges, with weights expressed as PLTSs. The authors propose a specific method for constructing these maps, including defining operational laws for PLTSs and an iteration algorithm that separates membership and probability. A key contribution is a novel similarity measure based on Euclidean distance and Z-mapping functions, used to rank the identified factors based on their influence on health outcomes. This similarity measure helps to quantify the impact of each factor and distinguish their relative importance.

Datasets

The study utilizes data from three major Chinese databases:

  • 2014 Chinese Longitudinal Healthy Longevity Survey (CLHLS)
  • 2015 China Health and Retirement Longitudinal Study (CHARLS)
  • 2015 Chinese General Social Survey (CGSS)

These databases provide comprehensive information on demographic, socioeconomic, health, family structure, and lifestyle aspects of rural older adults. The initial state vector for the PLFCM model was derived from statistical analysis of variables within these datasets.

Key Insights and Ideas

  • Beyond Simple Causality: The research highlights that health is a complex system where factors are not isolated but interact dynamically. This necessitates models that can capture these interdependencies, which PLFCMs aim to do.
  • The Power of Probabilistic Language: Traditional methods often simplify expert opinions. The use of PLTSs allows for a richer representation of uncertainty, where experts can express not only what they think but also how confident they are about it.
  • Education as a Health Driver: Counterintuitively, education emerged as the most significant factor influencing the health of rural older adults, surpassing more commonly assumed factors like lifestyle or direct medical access. This suggests a profound indirect impact of education on health outcomes.
  • Occupation's Hidden Role: The primary occupation of older adults before age 60 also plays a surprisingly significant role, likely due to its connection to manual labor experiences and lack of employer-provided pensions, impacting current health status.
  • Family Income's Indirect Influence: While family income is often considered a primary determinant of health, this study found its direct impact to be minimal, suggesting its influence is largely mediated through lifestyle and living environment.

Key Results

The study identified the following key factors influencing the health of rural older adults, ranked by their impact:

  1. Education (C10)
  2. Occupation (C9)
  3. Mentality (C5)
  4. Physical Exercise (C3)
  5. Intergenerational Relationship (C12)

Overall, socioeconomic status, living environment, lifestyle, and health management were identified as the most impactful first-level categories. The PLFCM model demonstrated robustness across different Z-mapping functions and parameter values, indicating reliable results.

Limitations and Future Work

The authors acknowledge that the effect degree of health factors can vary across different regions and age groups, suggesting that distinct health interventions should be tailored accordingly. While the PLFCM model effectively handles uncertainty, future work could explore further refinements in expert elicitation and the integration of more diverse data sources. The paper also suggests that while direct improvements to education and occupation might be challenging for current rural older adults, governments can foster well-being through social organizations, community engagement, and promoting intergenerational relationships.

Summary updated

BibTeX
@article{wu2022probabilistic,
  author = {Jian Wu and Yucheng Chen and Zengwen Wang and Guoheng Hu and Chen Chen},
  title = {{Probabilistic linguistic fuzzy cognitive maps: applications to the critical factors affecting the health of rural older adults}},
  journal = {BMC Medical Informatics Decis. Mak.},
  year = {2022},
  doi = {10.1186/s12911-022-02028-9},
}
Figure