Vulnerability Models Allowing for Incomplete Information and Imprecision Application to Eldercare

Personalized, multi-dimensional assessments of the risk of the condition of a person (or object) deteriorating, called vulnerability assessments, are critical in outcome-based care management. However, lack of time for information gathering, often coupled with urgency to take action, pose specific challenges for individual vulnerability assessment in the field compared with assessment in a clinical or laboratory setting. We describe an approach to develop vulnerability assessment models for use “in the wild” - as distinct from use in clinical/laboratory settings - borrowing a term from the visual emotion recognition terminology. Key elements of the proposed modelling framework for vulnerability assessment are that it (i) accommodates incomplete information about the person (ii) continuously adjusts as background information evolves (iii) can serve as a guide to prioritize information gathering (iv) can function with imprecision in the input parameters. Specifically, we integrate a Markov Chain model describing the evolution of the person into, and out of, vulnerable states together with a Bayesian network that serves to customize the dynamic model. We describe the extension of the framework to situations arising through the presence of imprecision in the model parameters. The examples that we present were developed and validated as practical tools for the analysis of the vulnerabilities of elderly persons in the context of a consulting engagement in China. The techniques presented are general however, and generalizations are discussed in the conclusion section.