Generalized Resemblance Theory of Evidence: a Proposal for Precision/Personalized Evidence-Based Medicine

Precision medicine emerges as the most important contemporary paradigm shift of medical practice but has several challenges in evidence formation and implementation for clinical practice. Precision/Personalized evidence-based medicine (pEBM) requires theoretical support for decision making and information management. This study aims to provide the required methodological framework. Generalized Resemblance Theory of Evidence mainly rests upon Generalized Theory of Uncertainty which manages information as generalized constraints rather than limited statistical data, and also Prototype Resemblance Theory of Disease which defines diseases/conditions when there is a similarity relationship with prototypes (best examples of the disease). The proposed theory explains that precisely-personalized structure of evidence is formed as a generalized constraint on particular research questions, where the constraining relation deals with averaged effect sizes of studies and its comparison to null hypothesis; which might be of either probabilistic or possibilistic nature. Similarity measures were employed to deal with comparisons of high-dimensional characteristics. Real examples of a meta-analysis and its clinical application are provided. This is one of the first attempts for introducing a framework in medicine, which provides optimal balance between generalizability of formed evidence and homogeneity of studied populations.

💡 Research Summary

The paper addresses a fundamental mismatch between the aspirations of precision/personalized medicine and the methodological foundations of traditional evidence‑based medicine (EBM). While precision medicine seeks to tailor interventions to the unique genetic, environmental, and lifestyle profile of each patient, conventional EBM relies on aggregated statistical summaries that often obscure individual variability. To bridge this gap, the authors propose the Generalized Resemblance Theory of Evidence (GRTE), a conceptual framework that integrates two previously separate theoretical strands: the Generalized Theory of Uncertainty (GTU) and the Prototype Resemblance Theory of Disease.

GTU reconceptualizes uncertainty not merely as a probabilistic distribution but as a set of generalized constraints. In practice, a research question is expressed as a constraint on the average effect size derived from a collection of studies, together with its dispersion. This constraint is then compared to a null hypothesis that can be framed either probabilistically (as a traditional null distribution) or possibilistically (as a possibility function). By allowing both probabilistic and possibilistic nulls, GRTE expands the interpretive space beyond the conventional p‑value paradigm, offering a richer description of evidence strength.

The Prototype Resemblance Theory of Disease treats diseases as prototypes—ideal exemplars—rather than fixed categorical definitions. Patient characteristics (clinical measurements, genomic markers, environmental exposures, etc.) are encoded as high‑dimensional vectors. Similarity between a patient and the disease prototype is quantified using multivariate similarity metrics such as cosine similarity, Mahalanobis distance, or kernel‑based measures. This similarity score becomes the bridge that maps population‑level evidence onto an individual’s context, thereby operationalizing the “personalized” aspect of precision medicine.

To demonstrate feasibility, the authors re‑analyze a meta‑analysis of several randomized controlled trials. They construct a constraint function from the pooled mean effect and its confidence interval, then evaluate this constraint against both probabilistic and possibilistic null hypotheses. Simultaneously, they compute similarity scores between the trial populations and a target patient profile. The results reveal that, although the pooled effect is statistically significant under traditional analysis, the constraint‑based assessment combined with similarity weighting can indicate negligible benefit—or even potential harm—for specific patient subgroups. This illustrates how GRTE can preserve the generalizability of evidence while simultaneously exposing heterogeneity that matters for individual decision‑making.

Strengths of the proposal include: (1) a unified treatment of uncertainty that accommodates sparse or heterogeneous data; (2) a principled method for translating population‑level findings to individual patients via prototype similarity; and (3) an expanded hypothesis‑testing framework that incorporates possibilistic reasoning. However, the authors acknowledge several challenges. Defining constraints and selecting similarity metrics involve subjective choices that may affect reproducibility. Possibilistic null hypothesis testing lacks established statistical standards, potentially limiting acceptance among clinicians and regulators. Moreover, high‑dimensional data require careful preprocessing, dimensionality reduction, and normalization to avoid spurious similarity estimates.

In conclusion, GRTE offers a promising theoretical scaffold for precision/personalized evidence‑based medicine, aiming to balance the twin goals of generalizability and homogeneity. Future work must focus on standardizing constraint formulation, validating similarity metrics across diverse clinical domains, and integrating possibilistic inference into mainstream statistical practice. If these methodological hurdles are overcome, GRTE could become a cornerstone for delivering truly individualized, evidence‑driven care.