Abstract

Several statistical packages, either commercials or open-source, provide many methods for multi-factorial and discriminant analysis; such a software is poorly used by physicians. Appropriate models and tests have to be used pending on the kind of experiment scheme, adequate distribution assumption are needed for variables and parameters and proper data validation have to be verified for historical records. These are but a few of many critical aspects for a robust and trustable data interpretation needed in the Evidence Based Medicine era. Clinicians always wish to be able to quickly interpreter diagnostic records to discriminate, or alternatively correlate, coherent groups of patient’s records according to either descriptive characters or variable units. Practically, patient’s records are stored in spread-sheet or database which change pending on the clinical trial scope; moreover, data entry and its validation is usually poor, hence physician are used to send raw-data to the statistician without contributing, for instance, with parametric and non-parametric indication on usable distribution. We address this problem by introducing a simple “weighted” model approached with the Unique Factorisation Domain theory: records can be compare by matching each other through a score overlap and clinician can modulate tolerance of closeness stringency criteria. An intuitive paradigm of records matching method (RMM) is presented and discussed with example, computational design and programming prototyping model; freely available material concerning real-world application, are also provided by the authors.

Disciplines

Clinical Epidemiology

Share

COinS