Multivariate refutation of aetiological hypotheses in non-experimental epidemiology

Area of Science:

• Epidemiology
• Biostatistics
• Scientific Methodology

Background:

• Traditional multivariate modeling in non-experimental epidemiology often prioritizes sensitivity in detecting causal agents.
• Current methods like forward selection and backward elimination may not align with rigorous scientific principles for establishing causality.
• Karl Popper's logic of refutation offers an alternative framework for scientific inquiry.

Purpose of the Study:

• To extend Karl Popper's logic of refutation to multivariate modeling in non-experimental epidemiology.
• To propose multivariate refutation as a more scientifically rigorous approach compared to traditional methods.
• To highlight the shift in objective from high sensitivity to high specificity in defending the 'innocence' of exposures.

Main Methods:

• Application of Karl Popper's logic of refutation to multivariate analysis.
• Utilizing a 'forward elimination' method instead of traditional selection techniques.
• Employing a likelihood approach for statistical inference, study design, and conduct, working backward from hypothesis to data.

Main Results:

• Multivariate refutation emphasizes maximizing specificity to rigorously test the 'innocence' of potential causal exposures.
• The likelihood approach guides both statistical inference and the design of epidemiological studies.
• Illustrative example using myocardial infarction triggers demonstrates practical differences from traditional modeling.

Conclusions:

• Multivariate refutation offers a more scientifically sound approach for non-experimental epidemiology.
• This method enhances the defense against false positives by focusing on specificity.
• The concept of multivariate refutation should supersede traditional multivariate modeling in this field.