A new measure for the analysis of epidemiological associations: Cannabis use disorder examples

TL;DR

This paper introduces a novel normalized effect size measure for epidemiological data analysis, improving the power of hypothesis testing and applied to cannabis use disorder data from US surveys.

Contribution

It proposes a new normalized effect size measure based on the maximum range of standardized log(OR) and derives its distribution, enhancing analysis of epidemiological associations.

Findings

New measure outperforms traditional OR tests in simulations

Applied to US cannabis use data from 2004-2014

Provides more accurate inference on epidemiological associations

Abstract

Analyses of population-based surveys are instrumental to research on prevention and treatment of mental and substance use disorders. Population-based data provides descriptive characteristics of multiple determinants of public health and are typically available to researchers as an annual data release. To provide trends in national estimates or to update the existing ones, a meta-analytical approach to year-by-year data is typically employed with ORs as effect sizes. However, if the estimated ORs exhibit different patterns over time, some normalization of ORs may be warranted. We propose a new normalized measure of effect size and derive an asymptotic distribution for the respective test statistic. The normalization constant is based on the maximum range of the standardized log(OR), for which we establish a connection to the Laplace Limit Constant. Furthermore, we propose to employ standardized log(OR) in a novel way to obtain accurate posterior inference. Through simulation studies, we show that our new statistic is more powerful than the traditional one for testing the hypothesis OR=1. We then applied it to the United States population estimates of co-occurrence of side effect problem-experiences (SEPE) among newly incident cannabis users, based on the the National Survey on Drug Use and Health (NSDUH), 2004-2014.