Most days I feel like the luckiest person alive. I get to travel as much or as little as I desire and along the way I meet the leading minds in healthcare, health policy, and even economic theory. You know what they say about being the smartest person in a room? Find another room.
But it isn't without challenges. Headlines herald new breakthroughs and advances especially in oncology--but the data is actually distorted to skew towards a cure rather than a potential stepping stone.
For example, when I read the statistics sections of clinical research reports I see the interchangeable use of multivariate and multivariable analysis. To begin to see light between these two terms, we need to understand the difference between a categorical and a continuous variable. Here is an excerpt from a small book I wrote as a guide, Improving Numeracy in Medicine.
You have heard of regression models. Outcome variables in linear regression models are continuous while logistic regression variables are categorical or have a binary outcome--survival analysis is time to event.
Multivariate analysis in a sense is referring to statistical models with 2 or more dependent or outcome variables. Multivariable analysis refer to multiple independent or response variables. A simple linear regression model will have a continuous outcome and one predictor while multivariable linear regression models have a continuous outcome and multiple predictors (continuous or categorical).
Whats the big deal? Maybe there isn't one. Remember I am not a statistician but when I see the wrong model being used or described in the methods or results section I begin to arch my eyebrow and try and unpack the mathematics. After all, the articles are written for informed decisions and often--important interpretation at the point of care.
Data is like a living organism. Depending on when you collect or how you collect the data you may only be capturing a "moment in time" and nothing more. We need to be more cognizant of making informed decisions at the intersection of multiple data sources and perspectives. Methods used to generate the data should be transparent and reproducible. I keep this series on Evaluation of Scientific Publications handy--Survival Analysis.
As a professional reviewer I often see the mishmash of terms and confusing model selection. Remembering the dynamic nature of the data we collect may go a long way in seeking robust statistics and open dialogue around what we can and can not infer...
Bonny is a data enthusiast applying curated analysis and visualization to persistent tensions between health policy, economics, and clinical research in oncology.