top of page

How to deal with missing data and outliers in observational studies

How to deal with missing data and outliers in observational studies

How to deal with missing data and outliers in observational studies

Saverio Verdone from the Thrombosis Research Institute and a research fellow at the London School of Hygiene and Tropical Medicine, presents an introduction to a problem that is ubiquitous in clinical research, missing data. He covers the basic concepts of missing data, the missing data lab mechanisms, and goes on to talk about the main conventional methodologies used to handle missing data, together with their advantages and disadvantages.

SUMMARY KEYWORDS
missing, data, analysis, methods, multiple imputation, imputation, dataset, outliers, talk, information, values, estimator, bias, results, patients, imputed, variable, complete, single, methodology

SAVERIO VERDONE:

Now, before we start talking about the different approaches how to handle missing data, it's good that we define what we mean by missing data. So, we're sure that we're all on the same page here. Missing data are defined as values that are not available and that would be meaningful for analysis, if they were observed. There is no point in collecting type of stroke data in patients who did not develop a stroke, for example. Okay, so yeah, information on type of stroke is missing. But they would not be considered information would not be considered missing. According to this definition we're talking about this data that would be meaningful for the analysis. Missing data are everywhere, in every type of study. They are in our cities, and missing data have seriously compromised influences from randomized clinical trials. And the topic has received legal attention in the clinical trial community. And the existing regulatory guidelines on the design conduct an analysis of clinical trials have little specific advice on how to address this problem.

But here, we are going to focus on observational study. Anything can be missing in an observational study really. The exposure information can be missing the outcome confounding variables can also be missing and missing data observational studies can truly impact the validity of a study. There are a large number of review papers on missing data and that they describe how they are poorly handled, and things are changing quite slowly. Are missing data big issue? Yeah, they can be they can be. So, we should care about missing data for two reasons basically. Reduce sample size. So, if we have missing data, we have less individual less patients. This means reduced statistical power, wider uncertainty, wider confidence intervals. And we will struggle if we want to do some subgroup analysis...

To see the full content, please click on the button below.

bottom of page