Application of Automated Statistical Methods applied to Public Health Surveys
We present an approach for automating the reporting of standard Epidemiological methods as applied to Public Health Survey datasets. We first examined the academic literature (PubMed) to see what statistical methods were applied to several surveys. We also looked at how this information was displayed in the form of tables and figures.
While conducting this research we ran into several issues with automating the analysis, such as the small sample sizes when segmenting different outcome and exposures. These issues needed to be addressed as when automating an approach is taken we need to account for all cases. Another issue that we are encountering is cpu performance as by automating the methods can lead to hundreds of thousands of statistical tests being preformed.
Several researchers in the epidemiological community have criticized this automated approach with the typical first response that “research isn’t conducted this way” or that research needs to be hypothesis driven. However, our approach would be the following:
Methodology – we are not changing the methodological approach and using new techniques, but instead are just automating the application of standard epidemiological methods already applied to these survey datasets.
Economics – we feel that automating standard methods improves the economics for research, as funding/time is not being utilized to answer questions that could be automated.
Research – we also feel that by automating the standard methods, that research can be pushed in new areas leading to new methods that could potentially be automated as well.
Validation/Reliability – we feel that an automated approach would improve the reliability of the research produced, as we currently have a research environment where reproducibility of research is a major issue.