The pioneers in epidemiology were justifiably called “shoe leather” epidemiologists for their strong field work and willingness to hike into the Netherlands in search of answers to epidemics. I recall a beloved teacher, Dr. George Comstock, with his ingenious knitting needle that sorted data cards and enabled him to analyze data while up in Alaska or investigating a TB outbreak. Early textbooks in epidemiologic methods ventured with an element of distrust towards multi-variable models, cautioning (and wisely) that they were a black box and ill used by the uninformed. As students, we learned to investigate infectious disease outbreaks in our beginning epidemiology classes, and the simple cross tabulations and chi square analyses were sufficient to incriminate the guilty food, and lead to remedial actions. At Johns Hopkins, no one ate egg salad sandwiches for years after that class.
Epidemiology has grown more complex today, and while the backbone of the discipline is still population based approaches to understanding diseases, the multi-factorial nature of health problems, tools at our disposal, and the growth in sophisticated biostatical methods have altered the analytic armamentarium at our disposal. This is especially true in Ophthalmology, where issues of clustering were faced decades ago with the need to accommodate the information from both eyes in a single head.
In an effort to keep our readership informed about some of the latest methods in use, we are inaugurating a “Methods Series” this year, written by some of the most talented methodologists in the field. Dr. Robert Massof leads off this series with an article on item response theory models. For investigators who use questionnaires to garner data on outcomes such as functional impairment or quality of life, or physical disability, this article is especially important for understanding the creation of rating scales. Any new instrument devised to measure, for example, “Quality of Life in Patients with Z syndrome” should be subjected to more rigorous testing of score creation than simply summing up responses. Rasch models are one approach to this issue, but not the only one.
In forthcoming issues, we will have articles on Structural Equation Modeling, Competing Risk models, and approaches to multi-level clustering of data. Applications to research in Ophthalmology will be stressed, and our hope is the articles will prove helpful to investigators faced with complex analytic problems. Furthermore, if readers also identify another methodological area that needs description and demonstration of applicability, we at Ophthalmic Epidemiology would be most responsive to such a submission. Reviewers are increasingly more sophisticated about appropriate methods for data analyses of complex problems, and those of us who wear our “shoe leather” proudly would be well advised to also advance our understanding of these new approaches.
By the way, I still do not eat egg salad sandwiches.