Page 179 - Computational Retinal Image Analysis

174 CHAPTER 10 Statistics in ophthalmology

2 Data classification, data capture and data management

2.1 Data classification
Measurements are everywhere in ophthalmology and vision science. As machinery
develops, we are able to capture things that previously were not possible. In past
times there was no topography and so ophthalmologists could not explore Kmax
measurements and how they change over time. While glaucoma researchers have for
many years looked at measurements of Intraocular Pressure (IOP) there are now new
methods for measuring this and they may now need to capture corneal thickness.
Measurements may be intuitively understood by many to be data but what about in-
formation such as whether or not a patient has improved following a treatment, how
the patient feels about a particular condition. This also has value and serves as data.
In order to understand how to use statistical methods to convert data into meaning-
ful information that can answer questions of interest in ophthalmology research it is
important to be able to classify data. Classification is needed as a step toward under-
standing which statistical method can be employed.
There are several ways that data can be classified. One approach that is found
within many statistical textbooks is to first consider data as being either categori-
cal or numerical data. Categorical data is viewed by some as perhaps the simplest
type of data. A categorical variable may have two categories (dichotomous) such
as whether or not a patient has a diagnosis of glaucoma (yes or no) or may have
several categories such as the quality of an image measured as excellent, good,
fair, or poor. In the second example, we can see that there is a natural ordering
to the data—an excellent image quality is readily understood by a researcher as
being better than a good quality image and a fair image quality might be readily
understood to be better than a poor image quality. Such quality data can be termed
ordinal. For other categorical data with several categories however there may be no
intuitive way of ordering data—for example, eye color. Such data may be termed
nominal. Measurement data may be continuous or discrete—discrete data occur
when the observation in question can take only on discrete values, such as the
number of events (e.g. number of lesions in retinal image). The numerical continu-
ous data are usually obtained by some form of measurement. Typical examples are
weight, height, blood pressure, cup-to-disc ratio, curvature of vessels, size of a le-
sion or the central foveal thickness or an average thickness of retinal layers across
whole macula, corneal topography. These numerical continuous measurements can
take on any value from a continuous scale range, except they may be limited by
the precision of the instrument or precision of the recording (e.g. we may decide to
store the weight to the closest 100 g). The type of data can be critically important in
determining which statistical method of analysis will be appropriate and valid [4].
In many ophthalmic studies many types of variables are collected, so that several
different analytical methods may be needed.
Another important way of classifying data is into whether it is independent or
paired. This information is also needed in order to best select the appropriate form