Research, Rapid Research Review, Advancement of EM

Rapid Research Review: Data Sets and Summary Indexes

Olga Kovalerchik, MD, Emergency Medicine Resident, EMRA Research Committee, Vice-Chair, Yale School of Medicine, New Haven, CT

You've just finished collecting data for your study on the effectiveness of a particular medication. Now you may want to use the information to make comparisons either internally or externally. You want to know if your control subjects are similar to each other or to the individual members of the treatment group. You also want to use what you found out about your population and extend it to the population at large. As part of this, you'll need to address whether your study population is younger, more obese, or has more cardiac disease than the rest of the general population. To allow for this comparison, you need to use the right summary indexes. Is the mean, median, or mode going to best capture your select population? Do these apply?

If the data cannot be ranked or has arbitrary magnitudes, it is considered non-dimensional. Non-dimensional data can be binary, nominal, or ordinal. Sex, generally referred to as male or female, is an example of binary data. If you documented the medical conditions of your patients such as hypertension, diabetes, or depression, you have data that is nominal. If you used a particular pain scale where pain is rated none, mild, moderate, or severe, you have ordinal data. Non-dimensional data has no absolute defined intervals between values. Ranking this data and trying to find a way to represent it in a central index is tricky for this reason. For example, is severe pain always twice as bad as mild pain? Not always.

Dimensional data, on the other hand, is more consistent in some ways. Data is dimensional if each of the values has an equal interval between the next, and also if the data is monotonic, where each value is consistently the same value more or less than the preceding category. If your data is dimensional, such as in age, weight, or height, you can use the conventional measures of central tendency such as mean, median, or mode. Examples of dimensional data include age, weight, or height.

Reference: Feinstein AR. Principles of Medical Statistics. College Park, MD: Chapman & Hall/CRC. 2002.