In this episode, I’ll discuss the different types of statistical data.
How to evaluate whether the correct statistical test is being used in a medical study is a necessary part of being able to properly evaluate a study. The application of biostatistics is a complicated field and in practice, many clinicians rely on the judgment about statistical test selection in medical studies to journal editors and peer reviewers.
Evaluating whether the correct statistical test was used in a study becomes much more straight forward when one realizes that it is the type of statistical data being analyzed that determines the correct test to use.
There are two broad categories of data to consider when choosing a statistical test: discrete data and continuous data.
Discrete data is further broken down into two sub-categories: nominal data and ordinal data.
Nominal data, also known as categorical data is that which exists in one of two or more states. Examples of nominal data are:
- Marital status
- Smoking status
- Presence of disease
If the variable being analyzed can only exist in two possible states it is called “dichotomous.”
Nominal data are summarized as proportions or probabilities and therefore do not have a normal distribution.
Therefore, statistical tests used on nominal data need to be ones that do not assume a normal distribution.
Ordinal data measure an attribute that may be put into a finite number of categories.
While the categories may be assigned a number, this is purely arbitrary and the categories could just as well be letters or symbols.
Examples of ordinal data are:
- Likert scale
- CHADs2VASC score
- NYHA functional classification
- CURB-65 score
A critical point to understand about ordinal data is that the difference from one category to the next is not constant. Take for example the mortality risk represented by the CURB-65 score:
CURB-65 score mortality risk:
0 = 0.6%
1 = 2.7%
2 = 6.8%
3 = 14.0%
4 = 27.8%
5 = 27.8%
A difference in CURB-65 score from 0 to 1 represents a 2.1% increase in mortality risk, but a difference between 1 and 2 represents a 4.2% increase. Furthermore, there is no difference in risk between a score of 4 and 5 as these are counted as one risk category.
Because of this property of ordinal data, applying descriptive statistics such as a mean value has no meaning. If a group of patients is described as having a mean CURB-65 score of 2, there is no way to tell if this is from a group that is composed of 50% ‘0’ and 50% ‘4’ or any other mix of CURB-65 scores. And the patient population represented by different mixes of patients that would result in an average score of 2 could be wildly different.
Continuous data is that which provides a measurement. Examples of continuous data are:
- Blood glucose
- Blood pressure
While continuous data can further be divided into interval or ratio data, this is largely irrelevant to the choice of a statistical test.
Interval data is that which has an arbitrary “zero” point (temperature).
Ratio data is that which has an absolute “zero” point (blood glucose).
Continuous data may be normally distributed, and this should be assessed prior to choosing a statistical test.
Once the type of data being analyzed has been identified, choosing the correct statistical test becomes a more straight-forward process based on additional details such as whether the samples are independent or paired, and how many groups are being analyzed.
Members of my Hospital Pharmacy Academy have access to an in-depth training on how to choose the correct statistical test based on the type of data being analyzed. This training is part 1 of a 3-part biostatistics primer designed to cover the statistics material that is part of the Board of Pharmaceutical Specialties exams.
Part 1: Choosing the correct statistical test (published August 1, 2019)
Part 2: Analyzing trials (to be published September 5, 2019)
Part 3: Understanding the relationships between variables (to be published October 3, 2019)
To get access to these and over 85 other trainings, as well as many more resources to help in your practice become a member of the Hospital Pharmacy Academy by going to pharmacyjoe.com/academy.
If you like this post, check out my book – A Pharmacist’s Guide to Inpatient Medical Emergencies: How to respond to code blue, rapid response calls, and other medical emergencies.