If I have a RCT where I have a control group (receiving funny data messages) and an intervention group (diet, physical activity tips reinforced with self-efficacy) with the aim of improving glycemic control.
I measured physical activity (PA) with an accelerometer. And I want to know the pros and cons of using my PA data as continuous and as a categorical variable.
Thank you for your help and time.
My hypothesis is that those participants receiving diet, PA reinforced with self-efficacy messages will have a better glycemic control than the participants receiving funny data messages. And PA and diet are variables for adjusting my HbA1c levels (glycemic control).
And my PA is in counts/week
Hi Alejandra,
In general, if you have a continuous variable (such as PA measured by accelerometry), it is much better to use it as it is, continuous. From a statistical point of view, categotizing continuous variables will lead to lost of information.However, depending on what you need to study, sometimes it is practical/necessary to categorize.
In your case, as I understand, you want to examine the relationship between receiving self-efficacy messages (or not) and glycemic control, after controling for diet and PA. I would then use physical activity as a continuous variable. You could use total PA, MVPA, or whatever literature suggests.
You can review this article for further information:
Altman DG, Royston P. The cost of dichotomising continuous variables. British Medical Journal. 2006;332:1080.
Hi Alejandra. Your choice really depend on your research question.
If possible I would use PA as a continuous variable. However, it may be valuable to you to categorize PA to see if PA and HbA1C are related as you expect - the higher the PA the lower the HbA1C.
When did you measure PA? Both at baseline and follow-up? I would not adjust for PA at follow-up because that might remove some of the effect you are looking for with your intervention.
Without knowing more details about your study, it appears that you will use physical activity as a predictor of the health outcomes (glycemic control), in which case using PA data as continuous would provide more detail than in a category (as mentioned by Alberto and Esther). I also agree with Esther that you could examine the PA data in categories (such as meeting PA recommendations or not) and look at those relationships as well. Sometimes you get interesting patterns when putting PA data in categories that you might not otherwise see when examining it as a continuous variable, and sometimes those patterns are what eventually shape our recommendations for activity levels in specific populations.
I agree with previous points made about the cost of categorizing data that are truly continuous.
Categorizing data in this way should only be done if there is strong rationale for the cutpoints you use to categorize the data. For example, you may categorize some health data using specified criteria for healthy/unhealthy levels.
However, too often data are categorized using arbitrary cutpoints or simply by splitting the data in the middle, or at the 33rd percentiles, or every 25th percentile or every 20th percentile. This is potentially misleading as you then create groups of "low active", for example, even if the whole sample was very active.
In general, you should leave continuous PA data in their original raw (continuous) form unless you have a very supportable rationale for doing otherwise.
Thank you all for answering my question, you have been very helpul.
Have a nice day. :)
Glycemic control through PA is also dependent on intensity of the PA. Short duration and high intensity may have a larger effect than a much longer duration at moderate intensity. Intensity will be as important a variable as duration. If you are using duration as a continuous variable, the intensity could be a confounding variable if the participants were not all doing similar PA.
Dear Alejandra,
I agree with Professor John Wells as the efficacy of the glycemic control in your research involves two different processes: impact of PA ( and its variables of frequency intensity duration and metabolic response to each session) and impact of a communication process i.e. receiving funny messages which implies in adequacy of the message content to the receiver and its decodification and degree of application, accepting and putting it into practice. Both processes are different in their nature. Maybe it shoud be better if you analyse them separately because there may be many confouding variables and you will not know how to deal with them, since these variables behave either as continuous and as categoric as well. Good luck!
If you are using your PA variable as counts per week, you will end up with a lot of time that is classified as sedentary behaviour (sitting/lying while awake and with low energy expenditure and typically characterised as less than 100 counts/minute). It is postulated that long periods of sedentary behaviour acts on glycaemic control in a way that is different than physical activity. It might be interesting to tease out the independent effects of both sedentary behaviour and physical activity.
i agree with the general statement that it is better to use continuous variables as continuous. this does not mean, however, that you should assume that the relationship is linear. one of the reasons for using categories for values of continuous variables is to avoid forcing a functional form on the relationship. in your case, you are dealing with PA as a potential confounder. i note that if your sample size is large enough and you have a randomized trial, you don't need to use any confounders (though doing so may sharpen the effect you care about).
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication