Change from baseline is a common way to measure treatment effect in many health-care research studies. For such studies measurements are taken before and after the intervention. Some examples can be: body weight before and after treatment, HbA1c before and after treatment, blood pressure before and after treatment etc.
The question arises, what would be the best way to measure of the change from baseline? Two of the most commonly used methods for this purpose are: absolute change from baseline and percentage change or relative change from baseline.
If we call themeasurement before exposing the intervention as ‘Pre’ and the measurement afterthe intervention as ‘Post’ then the absolute change from baseline is defined as the difference between the Post and Pre measurements. The percentage or relative change will be defined as (Post-Pre)/Pre.
Therefore, the percentage change will answer the question ‘By what percentage did the value change from the baseline value?’ and the absolute change will answer the question ‘What is the difference between the baseline value and the post baseline value?’
Both absolute change and the relative change have their own merits and demerits.
By its definition, the absolute change keeps the same unit as the variable being measured and therefore it is easy to interpret. On the other hand, the merit of percentage change from baseline is that it is a unit less quantity and independent of the units of measure. For example, percent change in cholesterol level for a sample will remain the same irrespective of whether cholesterol is being measured in milligrams per deciliter or millimoles per liter (example taken from Berry et. al., 2006).
Taking the above cholesterol example, we could get the following answers for absolute change and percentage change respectively:
Absolute change: cholesterol reduced from baseline by 16 units in the active treatment but only by 10 units in the control group.
Percentage/relative change: cholesterol reduced from baseline by 30% in the active treatment but only by 10% in the control group.
It is clear from above that the percentage change talks in terms of percentage which is intuitive to most of the people. However, even though it is tempting to use percentage change as a measure, it is also associated with some issues which we are going to discuss next in the article.
Percentage change of the small numbers can appear to be more significant than they actually are. For example, if a particular parameter increased from 12 to 60, it is 400% increase in percentage change which appears to be a big change even though the absolute increase is just of 48 units.
On the other hand, the percentage change on big numbers can appear less significant. For example, if we say that we have a population of bacteria with approximate number of bacteria as 20,000,000,000,000 and a certain chemical increases the number of bacteria by 5% then the percentage increase does not appear to be so big even though the absolute increase is 1,000,000,000,000!
Therefore, percentage change reporting may distort the magnitude of the effect of an intervention.
The question comes, should we never report the percentage change and stick to only absolute change?Andrew J Vickers (2001) performed a simulation study and concluded that percentage change from baseline as an outcome in a controlled trial is statistically inefficient and therefore it is not recommended to use in clinical trials.
It can also happen that the distribution of the percentage change data is highly skewed and therefore the basic assumption of normality may not hold. It is recommended thatone should not analyse percentage data with Analysis of Variance (ANOVA) because the basic assumption of normality for ANOVA may be violated when the response variable is percentage change.
Moreover, it has been shown in the past that the percentage change from baseline has much lesser power as compared to absolute change from baseline.
In spite of the above-mentioned drawbacks, percentage change is widely used in clinical domain because of itsintuitive easy interpretation as mentioned above. If the trial wishes to report percentage change or the relative change data, the researcher should be careful in interpreting the results.
The results should also be supported with using absolute change data so that the reader does not get wrong conclusions.
Written by Bhaswati Mukherjee, PhD.
References:
Andrew J Vickers, ‘The use of percentage change from baseline as an outcome in acontrolled trial is statistically inefficient: a simulation study’, BMC Medical Research Methodology (2001) 1:6.
Michael Hochman et. al., ‘Endpoint Selection and Relative (Versus Absolute) Risk Reportingin Published Medication Trials’, J Gen Intern Med 26(11):1246–52.
Donald A. Berry and Gregory D. Ayers, ‘Symmetrized Percent Change for Treatment Comparisons’, The American Statistician , Feb., 2006, Vol. 60, No. 1 (Feb., 2006), pp. 27-31.