Adjustment for covariates helps clinical trial statisticians demonstrate differences in drug effect between groups of people. In this blog, we discuss what this means and how it can help improve analysis and interpretation of clinical trial data.
Differences in patient characteristics like age, gender, medical history, etc. cause variance in clinical trial data that might obscure the true efficacy of a drug. These characteristics will not change or be influenced by drug treatment, meaning any drug response (or lack thereof) produced by them is not indicative of true drug efficacy.
Take age as an example. The age of a patient in a clinical trial could influence their response to treatment. However, this doesn’t mean the drug itself is ineffective. It just creates data variability that makes it more difficult to statistically demonstrate drug efficacy; unless, of course, if a covariate adjustment is used.
What is a Covariate Adjustment?
In clinical trials, drugs are evaluated in a diverse population of patients: a sample intended to be representative of the general population. This keeps results translatable. Yet, everyone is different and responds to drugs differently, but these differences don’t mean that a certain drug doesn’t work.
The covariate adjustment has two core functions:
- Reduce variability (noise) in clinical trial data
- Correct for unequal distribution between groups due to random sampling
Sometimes, there’s so much noise in the data that you can’t see anything else. Other times, there may be bias in the data because of unequal distribution of patients between groups. The covariate adjustment minimizes such distribution bias so you can “see” drug efficacy through the noise.
Age is a common covariate. Another example is the baseline severity of the disease. For example, in pain, someone who has higher pain at the start of a study may see a sharper reduction in pain throughout treatment. However, this doesn’t indicate drug ineffectiveness for those who have less pain at the start of the study.
How to Select Covariates for a Clinical Trial
Before the study begins, a clinical trial statistician will develop a statistical analysis plan as part of the clinical study protocol. During this process, the statistician can identify the covariates they intend to use in the statistical analysis. During regulatory clinical study protocol review (for example, during a pre-IND meeting), the regulators will have the opportunity to comment on and/or approve the intended analyses.
To select a covariate, three things need to be true:
#1. Covariate Is Relevant
The covariate identified needs to be relevant to the clinical trial indication.
For example, in a knee Osteoarthritis (OA) study, BMI is likely a relevant covariate. This is because heavier patients will likely experience more OA pain. Conversely, a study in dry eye disease wouldn’t consider BMI as a covariate because that characteristic has no influence on the results and would not muddy the drug efficacy data.
#2. Data Is Accessible
Speaking of data, a characteristic needs to be accessible to be used as a covariate. Many characteristics can be covariates, but if you can’t easily quantify it, you won’t be able to actually analyze it in the data.
#3. Covariate Is Not Affected by Treatment
Importantly, baseline covariates must exist and be collected at the beginning of the study and not be influenced by study treatment. This is of particular importance to properly evaluate drug efficacy and not introduce any bias in the analysis (such as increasing the probability of false positive results). The covariate must not change regardless of therapeutic intervention to meet the criteria of a baseline covariate.
Covariate Adjustment Follows Regulatory Guidance
The practical utility of baseline covariates has been covered by regulatory guidance issued by the FDA and EMA.
The FDA draft Guidance, EMA Guidance and the ICH Guidance for Industry all encourage the identification of “covariates and factors expected to have an important influence on the primary [endpoint].”
Trial statisticians determine the success of the experimental therapy vs. pre-defined criteria to understand drug efficacy. The covariate adjustment is there to help prove true efficacy by removing the characteristics that could be muddying the results. It’s important to pre-define success and the covariates to prevent any bias during data analysis. So, how does it all actually work?
Through an analysis of covariance, or ANCOVA.
Analysis of Covariance (“ANCOVA”) Adjustment
The statistical analysis plan included in the study protocol outlines the way the statistician will analyze results at the end of the study to prevent bias. This analysis tries to understand what would happen if the differences between patients (that don’t relate to drug effect) were minimized.
Going back to the age example, ANCOVA would adjust the data to see what an older patient’s drug response would look like if that individual was the same age as everyone else. Essentially, it mathematically moves age to the middle of the distribution to understand treatment effect in that scenario.
If you can determine that there is some variability in the drug effect data that can be attributed to the age of patients, ANCOVA allows statisticians to normalize that in the data. You are left with cleaner, more precise understanding of true treatment effect, regardless of the fact that it was tested in patients of varying age.
As the FDA puts it: “Sponsors can use ANCOVA to adjust for differences between treatment groups in relevant baseline variables to improve the power of significance tests and the precision of estimates of treatment effect.”
Covariate Adjustment for Placebo Responsiveness
Placebo responsiveness is another innate characteristic that causes noise in clinical trial data. But, before now, that data has been inaccessible—which means it couldn’t be used in a covariate adjustment. You couldn’t quantify placebo responsiveness for every patient the way you can easily quantify age or baseline pain level.
Thanks to the advent of predictive modeling, it’s now possible to have insight on placebo responsiveness accessible at the beginning of the study. By combining an understanding of individual patient psychology with a predictive machine learning algorithm, you can calculate a relative placebo responsiveness score for each patient. Then, you can use this score as a covariate in the statistical analysis.
While it’s important to think critically and follow proper guidance as it relates to selecting covariates, placebo responsiveness almost always impacts results, causing higher Phase II and III trial failures. Now that the characteristic can be accurately predicted before the study, the covariate approach can be utilized to better manage the data variability and increase study power.
Placebell©™ is a proven solution for assessing patient psychology, calculating placebo responsiveness and defining covariates in a statistical analysis. Powered by advanced machine learning models, the Placebell approach poses no mathematical or operational negative impact on your trial or data.
