In our last blog we discussed sample size choice and identified the key information that your statistician needs to know and why. In our second clinical trial design blog we shall look at the different types of endpoints in clinical trials and at some of the statistical issues they raise.
What is an endpoint in a clinical trial?
An endpoint is a trial outcome value that is measured for each subject, for example, objective tumour response (OR) or time to death (overall survival or OS).
The responses are summarised for a group, giving an Objective Response Rate (for OR), or median survival (for OS); the summary values can then be compared between groups. Note that a summary value is NOT an endpoint in itself.
Below we have detailed the four key end points used in clinical trial design. We have listed them from the most simple (binary) to the more complex (time to event) at the bottom. This is only meant as an introduction to this area so please get in touch if you have any more specific needs.
Binary endpoints are yes / no type measurements:
- Objective tumour response
- Dose limiting toxicity
- Achieving a preset level of change, eg increase haemogobin by 2g/dL.
Each subject is allocated to one category only:
- Tumour response catagories:
1 Complete response
2 Partial response
3 Stable disease
4 Progressive disease
- Likert-type scale for Quality of Life
These endpoints are measured on a numeric scale:
- Change in tumour size
- Visual analogue scale
- Walking distance in a certain time
They are sometimes displayed as “waterfall” plots typically showing each subject’s change from a baseline value:
…or boxplots summarising the data for a group:
Typical summary measures for continuous endpoints are:
- Average (Mean, median)
– Minimum, maximum, range
– Quartiles, interquartile range (IQR)
– Standard deviation (SD)
– Variance (= SD squared)
Time to event endpoints
Some continuous measures such as survival time may be limited by the length of the trial or because a patient is lost to follow up or dies from an unrelated cause.
In this situation the defined endpoint cannot be measured, but is known to be longer that the observed time on study.
In the chart above, the end of the study period was at 30 months (red dots). Green dots are deaths while red dots are ‘censored’ times due to a maximum follow-up time of 30 months. The graph shows the full survival time for all patients, even though some survived well beyond the end of the trial. The exact survival time would be unknown to the investigator at the time of analysis.
Time to event endpoints are usually measured from the time of randomisation because this is a common time point to all patients for their time on the trial, and therefore maintains the comparability between the treatment groups.
The most common graphical display for time to event endpoints is known as a Kaplan Meier survival curve (even if the endpoint is nothing to do with life-and-death, for instance time to recovery). The plot below shows Kaplan Meier curves for two treatment groups in a trial (solid and dashed lines). The curves show the probability of survival length in months. Each step down represents a time where a death happened. The black marked points indicate times at which a patient’s survival was censored. The probability of survival to 3 months is ~58% in one group (green dashed line), and 72% in the other (green solid line). The median survival (time to which 50% of patients survive) is 5 months (red dashed line) and 8.2 months (red solid line).
In the next blog we will look at how the summary results for a group can be compared, and how it is possible to estimate how well the results from the sample of patients taking part in the clinical trial reflects the likely results that would be seen if the treatment were to be used for all suitable patients.