Why we need to tackle the problem of missing data in palliative care research
Nearly nine out of every ten randomised controlled trials in palliative care report having missing data, which can mean these studies are biased, less accurate and less powerful than they should be.
It’s vital that we find ways to reduce the amount of data missing in these trials, and make the best use of the data we do have available. This is so we can make sure studies produce valid results that help reduce the uncertainty around how palliative care can help others in the future.
Why is missing data a problem in palliative care?
When a study sets out to gather information, but for whatever reason it’s never collected, this is known as ‘missing data’.
As a palliative care doctor, my patients tell me they want to take part in research in order to help others. These remarkable individuals donate their limited time and energy in order to give something back through research - even when this means spending less time with friends and family, or simply resting.
Sadly, some people become too poorly to finish a study or are unable to provide all the information needed for other reasons, such as moving home.
On average, just under a quarter of the information we need at the most important point in a study (known as the primary end-point) is missing in palliative care trials (see the full research article ).
This is over twice as much as that found in other areas of healthcare (see research ).
Why having all the data matters for valid results in medical trials
Missing data is common in randomised controlled trials – studies that investigate whether one treatment (eg treatment A) is better than another treatment (eg treatment B).
To carry out these trials, we have to make sure the people receiving treatment A are similar to those receiving treatment B.
This is so that if one group does better than the other group, we can say this is likely to be due to the treatment and not because of other differences between the people who were in the groups (see this article in Healthtalk about understanding allocation ).
How a randomised controlled trial works
In a randomised controlled trial, we make sure the two groups are similar by allocating participants at random to a treatment group. You can see the steps in a randomised controlled trial in the diagram below.
Why missing data causes biased results
Missing data can affect the balance of the groups in a trial so they are no longer similar. This means we are less certain that any differences we find between those who get treatment A and those who get treatment B are due to the treatment alone.
This can happen, for instance, if each group has a different number of people with missing data.
For example, if two pears and one apple have missing data in Group A and only one pear and one apple in Group B, we are no longer comparing groups made up of similar numbers of apples and pears:
In this scenario, because most of the data from Group A is from apples, we might mistakenly think that treatment A caused most of the fruit to shrink.
Results can also be biased if data from different types of people are missing from each group. For example, if all the apples have missing data in Group A and all the pears have missing data in Group B:
Even though this time we still have the same number of fruit overall in each group (ie three in each group), we are now comparing apples with pears.
So we might mistakenly say that treatment A helps all fruit to grow. But this would be wrong, because no data from apples are available in that group.
Missing data reduces the accuracy and power of a study
Before trials start, researchers calculate the minimum number of people they need to take part in the study. The more people there are in a study, the more precise the findings will be, and the more powerful the study is to find the effect the treatment has if one exists.
Studies usually try not to have more people than is necessary. Most also estimate how many people will have missing data and increase the minimum number needed to take account of this.
If there is more missing data than expected however, then the study will not have enough power (ie people) to tell whether, for example, treatment A is better than treatment B. So the study cannot do what it set out to do.
Failing to help those who need it most
In palliative care research, a common reason for missing data is that the person becomes too unwell to continue to take part. This means the information on how the treatment affects the people who are the most poorly is often missing.
This is a big problem for palliative care teams, because we are responsible for taking care of the people who become too poorly and need to make decisions about how best to treat them.
So if we don’t know if the treatment works for these people, we have to rely on information from people who are not as frail and unwell – and it's difficult to know if this will result in benefit or harm.
Incomplete data is wasted data
In 6 out of 10 palliative care trials, researchers don’t include any of the data from the people who didn’t provide a complete set (see research).
This means the information from those people who could only provide part of the data isn’t included in the results.
For the study, this is wasteful and will often bias the results. But more importantly, this means for those people volunteering their time to take part in research, but who can only provide some of the data, their time and energy is also wasted.
Fortunately, there are ways researchers can make sure this data is analysed. But we now need to find ways to get everyone to use them.
There is no complete cure for missing data, but it’s vital we identify ways to reduce the amount of missing data wherever possible and make the best use of the data we have.
Marie Curie recently hosted a workshop for researchers, clinicians and other partners in research on this subject. We discussed how to reduce missing data, what to do with missing data when analysing results of a study, and how to report missing data clearly so we can understand the impact.