Data mining medical trials – a 'game changer'

By Lynley Hargreaves 10/06/2015 1


Associate Professor Suetonia Palmer
Associate Professor Suetonia Palmer

Diabetic? Got heart disease? Which blood pressure treatment will help you most? With hundreds of thousands of medical trials to wade through, even your doctor might not know. But in a paper in the Lancet medical journal published recently, University of Otago Associate Professor Suetonia Palmer, as part of a worldwide team, uses a new statistical framework to show – for the first time – which drug combinations best protect kidney function. The Royal Society of New Zealand Rutherford Discovery Fellow explains.

How will this change what you can say to patients?

Figuring out what the medical research is telling us is nearly impossible because of the sheer number of trials available all the time. Plugging all the trials into a single analysis will allow me to be much more confident when I discuss with patients which treatments we think will help the most. Instead of eight or nine different drugs to choose from, I can say with more certainty, which specific drug combinations best protects kidney function, protects against dialysis, and may protect against death.

In terms of death, your Lancet paper says that no drug combination did better than a placebo. How can that be?

It isn’t because the treatment is bad. What it means is that we just don’t have the evidence that those drugs have important effects on how long people live. That’s probably because we’re not measuring the right things in our trials. Medical trials have to be big and long term and they are often very expensive to run. So trials commonly measure the effect of a drug on things that happen sooner than patient survival, such as blood pressure or blood tests for kidney function. They don’t tend to measure survival, even in big areas like diabetes, which seems surprising.

What this network approach also allows us to do is tell which drug combinations are nearly there, in terms of evidence. We can say, “these drugs are promising…this is where we need to put our time and money into new research”.

How do medical trials usually evaluate medicines?

Usually to understand the effects of treatments we use randomisation trials, where people are equally likely to receive the different treatments being studied. The only difference between patients should then be the treatment they get. This is helpful when we have a really big trial (involving several thousand people), but is not so helpful if two drugs have never been compared in a research trial, or if there are a hundred different trials all saying different things about two drugs – it becomes very difficult to understand which drugs work best.

With some drugs, such as antibiotics, we shouldn’t be randomising people to receive placebo to see if the antibiotic is better because we know they work. So trials will often compare an older drug with a new one. But we do still need to understand relative benefits of a treatment compared with no treatment. This new statistical framework to measure all drugs together really is a game changer. It allow us to understand the relative benefits of drugs and look at a range of options, instead of just comparing one treatment with another. We can say that only these four drugs are better than taking a placebo, even though we don’t have placebo in the trial.

Has this new approach been used in other areas of medicine?

There have been two really fantastic Lancet papers in psychology looking at antipsychotic medications and treatments for bipolar disorder. That’s great because there is such a huge number of treatment options in that area, and it’s really difficult for individual doctors and policy makers to know which drugs to choose.

These studies looked at all the available treatments and ranked them, giving people much clearer options. You can choose which drug will help you avoid certain side effects, such as sleepiness or agitation. So it really makes medicine much more personalised.

What are you working on now?

This approach is only possible because of electronic techniques and data sharing, and things are evolving rapidly – when I first used data mining I ended up with a cupboard in my bedroom filled with scientific papers. In addition, in medicine we just didn’t have the statistical know-how. I’ve realised – through networking as a Rutherford Discovery Fellow – that this approach has been done more in other areas of research, such as ecology. Medicine has finally come to the network approach.

Our techniques are still very labour intensive. A new medical question we’re working on involves 20-30 people on an international team, scanning 5000-6000 individual reports of medical trials, finding all the relevant papers, and entering data for about 100-600 reports by hand. We need to build an international partnership to make these kind of studies easier, cheaper, more efficient, and more relevant.

These interviews are supported by the Royal Society of New Zealand, which promotes, invests in and celebrates excellence in people and ideas, for the benefit of all New Zealanders.


One Response to “Data mining medical trials – a 'game changer'”