Skip to content
#UnblindingResearch: Randomisation episode artwork

Nov 23, 2018 20 min

#UnblindingResearch: Randomisation

from Take Aurally Hosted by Take Aurally

In this live recording of the ninth #UnblindingResearch session on 'Randomisation' from DREEAM through a randomising the audience to wine gum or jelly babies. By the end of this you should have randomisation sorted! The blog entry containing the slides, #TakeVisually and more info can be found at https://www.takeaurally.com/new-blog-1/2018/11/12/randomisation-1 Subscribe to Take Aurally on SoundCloud and iTunes Remember you can find Take Aurally and NUH DREEAM on both Facebook and Twitter The amazing DREEAM Research Team can be found on both Facebook and Twitter Our next #UnblindingResearch session will be on Blinding on 19th December 1000-1100 in DREEAM. Next week's episode will be our third Redthread special!

Episode details come from the podcast's RSS feed.

Embed this episode

Ready to play

#UnblindingResearch: Randomisation

0:00 20:44
of matches

Transcript Auto-generated, may contain errors

This isn't Harry Potter, I'm mixing up my themes. What's this about, Louise? What does this slide mean? Oh, it's a nice friendly environment and we're all free to say what we want. Because there are two stormtroopers holding hands looking at the sunset, which is us, metaphorically, looking at the sunset of research. Or the sunrise of research, because the sun never sets on research. So this won't take up a full hour, but I'm just going to talk about some of the randomisation principles and some of the terms you guys may have heard about.

gathered and I'm going to randomise you several times and we'll see what you get so we've got a choice of two today winegums and jellybabies who prefers winegums yes come on two people who prefers jellybabies that means we get two babies yes there we go come on Chris don't care I eat sweets don't care you're going into this trial completely open-minded I don't care I just want to help research anything do you have a preference I don't particularly like either okay so you're not particularly sure if you want to be randomised now oh yeah I eat sweets So why do we randomise guys? Do you have a scary part? So it's fair.

The risk of randomisation is random. Yes. So you don't get people picking up which intervention they want. So you probably have everybody picking up the actual treatment and not the placebo. You don't get the study people go, I think this person's not going to do it anyway. So which trials use randomisation? RCTs. RCTs, randomised controlled trials, OK. And what do they normally involve? What's the trial model?

Intervention versus standard care of placebo. Yeah, absolutely. So it's usually, this is our current standard, and here is a new thing that we think could be better, or at least not inferior. Or we're going to try TXA, because everybody does TXA at the moment, versus placebo to see if that helps in whatever need. Good, cool. And if you say to patients, there's this new round of treatment, they may well always want to go for the new treatment, so there'll be a bias there. But like Chris said, you may be setting up your trial, and you may know that you only want nice, young, fit people to be in your trial, and the older, sicker people, they can have a placebo, and then it will show that you're an amazing new treatment.

Cool. What models of randomisation do we use in our current trials, you guys, when you're randomising a patient, what usually happens? Telephone. Telephone? How does that work, Lucy? So you usually have a generic phone number that takes you to an automated randomisation service and you'll have a site number.

I'm not sure how that works, but then it depends what blocks they are, so sometimes they might be randomised one to one, but they may be randomised in a different way. There's also website randomisation, so you can put in the patient's details, and they'll say this is the code, and you just ask the code, that's fine. It's like a big bingo ball thing. Although it's not as random as that sometimes, so sometimes you get randomised, it is still random, but you'll get randomised based on age. So you don't get a lot of, so they want to spread the intervention out across the whole agency, a stratification. So depending on what your age is, then you'll get, so it's kind of, yeah. Does that make sense? It's like a bingo ball, but it can be possibly. It's a very complex algorithm that the computer can do instantaneously. It would take us a long time. From what you stratify, what patients you do. So from a running a trumpet, you don't want many stratifications. Whereas if you think you're going to get a thousand people in, you'll say, oh, yeah. Yeah, that's it. Cool. Envelope.

Randomisation, that's the thing as well. So you just pick out the envelope, you agree to be in the trial, that's great, and then you open up the envelope and it says what you're going to give the patient. It can easily be compromised. You can always hold an envelope up against light. Well, actually, the tips one's made of cardboard. There you go, exactly. So you can have a look and go, and I don't quite like the look at that one unless you make it completely opaque. Good. There's also next box randomisation, so you literally just have a store of boxes, and it's just literally the next one you pull out, and that's going to be...

Good. And obviously there's a telephone, there's a website, there's an Xbox, there's an envelope, it's all different things. Cool. So we're going to use some terms already, there's just stratification stuff. So we're going to do some of those randomizations now. So what I'd like you all to do is stand up, please.

So, if you get heads, you go and have a whack-on. If you get tails, you go and have a whack-on. If you get tails, you go and get a giddy baby. Does that sound alright? Heads, whack-on, tails, generally. You can have one, yeah, have one, please. That's what they're there for, tails, great. There you go. So take a seat, take a seat where you're at. Have a suite.

Do you have to randomise these two results? You can have whatever suite you want. I'm not adding extra layers in randomisation, that'd be too bad. So, good randomisation? The best. So if I was doing a trial on which suite makes you the most productive, was this a good form of randomisation? You were my cohort, you were my participants, and I randomised you.

It's not quite even though. It's not quite even. So this is what's called simple randomisation. Essentially it can be just a coin toss. You can pull out a card, for example. It's a binary choice. You could use a random number generation. You could have 80 in this and 20 in this. Well, absolutely. So this is great. A bit like Chris was saying about sample size. This is great if you've got a thousand people. Because eventually you are going to get a 50-50 split. Eventually, close enough. Whereas on a small group, I've got three over here in this sample having wine gums. And you guys are over here in this arm for my trial having j-babies. So we've got five to three split.

So if I find that jelly babies make you the most productive, I've skewed my data because there's more of you in that trial, so you'll be doing more work. So simple, good for large, not so good for small. Have you all had a sweet? Thank you. Good, marvellous. If you want to come back into the middle, let's do another source of randomisation.

So this looks a bit better, doesn't it? I've got a four and four. So this is what we call block randomisation. So block randomisation is say you've got a population like this and it often blocks a four. So what you do is you split your participants into fours, and you'll have random codes. So say A, B, B, A, so ABBA, ABBA, ABBA, and ABBA. But essentially, if you notice, after each block, you get an equal number of people going to each arm of your trial. So what this guarantees is you get an even split between both arms of your trial. So say A is wine gums, B is jenny babies, you've got an equal split, as we have here, four and four.

Everybody had another spoon? Good. Do you want to stand back up in the middle? Do a bit more. We are. Right. So, I need you to... People who prefer wine gums on this side, please. People who prefer jelly babies on this side, please. For those of you who are a bit in the middle, you need to make a choice.

So once again, heads for wine gums, tails for jelly babies. So that's an example of what we call stratified randomisation. To say you know that this is potentially a confounding factor in your trial, age could be one of them. To say you're doing a trial that's looking at rehab outcomes in surgery, and you were doing a new intervention and wanted to see the rehab outcomes. But you know that age is going to be a factor in that. If you're older, you've got more home mobility. So what you do is you first do a stratification. So you separate those people out into a different arm. And then you do a simple randomisation based on that. So here I just went on who prefers which suite and then randomised you. So that could then go. So I know that some nurses prefer wine gum, some nurses prefer jelly baby. So if I was looking at nurse productivity, I would first stratify you and then randomise you as a way of getting around that. Does that make sense? So alternatively, you could have a cohort of Hogwarts students and you want to see if you can stop some of them becoming evil.

and something's staying good, you know that Slytherins are more likely to be evil than Gryffindors, so you separate out your Gryffindors and your Slytherins before you then run to life. Does that make sense? This is like the coolest whole game ever. Thank you. Does that make sense so far? So we've looked at simple, we've looked at block, and this is stratified. So, kind of classic. Yeah. So you can have your separate stratified results, but then you look at your overalls, so you can say that I've got my intervention versus my placebo arm, but you mentioned that you stratified. Yeah. Yeah.

They're called covariates, so it's just a way of getting around those, basically. Because otherwise, you can say, well, it's been messed up because I've got patients who are younger and fitter mixed in with patients who are older and not as fit. But if you do it this way, it sort of helps that. So what is cluster randomisation? So in cluster randomisation, you have a group of participants, say it could be the first 10, then the second 10, then the third 10, etc. And that is your cluster. So you would be my cluster. And that is a group of participants. And you would get either my guns or jelly babies.

and then the next group who came in would get the other one, etc and so on. When do you think you might do a cluster like that when you've got a large group? So we did the POPI study, which was looking at psychological interventions for patients and intensive care, and that was cluster randomised for sites. The intervention was providing psychological support for patients and intensive care. So if you randomised within a site, so some of the patients received psychological support and some of them didn't.

the people that are trained to give psychological support, if they were then looking after a patient that was, it was randomised to not have any support, you can't guarantee that they are subconsciously giving them their support, so it was done by sight, so that you're either given psychological support or you won't. So that's a really good example of that. So, yeah, I think it's those ones where you've got an intervention that needs a large number of people. So another one would be, say you wanted to do an educational event for people with diabetes, for example. Having one person going to every body participants' house to give them a talk on how to optimise their diabetes control would just completely be inefficient, you couldn't do it. But if you had an educational event and you invited 100 people to that event and held it, and then you had another 100 patients who didn't have that event, those would be your two clusters. Because you have that one single event, you know, psychological intervention, you can then have that place. Does that make sense?

So if you guys come back into the middle, you're going to be my cluster. So heads for wine gums, tails for jelly babies. See what everybody gets. Everybody gets jelly babies. That's an example of cluster randomisation. You would be my cluster, and we'd see how you would all do now, giving you a jelly baby, versus a cluster of nurses who've been given wine gums. So that's how we would do it. And there are scales for looking at how varied individuals are within a cluster. It's called the intra-cluster coefficient. And then finally, we'll look at...

One to two or one to three, so we've looked really at trying a one to one randomisation, so one goes to one group, one goes to another group, and trying to get things as evenly matched as we possibly can. One to two, as the name implies, one goes to one arm, two to another, or even one to three, so one goes to one, three goes to the other arm. Why do you think we might use this sort of randomisation? Because it seems a little bit weird. If the placebo are really well documented as to what the results are for that, it was actually...

So if you're more interested in your intervention, especially if you've got different doses for your intervention, so this is often used if you're not entirely sure of the efficacy, the correct dose. So this way, if you've got more into one intervention, you can then look at different doses of whatever your new drug is in that arm. Why else wouldn't you do it? You have three intentions.

Absolutely. So again, you're into dosing regimes, if there's some uncertainty about that. Another one is cost. So if one arm is particularly cheaper than the other, you might go more towards the cheaper arm, so as a way of keeping the cost of your research down. Sounds a bit dubious. More than likely, the intervention is going to be where the cost is. Although sometimes the placebo is more expensive. So if there's a great discrepancy, you might put more towards the more expensive arm. I'm not entirely comfortable with that idea, but that is one reason.

Another one is recruitment. So if you think that if your subject has a three times greater chance of getting the new wonder drug, they are more likely to sign up for your study, and that's opened up to them in the literature, that is one reason to do three to one. Again, you can question how ethical that is, and if there is that much uncertainty about the benefits and the risks, I wonder why you're doing that trial in the first place. I'd imagine that it's all good RCTs as soon as you want to run, is it? I imagine so, yeah. I think this is one of those things that you have to be very, very clear in your research as to why you do what you're doing. So can you all stand up? So if we'll do a, shall we do a two-to-one?

Shall we do a two to one randomisation? So if you go wine gum, wine gum, jenny baby, wine gum, wine gum, jenny baby, wine gum, wine gum. There we go, do you want to go to your places? Do you want to take a seat and have whatever sweets you want now? Our next session is the 19th of December. I've got to work out a Christmassy theme for blinding. What do you think the pun is this month? There's always a pun. There we go. Thank you guys. Thank you for letting me randomise you.

Frequently asked questions

How long is this episode of Take Aurally?

This episode is 20 minutes long.

When was this Take Aurally episode published?

This episode was published on November 23, 2018.

Is there a transcript available for this episode?

Yes, a full transcript is available for this episode. You can read the complete transcript on the episode page.

Can I download this Take Aurally episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.