Episode 300: 299. When Do You Use a Cluster RCT?

2024-06-07

When you want to do an RCT but you realize there might be some cross contamination so instead of randomizing individuals you randomize cluters or groups Lets say you want to do fluids--- what is better LR or NS instead of doing individuals and room you just put whole hospitals in either LR or NS Or you want to test if mask work for some virus—you would say this city gets all the mask in the world and this city gets zero mask…you know that If you just gave some individuals in the city ...

When you want to do an RCT but you realize there might be some cross contamination so instead of randomizing individuals you randomize cluters or groups

Lets say you want to do fluids--- what is better LR or NS instead of doing individuals and room you just put whole hospitals in either LR or NS

Or you want to test if mask work for some virus—you would say this city gets all the mask in the world and this city gets zero mask…you know that If you just gave some individuals in the city a mask then maybe someone who give their extra mask to their friends and you would have cross contamination meaning that that individuals in the group end up getting the intervention you are trying to test.

Cluster RCT are GREAT for logistics and trying to figure out logistically can something or does something actually work

However cluster RCT can be harder to analyze because the clusters may not always have the same exposures or the confounders might not be equal between clusters

In our mask example that we just talked about lets assume that one village or city never has a case of covid—will it doesn’t matter if they get mask or not because they are so isolated they never get a case of it. However if you had a perfect individual RCT the number of people in that never covid village would be exactly equal.

In the IV fluid example lets assume that one hospital gives their patients on average 20L of fluid per admission while other hospital is more reasonable in their fluid management. We know the more fluid or volume overload you are the worse your outcomes. On an individual level this wouldn’t be a problem as the same number of people would be randomized to the hospital with excess fluid exposure. So for cluster RCT while you might be able to randomize the type of fluid that cluster gets you cant control for other confounders which may throw off your result, things like exposure to the amount of fluid or type of nursing care. These excess confounders balance each other out in a individual trial as ideally individuals will have everything else the same, from volume of fluid to nursing care and the only difference will be the fluid.

Comments (3)