OHDSI Home | Forums | Wiki | Github

How to explain preference score , the results of PLP,?

(panpan) #1

When I read the book of ohdsi and use the atlas to make a prediction, I found that I cannot understand the results, just like the picture show, what is the mean of preference score? where it is from?

(Chris Knoll) #2

I think this video is informative on how you would use the preference score.

The preference score is the probability that a person is in a target population. For prediction, it would the the probability of the person having the outcome, and what these graphs are showing is the amount of ‘overlap’ you would find in preference scores at different preference score levels (ie: the density of a preference score of .25 in between the groups compared to a preference score of .75 between the groups.

In the example you presented, you see that there is overlap between the two groups. This is indicating that the prediction could not discriminate between the outcome and the non-outcome groups. This is bad for prediction, because you would want to have a predictive model that would discriminate from those with an outcome from those without the outcome (that’s the point of a prediction).

For effect estimation, we’d use the same kind propensity score model (where the propensity score is the probability that a person would receive the target treatment), but we would want to see an overlap like in your example because it means that the model can not discriminate between those who would be selected for treatment. If it can’t discriminate, this indicates that the population is similar enough that the comparison group could be used as a counterfactual for purpose of performing a comparative effect analysis.

Hope this helps.


(panpan) #3

Thank you very much, @Chris_Knoll, because current I can not open this web, when I have the access to watch the video, I think I can better understand what you said.