I am trying to determine which is the best statistical test to evaluate the validity of an outcome probability predicted by a clinical calculator. I have the mean (+/- SD) outcome probabilities for the cohort (continuous variable) and also the actual outcome data (dichotomous variable). e.g.- Mean (+/-SD) Predicted Probability of respiratory disease= 45.4 (+/- 8.6)%
Actual incidence= 63
Cohort size 212 subjects.
If data related to a design you need to follow it, if data on two variables, paired t test or chi square fit. Regards
Hi, Bharath Srivatsa . I suspect to get any answers to your question, you would need to explain a little more about what are you trying to do. Can you provide some toy data similar to your data ? Where is the calculator you are using ? What is it you are trying to do ? Calculating mean and standard deviation don't require a statistical test.
Hello Bharath Srivatsa
See column 1 and 2 in the attached image. I think you have something like that.
Column 2 (PRE) is the probabilities you are talking about. Note, we do not need their mean or standard deviation.
Convert column 2 to column 3 (PGR). Based on Classification cutoff = 0.5.
It means that it is very simple if PRE > 0.5, it will be converted to 1 in column PGR and if PRE < 0.5 it will be converted to 0 in column PGR.
After that, you have a simple task ahead of you. Draw a Classification table and get the sensitivity and specificity in it. See the attached image.
good luck
Hi Mr Ghoodjani,
Thank you for your response.
I understood your methodology, however, needed to clarify these items-
1. Can you please spell out "PGR"? What does it stand for?
2. Can you give me some reference in support of this methodology i.e. using 0.5 as the cutoff? I have not seen this approach in my limited experience of reading medical journals.
Best!
Bharath Srivatsa, MD
Hello Bharath Srivatsa
1- PGR is the same as Predicted Group
2- In this context, study Binary Logistic Regression (BLR). Of course, you are in the middle of a story. Because you have suddenly encountered numbers called PRE (Predicted Probability).
They are derived from a logistic regression model. Of course, as far as I think.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS