statistics.com

[sharon19]

For medicine and AUC >700 is considered good, so I am okay. I have AUC of 760
BUT my H-L test is p=0
I have read about using clustering as a way to increase calibration, but the paper was complex. I read somewhere that the calibration does not matter. I want to write up a paper so I do not know what to do.

My equation is simple
Resuscitation Code Status = AGE + GENDER + SAPS SCORE

Besides making AGE binary, which I have had trouble doing, I do not know what to do.

Ultimately I am interested in the model to derive Odds Ratios to determine ageism or gender bias

Many thanks.

Sharon


Oh, and though I do not think this will work, if AGE (continuous) does not have an interaction term, do I have to repeat the interaction study if I change AGE to binary?

[alethephant]

1. What is "AUC"? Is this "Area Under Curve" in a ROC plot? If so, does "760" mean 0.76?

2. What is "H-L"? Is this the Hosmer-Lemeshow test? If so, you need to describe more about how you carried it out and how many classes you used and how many data you have.

3. AGE is a complicated variable, and you don't describe what range and ages you have, and how you expect them to relate to your RCS response.

4. What are your "clusters"? Are these institutions you sampled at?

Etc.

[sharon19]

Yes to numbers one and two, sorry for my abbreviations.
Age ranges from 15 - 100 exactly (due to de-identification). The mean is around 65 years
I have approximately 17,000 subjects.
I do not have any clusters, it was just a thought. All the data comes from one hospital, but it does come from different ICUs (about 7 of them)


Thank you

[alethephant]

The 7 ICUs are probably clusters (correlation among subjects in same ICU).

You have a wide range of AGE, so if you are going to categorize it, you should use more than 2 categories. Perhaps 4 or more to capture the association.

As to your question of "calibration", which I presume means "goodness of fit" or "stability of fit", validation is a good idea if you have a lot of data.

[sharon19]

what do you mean by validation? Yes I mean goodness of fit.
Thank you,
Sharon

[alethephant]

Validation usually means verifying stability of a model by holding out portions of data and seeing if the model remains essentially the same when fit on different portions.

This can be a dichotomy, such as 70% for training and 30% for validation, or a trichotomy of 50% for training, 30% for design level validation and 20% for final independent validation of the selected model.

Or it can be a more systematic approach, such as jackknifing, bootstrapping or other techniques with a sampling component.

[sharon19]

thank you very much. I have used validation for classification, but I did not realize it meant the same thing here. Thank you

[LeeRain]

You have a wide range of AGE, so if you are going to categorize it, you should use more than 2 categories. Perhaps 4 or more to capture the association.
_________________
import trade lead
brass disks

[Sam13]

Thank you everybody...
Igot many useful information from you....
_________________
best way to get rid of acne

[zihad36]

I have approximately 17,000 subjects.
I do not have any clusters, it was just a thought. All the data comes from one hospital, but it does come from different ICUs (about 7 of them)



online bingo

[raat475]

I read somewhere that the calibration does not matter. I want to write up a paper so I do not know what to do.

My equation is simple
Resuscitation Code Status = AGE + GENDER + SAPS SCORE

Besides making AGE binary, which I have had trouble doing, I do not know what to do.



retouching

[zia111]

I have read about using clustering as a way to increase calibration, but the paper was complex. I read somewhere that the calibration does not matter. I want to write up a paper so I do not know what to do.




no deposit bingo