PakMediNet Discussion Forum : Biostatistics : which test to use
i'm writing a synopsis on EVALUATION OF APACHE II SCORE as a predictor of mortality and legth od stay in med icu patients.it has been sent back for correction n i'm confused in a few areas
1)how do i show in graphical form the no. of days spent by the patients vs thier respective apache ii score?
2)which association test should i use to show relation between a high apache ii score n motality and which to show association between high apache ii score and a lengthy icu stay
3)how do i know how much is the sample size.i only know that in most of the studies it is mentioned that apache ii score shows with 100% proabability that the patient will be shifted out og icu if the score is <10.and the area under reciever operator curve is 0.82
4)i'm supposed to show that the confounding variables are controlled.but i dont know what could be the confounding variable in this study?improper clinical techniques?faulty apparatus??
so plz help
Posted by: saadia79 Posts: 4 :: 29-10-2006 :: | Reply to this Message
Hi saadia;
I will try in a couple of days to help you in statistics, i will try to ask a statician for you and i have a good backgroud in it.. but i have a small problem in understanding the purpose of the study.. so would you please email me more deatils on your case, and i will try to help you.
thank you
younis
Posted by: palestine Posts: 7 :: 29-10-2006 :: | Reply to this Message
If I understand your questions correctly, here is what you need to do:
1. Your outcome measure (Days) will be on y-axis, your exposure measure (APACHE) will be on x-axis. MS Excel should be able to make this graph.
2. You can use odds ratio, or correlation coefficient, which ever you like better/easier.
3. smaple size calculation is more complicated. Looks like you don't know what it mean. Read a good article on it. You will need estimate of event rate for mortality, and estimate of variance of days in ICU in addition to many other, relatively smaller things.
4. You will need to use logistic regression for controlling for confounding in estimating mortality, multiple linear regression for controlling for confounding for days in ICU.
Posted by: rqayyum Posts: 199 :: 30-10-2006 :: | Reply to this Message
RQayyum has already answered most of your questions.
Regarding point 2. I am assuming you will be using apache score as a continuous variable. If true, you cannot compute Odds ratio, instead you will have to do a t-test to show association between mortality and apache score. However, if its acceptable practice, then you can split apache score into a two level categorical variable, and calculate odds ratio.
For association between score and length of hospital stay, you can calculate pearson's correlation coefficient, or if you dchotmize your apache score, you can do a t-test.
Sample size is not something that can be adequately answered in a short post. You will need to consult a book of statistics. I will recommend Biostatistics by Bernard Rosner.
If you need to control for one or two confounders, stratification (assuimng a large enough sample size) will be as adequate as logistic regression.
I will strongly recommend seeking help from a statistician or 'qualified' epidemiologist.
Good luck!!
Posted by: siddiqi Posts: 19 :: 01-11-2006 :: | Reply to this Message
THE apache ii score is calculated with in the first 24 hours and it can be from 0-71.for one patient only one score is calculated in that admission .no new score can be calculated again unless the patient is re admitted. but its divided in 8 gruops 0-4 , 5-9 ,10-14 ,15-19 ...30-34 & >34.the legth of stay will be in days.the standard table used for interpretation of score for predicting mortality is :
INTERPRETATION OF APACHE II SCORE
Apache II Score Interpretation
0-4 ~4 % death rates
5-9 ~8 % death rates
10-14 ~15 % death rates
15-19 ~25 % death rates
20-24 ~40 % death rates
25-29 ~55 % death rates
30-34 ~75 % death rates
over 34 ~85 % death rates
so i'll be showing %age of mortality in each of these 8 groups in the form of a simple bar graph .the objectives of the study are:
•To correlate APACHE II score & the length of stay of patients in Medical Intensive Care Unit shifted from Medical D Ward.
•To compare the expected outcome(mortality) with the actual outcome.
i've gone through lots of websites after bieng advised to read a good article about sample size but realy!nothing helped or may be statistics's realy difficult.i've lots of formulae by now but dont know for sure which one should i use.i've finished correction of my synopsis but i'm realy stuck with sample size and i'd appreciate if there's something more to analysis of the data & tests used coz i guess i gave too scanty info about my study the first time .thanks for the help
Posted by: saadia79 Posts: 4 :: 01-11-2006 :: | Reply to this Message
Saadia,
There is no substitute for books. Internet is great, but cannot replace books. The widespread dissamination of information via internet is its greatest advanatge, but also the disadvantage.
Also, there is no substitute for hard work either. For sample size you have to either consult a textbook or a statistician, preferably biostatician. I would have been happy to help you out but I am sitting several thousand miles from Pakistan.
You used the word Medical 'D' ward. Are you in Karachi at Aga Khan. If yes, you can send an e-mail to statistical consultanats group based at CHS. A few very competent faculty members voluntarily help out AKU affiliates in statistical issues, free of cost. Look up their e-mail and setup an appointment. Be respectful please. The faculty doesnt have to do this, they are being kind to offer this service.
Going back to your question. It seems you dont have any problem in putting together the graph. As for the number of days spent in hospital, do you want to see that with higher score the length of stay goes up (or down, since high mortality also means shorter stay), or do you want to compare if the number of days spent in hospital by patients within each group is coniostent with what is reported in literature or not? The difference in research question traslates into use of different statistical tests.
A simple way to compare whether your observed and the expected mortality rates are the same within each category or not would be to compare two proportions, within each category. Which has its own limitations but will provide some evidence to support or refute your hypothesis.
Like I said, you do need to consult a statistcian, I am sorry, but short cuts will not help.
I wish you best of luck!!
[Edited by siddiqi on 06-02-2007 at 03:20 PM GMT]
Posted by: siddiqi Posts: 19 :: 03-11-2006 :: | Reply to this Message