So you have the following stata logistic regression results: OR 1.28, 95% CI 0.5999 – 2.7341; and you are looking at the relationship between total alert message received for suspects cases vs mortality. How do you interpet such result? Alert messgae is your predictive variable while death is your outcome variable (death=1. alive=0). your hypothesis is : alert messages received for suspect cases = mortality (2 tailed).
Alert messages cannot predict the occurrence of death in terms of statistical significance.
Since the value 1 falls with in the confidence interval (CI), ie since the CI extends from below one to above one, the observed relation of OR 1.28 may have been just due to chance; ie there is no association between alert messages and death other than just by chance. If the CI was fully above 1 (ie say from 1.05-2.7341, then you can say there is positive relation between aler messages and death. If you still believe that there should be some positive relation, what you can do is make sub groups depending on types of laert messages,type of conditions (heart attack, accident with bleeding, without bleeding, drug overdose, etc, etc,) and analyse again. You can do this only if sufficient number of cases are available.
Narayanan Bhattathiri
Mortality is a time-dependent type variable, and should not be dichothomized (Y/N) in order to perform a logistic regression: it unnecessarily reduces the statistical power. Instead, try a Cox regression using time to death as dependent variable. As "total alert messages" may also be a time-related variable, you could include a time-dependent explanatory covariate to split and discriminate its putative effects (time and alert).
In addition you can find several examples of the above mentioned hints in a lot of published clinical trials. As one of our results you can find it as download in our papers (but unfortunately just in German as an older publication; the newer results in English are not available as downloads): Ch. Nattermann, A. Goldschmidt, H. Dancygier: Endosonographie bei chronischer Pankreatitis. Ultraschall in Der Medizin - ULTRASCHALL MED. 01/1992; 13(06):263-270. See here esp. page Tab. 4 on page 267.
I believe my interpretation is essentially OK;
From Wiki: If a corresponding hypothesis test is performed, the confidence level is the complement of respective level of significance, i.e. a 95% confidence interval reflects a significance level of 0.05.[4] The confidence interval contains the parameter values that, when tested, should not be rejected with the same sample. Greater levels of variance yield larger confidence intervals, and hence less precise estimates of the parameter. Confidence intervals of difference parameters not containing 0 imply that there is a statistically significant difference between the populations.
I had earlier erroneously put 1 instead of 0; but here it is also almost same.
In logistic regression a result is between 0-1, than it closes to 1 there is strong relationship. Your data OR 1.28, 95% CI 0.5999 – 2.7341 is related to OR. It is determination of affecting one risk factor to outcome. In this case it is weak level of impact.
Summary for you data is predictive variable impacts to death, but has the weak level.
So, there can be another predictive variables.
Dear Elmer
We calculate CI with 90%, 95%, 99% etc. probability. In the present case it was 95%. An OR of 1.28 and 95% CI of 0.5999 to 2.7341 means that in 5% of samples the OR will be either 2.7341; i.e. there is at the least 5% probability that the OR is not 1.28. Actually, the probability of OR not being 1.28 is higher because in 95% of samples evaluated the OR may be any value between 0.5 and 2.0, and it was just by chance that in the present sample it happened to be 1.28. What else does the '95' in 95% CI indicate?
I interpret the result at its face as we seem to all agree - the result provides no convincing evidence of effect. But how might stronger evidence be arrived at/lack of strong evidence appropriately handled? Some possibilities: increase the sample size, improve the logistic regression model (e.g. add an interaction term), try a survival model, consider modifying the hypothesis, drop the hypothesis testing part of your reporting, etc. Of course, if you do a lot of hunting around for a significant result you have to make adjustments: state findings as exploratory detail the explorations or use 0.01 rather than 0.05, or adjust for multiple comparisons, or ... You have lots of options to consider.
Interesting question, I have exactly the same problem but for a different topic. I still cannot figure out how to interpret the " 28% increase in the odds of mortality"... can this be converted to "risk of mortality" or absolute numbers?
- Badges
- Science topic
- Similar topics
- Medicine
- Disease
- Infectious Diseases