Dealing with cohort study data, I have done logistic regression to find risk factors associated with acquiring a particular disease. I have expressed it as adjusted Odds Ratio and 95% confidence interval. However I am confused as to what should be considered as valid or invalid data?
1. For Odds ratio: I know that below 1 is protective and above 1 is risk. But how much below or how much above? Is 0.5 or 0.8 considered preventive; is 1.2 or 1.5 considered a risk?
2. For confidence interval: What range is considered too wide? If the difference between means is greater than 0.5, does it become insignificant? I have found 95% CI range from 0 -14, which is obviously too large. However, is a 95% CI of 0.5 - 2 considered too large also?
If OR is less than 1 , this factor will be preventive, and opposite if OR is more than 1, it means that this factor will be a risk factor. Good luck !
This calculator gives oveview:
https://select-statistics.co.uk/calculators/confidence-interval-calculator-odds-ratio/
For diagnostic purpose You need an OR > 10 (f.e. predicition of Rheumatoid Arthritis by ACC-antibodies).
Under some circumstances You will accept an OR > 5 (f.e. prediction of Rheumatoid Arhtritis by chopstick arthrosonography at the best target).
For therapeutic purpose You will choose the therapy with reduces the damage by an OR < 0,1 (f.e. MTX reduces the mortality by ILD-RA to 0,13).
Researchers often report ORs with the 2s-boarders of confidence. Certainly this creates some knowledge for new studies. Mostly epidemiologists are interested in changes of ORs near 1. Within big populations a slight increased risc can get impact for the the persons in power.
If a predictor is a risk factor then the 95% CI of the odds ratio needs to be greater than 1 and if it is found not to be a risk factor then the 95% CI of the odds ratio needs to be less than 1.
I do not share this decision making. The discrepance towards the location "one" is object of comparison of two distributions. It is a common misunderstanding to interpret every OR different from "one" as a given factor.
You know the mean and the standard deviation of Your Gaussian distribution. For a rand propability of 0,1 of Your healthy group You find a Z-value of 0,6745. At this Z-Value the group at risk shows a propability of 0,5. The risc ratio at this point is the OR=5,0.
Thumb rules
ORs higher than 2 win significance.
ORs higher than 5 get admission for clinical tests.
ORs hgher than 10 are good for established praxis for decision making.
Scales of difference
Look:
CRP shows an upper normal border of 0,5 mg/dl.
CRP indicates coronary riscs at a level 3,0-10 mg/dl.
Bronchitis shows the CRP cut-off of 50,0 mg/dl.
Pneumonia shows the CRP cut off of 100,0 mg/dl.
Burning injuries show CRP levels > 200,0 mg/dl.
My good fellow WILLI so often explains the basics of probabilities, riscs, distributions. Yet the researchers forget such basics. There is a wrong believe of differences so very near to NOTHING.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics