We had a nice way of memorizing this, finding the ad/bc method somewhat difficult to remember at times, especially when lumping it with specificity and sensitivity, which were also taught with the ABCD box. Instead, we would say, "(BOTH x NONE)/(SOME x SOME)." where BOTH = exposure and the disease, NONE = no exposure and no disease, SOME = either exposure without the disease or disease without exposure (it doesn't change the outcome in the denominator since it's simple multiplication, for example 2 x 3 is = 3 x 2). Can't go wrong.
The reason the cross-product works is that you are dividing the odds of having been a smoker 90/10 by the odds of not being a smoker 10/90 (it's a ratio). Division is multilication by the reciprocal which is that you do with the cross-product. and as comments say, any numbers will work. Very nice presentation.
Terry, could you please explain the difference between, "If you have lung cancer, then you are 81 times more likely to have smoked" AND "if you smoke, you are 81 times more likely to get lung cancer." Or is there not a difference? I ask because the way you phrase it does not infer that smoking causes cancer, but that the behavior and disease are associated. However, there's an example in the text book that interprets the OR this way: "smokers are 1.6 times AS LIKELY to develop LC as nonsmokers." Any clarity? Thank you.
Hey Anne, I think you might be mixing up "Odds Ratio" (presented here), which is the mathematical inference drawn from case-control studies - "how much more likely were you exposed to the risk factor if you have the disease?", versus "Relative Risk", which is the mathematical inference drawn from cohort studies - "how much more likely would you have the disease if you are exposed to the risk factor?" The first study is retrospective with the latter being prospective
The explanation which you mentioned is that "If you had lung cancer, there is 81 times more likely to smoke if you did not have.". I find it difficult to accept. Would not it be another way around. If you smoke , there is 81 times more chance (odds/likely) to get lung cancer than if you do not smoke. Please correct me If I am wrong.
I was stuck there too. I just assumed he meant if you smoke, you're 81 times more likely to get lung cancer than if you don't smoke.. OR people who have lung cancer are 81 times more like to have developed it from smoking than from other causes. May be he is using that language because it was a case control study, we go from disease back to exposure.
How do you add multiple odds ratios? Let’s say a patient has multiple risk factors for developing schizophrenia, each risk factor with a 2 fold possibility. (Ie having an older father, smoking cannabis, birth complications). Would you add them all together or multiple them?
Is it necessary to have the same number of test and control group? I have 60 CRC patients as test group and 46 healthy persons as a control group. does it works the same way?
Can you do a video on retrospective cohort and prospective case control studies. I thought all retrospective studies are case control studies and all prospective studies are cohort. I got that question wrong on a Q Bank.
A cohort study can be retrospective or prospective depending on when you assemble your cohorts (in the past is retrospective and now is prospective). All cohorts are followed from cohort assembly forward looking for outcomes. All case control studies are retrospective even though some label them as prospective. You assemble your cases in the present and go back in time to look for exposures. Hope this helps. I cover this in a video on observational study design.
Thanks for the great explanation. You conclude that if you have lung cancer, you are 81 times more likely to have smoked than if you didn't have lung cancer. Is this bi-directional? i.e. if you smoked, you are 81 times more likely to have lung cancer than if you don't smoke?
1000 patients with HIV infection with TB: 100 patients having relapse compared with 1600 patients who do not have HIV infection & 80 patients from that also having relapse of TB within 1 year of completion of treatment. Find odd's ratio of above samples
The odds ratio calculation is correct, but the interpretation is incorrect - all you can say is that there is a very strong correlation. I do not think the "81 times more likely to smoke" is correct. The chances of having lung cancer if you smoke is 90/100 = 0.9 The chances of having lung cancer if you do not smoke is 10/100 = 0.1 Relative risk is 0.9/0.1 = 9 - So you are 9 times more likely to have smoked if you have lung cancer.
Why would you calculate a relative risk in a case control study? It can't be done as a relative risk is a ratio of incidence of outcomes in exposed and unexposed groups. There is no incidence in a case control study
Terry Shaneyfelt As you said . . . "odds ratio = odds of being exposed to smoking if you have lung cancer/ the odds of being exposed to smoking if you do not have lung cancer" Odds of being exposed to smoking if you have lung cancer = 90/10 = 0.9 odds of being exposed to smoking if you do not have lung cancer = 10/90 = 0.111 So the ratio of the odds . . . OR = 0.90 / 0.111 = 81.0 So the *odds* of being exposed to smoking if you have lung cancer(0.9) certainly are *81 times* the *odds* of being exposed and not having lung cancer (0.111) . . . However, as soon as you say "more likely to smoke" you are talking probabilities (risk) and not odds so you cannot use the odds ratio. As calculated above, the probability or risk of getting lung cancer if you smoke is 9 times that if you did not smoke. If you say that in a population of 100 non smokers 10 will get lung cancer (i.e. a probability of 10%) then if you say that "smokers are 81 times more likely to get lung cancer then the probability is 81*10 = 810% . . . which cannot be correct. Other than giving the direction and degree of an association it is very difficult to give ORs a simple meaning.
Damien the relative risk of having lung cancer given you have smoked is equal to: P(lung cancer | smoked)/P(lung cancer | not smoked) whereas you are finding: P(smoked | lung cancer)/P(smoked | don't have lung cancer). So although you're 9 times more likely to have smoked if you have lung cancer that doesn't mean that you're 9 times more likely to have lung cancer if you've smoked (you could be 81 times more likely to have lung cancer if you've smoked, or 5 times, or 3 times etc.). That said you're right that the relative risk of having lung cancer is not equal to the OR.
Terry, you conclude “If you had lung cancer, you are 81 times more likely to smoke, than if you didn’t have lung cancer”. Interpretation of OR: “The ODDS of a lung cancer patient having smoked are 81 times higher than the ODDS if you did not have lung cancer.” Here the cancer patient risk of having smoked was 90%, and the non-cancer patent risk 10%. The relative risk (RR) is 0.9/0.1=9: “If you have lung cancer, you are 9 times more likely to smoke, than if you did not have lung cancer.“ Or…?
I was going to say the same thing. The video is nice but that one line is wrong. It's not 81 times more likely (as you, AK here notes), it's that the odds are 81 times greater.
I don't think so. I'm by no means an expert in the field, but with the given numbers, I think that "If you smoke, then you are [9x] more likely to get lung cancer than if you do not smoke". This is straight from the table in the video, as of the people that smoke, 9 times more have lung cancer, than don't have lung cancer.
Daniel Petersson NOT in a case control study. You don't calculate risk, just odds. In a case control it's simply finding a group of people with a disease like cancer and seeing if they smoked. A cohort study on the other hand would follow smoking and the risk that has to getting lung cancer. Disease comes first in a case control and past exposure is detected. In a cohort exposure comes first and then you follow to see if that led to a disease. That's where the risk comes from.
This explanation restricts the understanding of odds ratio's to basically using a certain formula. Thus it is not a good lesson. You will learn how to apply a certain formula, but you will not understand the principle of odds ratio's from this.
We had a nice way of memorizing this, finding the ad/bc method somewhat difficult to remember at times, especially when lumping it with specificity and sensitivity, which were also taught with the ABCD box. Instead, we would say, "(BOTH x NONE)/(SOME x SOME)." where BOTH = exposure and the disease, NONE = no exposure and no disease, SOME = either exposure without the disease or disease without exposure (it doesn't change the outcome in the denominator since it's simple multiplication, for example 2 x 3 is = 3 x 2). Can't go wrong.
Thank you thank you thank you!
Thank you so much for this unexpected help. 🙏🏻🙏🏻🙏🏻🙏🏻
Theodore Nelson now that is an awesome mental tip
OMG you are amazing. thx
Thanks man!
Brilliantly simple exposition of a concept I have found daunting.
The reason the cross-product works is that you are dividing the odds of having been a smoker 90/10 by the odds of not being a smoker 10/90 (it's a ratio). Division is multilication by the reciprocal which is that you do with the cross-product. and as comments say, any numbers will work. Very nice presentation.
Thank you for this!! It was posted 8 years ago but still helping students like me
Such a simple way of explaining this! thank you
I don't know why my teachers like complex formulas when getting OR was this easy !! . Cross product works magic. Thanks
It does help me, soooo helpful!!! A million thumbs up from me.
Thanks!
From Nigeria
Great vid, and a way better explanation than my professor gave in class. Thanks a bunch for posting this vid :)
thank you! it took me hours to study this on the book, only 3 minutes here. :)
You`re a legend in this game.
Thanks for the help.
Thanks for a short and to the point explanation! In a Epi class this helped me!
Terry, could you please explain the difference between, "If you have lung cancer, then you are 81 times more likely to have smoked" AND "if you smoke, you are 81 times more likely to get lung cancer." Or is there not a difference? I ask because the way you phrase it does not infer that smoking causes cancer, but that the behavior and disease are associated. However, there's an example in the text book that interprets the OR this way: "smokers are 1.6 times AS LIKELY to develop LC as nonsmokers." Any clarity? Thank you.
was thinking the same..
Since this is a case control study you can only estimate "risk" of exposure
i get u
Your videos on OR and RR are helpful for my Epidemiology class. My professor sucks. Thanks
Excellent...no nonsense video...right on the money!
Hey Anne, I think you might be mixing up "Odds Ratio" (presented here), which is the mathematical inference drawn from case-control studies - "how much more likely were you exposed to the risk factor if you have the disease?", versus "Relative Risk", which is the mathematical inference drawn from cohort studies - "how much more likely would you have the disease if you are exposed to the risk factor?" The first study is retrospective with the latter being prospective
Thank you, a nice way to understand OR.
Your videos are nice and helpful. Keep up the good work
That is so uplifting, You could also used as an example the odds of finishing in an oven if you lived in the Warsaw ghetto.
Since this is a case control study you can only estimate "risk" of exposure
altough I am not English or American I did quite understand what you where saying here, Thanks a lot for the explanation.
The explanation which you mentioned is that "If you had lung cancer, there is 81 times more likely to smoke if you did not have.". I find it difficult to accept. Would not it be another way around. If you smoke , there is 81 times more chance (odds/likely) to get lung cancer than if you do not smoke. Please correct me If I am wrong.
I was stuck there too. I just assumed he meant if you smoke, you're 81 times more likely to get lung cancer than if you don't smoke.. OR people who have lung cancer are 81 times more like to have developed it from smoking than from other causes. May be he is using that language because it was a case control study, we go from disease back to exposure.
Thank you Terry, very informative and succinct.
Great lesson! I would like to know if you could show us how to calculate IC95% for OR & RR. Thank You!
Thank you, brilliantly explained.
Very helpful video. Thank you!
With case control studies, do we also need an inferential statistic (e.g. Chi Squared, Fisher Exact Test) to establish statistical significance?
Thanks! Helped me a lot for my exam
This was very simple and great. I understand.
How do you add multiple odds ratios? Let’s say a patient has multiple risk factors for developing schizophrenia, each risk factor with a 2 fold possibility. (Ie having an older father, smoking cannabis, birth complications). Would you add them all together or multiple them?
Is it necessary to have the same number of test and control group?
I have 60 CRC patients as test group and 46 healthy persons as a control group. does it works the same way?
what is ur research area? i am working on rectal cancer
Thanks for you help , you explain very well
Perfect explanation. Thanks!
Yes it does. He is just making an example here but it will work for any combinations of numbers.
HOLY SHIT thankyou, haha this is a lot better than my lecturer
Perfect video thank you
Thanks a lot Sir ! Very well explained.
Thank you sir ❤
I think its better to use (90/10)/(10/90), instead of multiple like that, I dont really understand the meaning by that formula
Thank you sir.
Thank you so much! Very helpful!
Thank you very much
An excellent video
Can you do a video on retrospective cohort and prospective case control studies. I thought all retrospective studies are case control studies and all prospective studies are cohort. I got that question wrong on a Q Bank.
A cohort study can be retrospective or prospective depending on when you assemble your cohorts (in the past is retrospective and now is prospective). All cohorts are followed from cohort assembly forward looking for outcomes. All case control studies are retrospective even though some label them as prospective. You assemble your cases in the present and go back in time to look for exposures. Hope this helps. I cover this in a video on observational study design.
good explanation! I understand that a case control study looks for exposure so what does a retrospective cohort study look for?
+Spicy Mocha I feel its prevalence while prospective looks for incidence rates
Your letters with the cursor look really cool its sort of like graffiti writing.
Amazing, Thank you very much
explained well, good man
Thanks for the great explanation. You conclude that if you have lung cancer, you are 81 times more likely to have smoked than if you didn't have lung cancer. Is this bi-directional? i.e. if you smoked, you are 81 times more likely to have lung cancer than if you don't smoke?
Since this is a case control study you can only estimate "risk" of exposure
thank u! great explanation
thanks for the video!
nice explanation
1000 patients with HIV infection with TB: 100 patients having relapse compared with 1600 patients who do not have HIV infection & 80 patients from that also having relapse of TB within 1 year of completion of treatment. Find odd's ratio of above samples
Thanks!
thank you so much
The odds ratio calculation is correct, but the interpretation is incorrect - all you can say is that there is a very strong correlation.
I do not think the "81 times more likely to smoke" is correct.
The chances of having lung cancer if you smoke is 90/100 = 0.9
The chances of having lung cancer if you do not smoke is 10/100 = 0.1
Relative risk is 0.9/0.1 = 9 - So you are 9 times more likely to have smoked if you have lung cancer.
Why would you calculate a relative risk in a case control study? It can't be done as a relative risk is a ratio of incidence of outcomes in exposed and unexposed groups. There is no incidence in a case control study
Terry Shaneyfelt As you said . . . "odds ratio = odds of being exposed to smoking if you have lung cancer/ the odds of being exposed to smoking if you do not have lung cancer"
Odds of being exposed to smoking if you have lung cancer = 90/10 = 0.9
odds of being exposed to smoking if you do not have lung cancer = 10/90 = 0.111
So the ratio of the odds . . .
OR = 0.90 / 0.111 = 81.0
So the *odds* of being exposed to smoking if you have lung cancer(0.9) certainly are *81 times* the *odds* of being exposed and not having lung cancer (0.111) . . .
However, as soon as you say "more likely to smoke" you are talking probabilities (risk) and not odds so you cannot use the odds ratio. As calculated above, the probability or risk of getting lung cancer if you smoke is 9 times that if you did not smoke.
If you say that in a population of 100 non smokers 10 will get lung cancer (i.e. a probability of 10%) then if you say that "smokers are 81 times more likely to get lung cancer then the probability is 81*10 = 810% . . . which cannot be correct.
Other than giving the direction and degree of an association it is very difficult to give ORs a simple meaning.
This is the correct interpretation. I hope people have read this.
Remember odds are a ratio of probabilities. Odds = probability÷ 1-probability and probability = odds ÷ 1+odds.
Damien the relative risk of having lung cancer given you have smoked is equal to:
P(lung cancer | smoked)/P(lung cancer | not smoked)
whereas you are finding:
P(smoked | lung cancer)/P(smoked | don't have lung cancer).
So although you're 9 times more likely to have smoked if you have lung cancer that doesn't mean that you're 9 times more likely to have lung cancer if you've smoked (you could be 81 times more likely to have lung cancer if you've smoked, or 5 times, or 3 times etc.).
That said you're right that the relative risk of having lung cancer is not equal to the OR.
Thanks a lot
wow thank you!
it should bi-directional. but it is correlation rather than causation
Terry, you conclude “If you had lung cancer, you are 81 times more likely to smoke, than if you didn’t have lung cancer”. Interpretation of OR: “The ODDS of a lung cancer patient having smoked are 81 times higher than the ODDS if you did not have lung cancer.” Here the cancer patient risk of having smoked was 90%, and the non-cancer patent risk 10%. The relative risk (RR) is 0.9/0.1=9: “If you have lung cancer, you are 9 times more likely to smoke, than if you did not have lung cancer.“ Or…?
I was going to say the same thing. The video is nice but that one line is wrong. It's not 81 times more likely (as you, AK here notes), it's that the odds are 81 times greater.
Good
Thankyou :)
thanks
thanks.
I don't think so. I'm by no means an expert in the field, but with the given numbers, I think that "If you smoke, then you are [9x] more likely to get lung cancer than if you do not smoke". This is straight from the table in the video, as of the people that smoke, 9 times more have lung cancer, than don't have lung cancer.
Is it not more convenient to calculate the OR of lung cancer & exposure? So you formulate the risk of developing lung cancer if you smoke.
Daniel Petersson NOT in a case control study. You don't calculate risk, just odds. In a case control it's simply finding a group of people with a disease like cancer and seeing if they smoked. A cohort study on the other hand would follow smoking and the risk that has to getting lung cancer. Disease comes first in a case control and past exposure is detected. In a cohort exposure comes first and then you follow to see if that led to a disease. That's where the risk comes from.
This explanation restricts the understanding of odds ratio's to basically using a certain formula. Thus it is not a good lesson. You will learn how to apply a certain formula, but you will not understand the principle of odds ratio's from this.
Thanks but you always forget that there are people whose native language is not English. Could you please speak slower and clear next time
Im 10 im in this subject and i dont underatand anything this guy said
Not professional.
Thank you so much