PakMediNet Discussion Forum : Biostatistics : use of odds ratio to establish association in cross sectional study
use of odds ratio to establish association in cross sectional study
could anyone help me explain how the cross sectional study below was used to establish an association without the use of contols? I am not too good with biostatistics.I thought one needed to use a case control study to calculate odds ratio. How did the researchers adjust for age,weight etc
Association of androgenetic alopecia and hypertension.
Ahouansou S, Le Toumelin P, Crickx B, Descamps V.
Department of Public Health, Avicenne Hospital, Assistance Publique-Hôpitaux de Paris, Bobigny Cedex. France.
Androgenetic alopecia is considered to be associated with coronary heart disease but the explanation of this association remains unknown. Hypertension is highly prevalent in patients with coronary heart disease. Essential hypertension is linked to hyperaldosteronism and spironolactone, an antihypertensive drug which is a mineralocorticoid receptor antagonist, has been used for a long time in the treatment of androgenic alopecia. We recently observed in a double transgenic mouse model that overexpression of a mineralocorticoid receptor targeted to the skin induced the development of alopecia. We prospectively studied the association of hypertension and androgenetic alopecia in Caucasian men. Two hundred and fifty Caucasian men aged 35-65 years were consecutively recruited by 5 general practitioners (50 per practitioner). Data collected included age, androgenetic alopecia score with a simplified Norwood's score (0-4), blood pressure or history of hypertension, smoking, history of diabetes mellitus or hyperlipidemia, familial history of androgenetic alopecia, and treatment. Chi-square, Fisher exact tests and linear regression model were used for statistical analysis. Hypertension was strongly associated to androgenetic alopecia (p < 0.001). Linear regression tests confirmed that this association was independent of age : odds ratio was 2.195 (95% CI : 1.1-4.3). Familial history of androgenetic alopecia was also strongly associated with androgenetic alopecia : odds ratio was 10.870 (95% CI : 4.3-27.1). Other variables (diabetes mellitus, hyperlipidemia, smoking, treatment) were not associated with androgenetic alopecia. We were limited by a relatively small study sample but in this study androgenetic alopecia was strongly associated with hypertension. Association of androgenetic alopecia and hyperaldosteronism warrants additional studies. The use of specific mineralocorticoid receptor antagonists could be of interest in the treatment of androgenetic alopecia.
Posted by: SANKOPosts: 3 :: 12-02-2010 :: | Reply to this Message
Re: use of odds ratio to establish association in cross sectional study
Odds ratio can be calculated in cross-sectional, case-control and cohort studies. Odds ratio is just basically the odds of exposure to non-exposure. Now in a cross-sectional study,you can't establish CAUSATIOn ( you don't know what came first), but you can establish ASSOCIATION ( whether the two factors have some relationship). Thus, you can calculate an odds ratio for association.
I'd like you to also learn an another important fact. You CANNOT establish risk ratio or risk difference from a case control study. The reason being you choose your cases after they have occurred and you get your controls independent of exposure. In fact the only thing that you can calculate is the ODDS ratio. So if someone does a case-control study and describes a risk difference or risk ratio, expect the paper to be rejected.
From a cohort study, you can calculate all three: risk ratio, risk difference and the odds ratio. Since you are following from exposure to disease.
Posted by: aftab.iqbaPosts: 7 :: 14-02-2010 :: | Reply to this Message
Re: use of odds ratio to establish association in cross sectional studyThere are several ways to get odds ratio after adjusting for confounding variables. However, authors have not described any method through which they could have calculated adjusted odds ratio (at least in the excerpt that you posted). I think, instead of linear regression, they had used logistic regression - exponentiation of beta-coefficients from logistic regression gives odds ratio - very easy mathematically to derive this relationship. Although logistic regression is also a type of linear regression on a logistic scale, but no one in literature reports it as linear.
Posted by: rqayyumPosts: 199 :: 14-02-2010 :: | Reply to this Message
Re: use of odds ratio to establish association in cross sectional studyHaving said that, it appears that they have used a scale (Norwood) with only 5 values (0-4), which means that simple logistic regression is not appropriate; they should have used ordinal logistic regression. Now, it is possible that they have used the values of this scale as a continuous outcome variable and used linear regression, in that case, there is no straight forward way to convert beta-coefficients of a linear regression into odds ratio. Perhaps it will be helpful if you can post the statistical methods section of the paper here; otherwise, everything else is speculative.
Posted by: rqayyumPosts: 199 :: 14-02-2010 :: | Reply to this Message
Re: Re: use of odds ratio to establish association in cross sectional study
here is the details of methodology
Methods
White men aged 35-65 years were consecutively included in this study by five physicians (general practitioners). A total population of 250 Caucasian patients were recruited (50 per physician). The physicians were informed that the aims of this study were to evaluate the prevalence of androgenetic alopecia in this population and to study the association with cardiovascular risk factors (history of personal hyperlipidemia, blood pressure or history of hypertension, history of personal diabetes mellitus, smoking), familial history of androgenetic alopecia and antihypertensive drugs.
A pre-printed examination form with a scheme of a simplified version of the Norwood’s classification of androgenetic alopecia in 5 grades (0-4) was used to collect standardized information [8].
Anonymously collected data included age, 5 grades of androgenetic alopecia (0-4), hypertension, history of familial androgenetic alopecia, history of personal diabetes mellitus, smoking, history of personal hyperlipidemia, antihypertensive drugs (name). Excluding criteria included history of cancer, some treatments (corticosteroids, interferon, antimitotic, retinoid, lithium, finasteride, testosterone, danazol).
Hypertension was defined by a systolic and diastolic blood pressure over 140 and 90 mmHg, respectively. Patients with normal blood pressure but treated for hypertension were considered as hypertensive. Hyperlipidemia was defined by history of abnormal values (cholesterol or triglycerides).
Three groups of patients were defined 35-45 years, 46-55 years, and 56-65 years.
Patients’ data were recorded in a computerized database. Statistical analyses were conducted using SAS v8.2 (SAS Institute Inc, Cary, NC). The baseline characteristics of the study patients were expressed as numbers and percentages for categorical variables (hypertension, hyperlipidemia, history of familial hypertension, history of familial androgenetic alopecia, diabetes mellitus, smoking, treatment) and as means ± standard deviations (SD) for continuous variables. For univariate analysis, the Chi-square and Fisher exact test were used for categorical variables and a Student test was used for the comparison of the age according to the presence or not of alopecia. The variables with the threshold of 0.1 in these univariate analysis were selected, like explanatory variables of the alopecia in a logistic model of regression step by step. Only the results with the threshold of 0.05 were regarded as significant with the logistic model of regression.
Posted by: SANKOPosts: 3 :: 15-02-2010 :: | Reply to this Message
Re: use of odds ratio to establish association in cross sectional studyok, so they dichotomized their outcome into presence or absence of alopecia and then used logistic regression; pretty standard method - and exponentiation of the beta-coefficients from the regression equation gives odds ratios. Is there still a question?
Posted by: rqayyumPosts: 199 :: 17-02-2010 :: | Reply to this Message
Re: Re: Re: use of odds ratio to establish association in cross sectional study
Hey Sanko buddy,
Sorry I haven't been able to read your detailed email and answer yet. I'll try going over it this weekend. I hope Dr.Rehan Qayyum's answer helped you.
Remember Statistical correction can only account for some of the problems of confounding. You have to take into consideration the limits of the study design based on your research question before you carry out the study.
For example, a cross sectional study gives a snapshot of a disease or exposure in a certain time. So for certain diseases with a fast recovery and low incidence, it will probably miss many of the cases. Also, the location of where you carry out the study ( the hospital e.g can miss the more severe cases that died on the way or the latent cases that caused minimal symptoms) can cause a bias.
So, it basically boils down to how much bias can you afford in the study. A cross sectional study is best if very little is known about the disease or you want to give something to look at for policy makers.
If information is plentiful and resources are available, a more powerful study like a case-control, a cohort or ideally a randomized control trial is appropriate.
So, it would help us more if you would tell us what is your research question and what is your hypothesis. You could submit your synopsis and we can guide you as to what we would think is a more better design.
Posted by: aftab.iqbaPosts: 7 :: 18-02-2010 :: | Reply to this Message
Re: use of odds ratio to establish association in cross sectional studyNot a comment, but a question: Is there a specific formula to convert OR into PR? rather than ac/bd to a/(a+b) : c / (c+d)? Just in case there is not specified the number of a,b,c,d? Thank you
Posted by: Grossville (Guest) :: 20-10-2011 :: | Reply to this Message
Re: Re: use of odds ratio to establish association in cross sectional studyYes you can use this formula for the calculation of RR. (ac+ad)/(ac+bc). for any assistance call 0300 4668681
Posted by: ibrahim_apPosts: 138 :: 20-10-2011 :: | Reply to this Message
Re: Re: use of odds ratio to establish association in cross sectional study
The answer is yes. There are statistical methods (based on heavy mathematics) to estimate the missing freuencies i.e. a, b, c, d or the marginal frequencies (column and row toatals) ie. a+b, a+c ...
Risk ratio (RR) and odds ratio (OR) are two measures for comparing risks; the former relates the chance or risk of developing outcome for exposed versus unexposed groups, the latter relates their odds. Because odds are not ordinarly used in common language as chance or probability or risk, many people find the concept of OR a bit more difficult to interpret than that of RR.
A word of caution:
You have written OR=ac/bd, which is not the form we statisticians use. Let me explain;
Our sample data in a prospective study looks like as
Exposed Unexposed Total
Cases a b a+b
Noncase c d c+d
Total a+c b+d n
for which OR=ad/bc and not what you have written. To find RR we can use the fact
RR=[(a/a+c)/(b/(b+d)]=(ab+ad)/(ab+ac)
For details you may consult my book
Sahai, H. and Khurshid, A.(1995). Statistics in Epidemiology: Methods, Techniques and Applications. CRC Press, New York.
Posted by: anwer_khurPosts: 30 :: 20-10-2011 :: | Reply to this Message
Re: Re: Re: use of odds ratio to establish association in cross sectional study
Dear Anwar Sb, Can i got your email address
Muhammad Ibrahim
ibrahim_ap98@yahoo.com
Posted by: ibrahim_apPosts: 138 :: 21-10-2011 :: | Reply to this Message
Re: use of odds ratio to establish association in cross sectional studyanwer_khurshid@yahoo.com
Posted by: anwer_khurPosts: 30 :: 21-10-2011 :: | Reply to this Message