We are want to conduct a study to identify the risk factors for developing ventilator associated pneumonia in our intensive care unit. After identifying the control group, which test is suitable to study the risk factors? On which bases the selection of test should be made?
Hi,
A consideration would be to utilize logistic regression modalities( simple or multivariate). After creating a dichotomous outcome (PNA or not), independent variables (risk factors-ventilated, diabetic, age, etc...) can be analyzed to determine their relevance to the outcome in question. As you proceed, novel dependent variables can be created to address new considerations (PNA AND a MDRO as causative organism... for example) while testing the independent variables for relevance.
In some instances-especially as you create new dichotomous dependent variables, mitigation of sparse data bias may need to be performed (Fisher's exact, etc), but overall this is a reasonable approach (albeit there are multiple) to determine the relevance of risk factors.
Hope this helps
it depends upon ur study design. if u r conducting a case control study, go for ODDS' ratio. if u r proposing for cohort design, go for relative risk, incidence rates etc. these u can calculate using EPI INFO or SPSS.
Personally, I prefer to assess risk through probit or logit regression. Using a general MANOVA or ANOVA. causality seeks more experimental. and what I feel and I "guess", the question that should be behind this, is what are the chances that a subject with x1, x2, x3, x4 .... can acquire certain disease characteristics or conditions?.
MANOVA is used for multiple dependent variables, just i see one. but many independent.
Hi,
A consideration would be to utilize logistic regression modalities( simple or multivariate). After creating a dichotomous outcome (PNA or not), independent variables (risk factors-ventilated, diabetic, age, etc...) can be analyzed to determine their relevance to the outcome in question. As you proceed, novel dependent variables can be created to address new considerations (PNA AND a MDRO as causative organism... for example) while testing the independent variables for relevance.
In some instances-especially as you create new dichotomous dependent variables, mitigation of sparse data bias may need to be performed (Fisher's exact, etc), but overall this is a reasonable approach (albeit there are multiple) to determine the relevance of risk factors.
Hope this helps
I think the regression analysis is one of the common statistical test
Since your outcome variable - developing pneumonia or not is a binary variable you will estimate odds ratios using logistic regression. Make sure you compare adjusted and unadjusted ORs. Choose your reference group carefully to estimate the odds in logistic regression. If you suspect if there is a gender difference you might want to conduct stratified analyses.
It depends on exposure & outcome in your study. If your case groups are patients with pneumonia (independent variable or exposure) and outcome (dependent variable) is ventilator, and at beginning of the study, any of patients in case & control groups have no outcome (ventilator), you follow them and whether or not they need to ventilator during follow up, so your study is cohort and you should calculate relative risk.
Careful logistic regression not only regression which measures more the degree of interaction between variables, Odds Ratio and always calculate confidence intervals as well
You can use univariable and multivarible cox proportional hazards regression analysis.
Dear Behnaz, what is the advantage of multivarible cox proportional hazards regression analysis on logestic regression modalities, which was mentioned by S. Jason Moore?
It very much depends on your study but overall, regression analyses are the proper tests
This has been fairly well discussed. Risk factor analysis depends mainly on strength of association i.e. how strong is the risk in explaining the outcome. Logistic regression analysis stands out. Most of the other statistical methods that can be used are not as explanatory as logistic regression. the simplified method for health surveillance basically is 2x2 contingency table which provides odd ratio , relative risk and confidence intervals.
We can use odd ratio, relative risk, chi squire test for description of risk factors causing disease
Logistic regression is the best option to explore the importance of each risk factors.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software