Topic Review - Newest First (only newest 5 are displayed)
doctoc
Re: How to control / eliminate bias and confounding in my synopsis
@rqayyum. Thank u for ur help regarding the books, id try to study it myself.
@chameed.
1. Stress hyperglycemia wont produce any effect on HbA1c bcoz it is fairly short lived.
2. Whenever there is hyperglycemia per se it elevates the production of ketone bodies, and consequently fatty acids within the blood as the body d try to find alternate source of energy.
3. Stress hyperglycemia is not related with BMI or HTN
Its only importance is that the pts who r admitted in surgical icus, bcoz their outcomes depend on it.
GLUCOTROL study i think is based on stress hyperglycemia but the results r not published .
all new information d b welcome.
chameed
Re: How to control / eliminate bias and confounding in my synopsis
Hyperglycemia will eventually affect HbA1c but is there is a pre-determined level and/or duration after which HbA1c will be affected?
Does hyperglycemia affect lipid levels or its the other way around? Will that happen in stress hyperglycemia?
Is level of stress hyperglycemia related to the BMI?
People who have tendency towards hypertension, will their level of hyperglycemia be different than those of nomotensives?
Any one?
rqayyum
Re: Re: How to control / eliminate bias and confounding in my synopsis
If you allow me I will recommend not one but four books.
1. "Designing Clinical Research: An Epidemiologic Approach" By Stephen Hulley and others. ISBN-10: 0781782104
This is a very nicely written book for beginners to help in designing clinical studies. I will highly recommend it.
2. "Primer of Biostatistics" by Stanton Glantz. ISBN-10: 0071447814.
This another nicely written book on Biostatistics for beginners. It discusses basic issues such as t-test, chi-square test, basic non-parametric tests, basic sample size calculation, and simple linear regression.
3. "Primer of Applied Regression & Analysis of Variance" by Stanton Glantz and Bryan Slinker. ISBN-10: 0071360867.
Once you know some basics about biostatistics and want to learn multivariate linear regression, this is a very good book written in the same style as the above primer.
4. "Logistic Regression" by David Kleinbaum and others. ISBN-10: 0387953973. A very nice book for learning logistic regression. It is especially helpful if you already have some understanding of multivariate linear regression.
These books should be more than enough to guide you through most of your needs, unless you want to do survival analysis, genetic analysis, or design studies with complex designs.
[Edited by rqayyum on 17-12-2007 at 01:18 AM GMT]
doctoc
Re: How to control / eliminate bias and confounding in my synopsis
Thank u for answering my query docosama and rqayyum. Getting the help of biostatician is impossible bcoz i work at Hyderabad and there is not a single biostatician present in this city. Please suggest a stats book which i can study myself, otherwise i ll v to scrap this synopsis and write a new one altogether.
rqayyum
Re: How to control / eliminate bias and confounding in my synopsis
First, these variables on which you plan to collect data can all be construed as outcome variables (or dependent variables). Hyperglycemia is likely to affect all these variables. What predictor (or independent) variables do you plan to collect data on?
Confounding is due to a variable that is related to both outcome variable and predictor variable and produces a false association between outcome and predictor variables. For example, if age is associated with both hyperglycemia and mortality, and if you don't control for age in your analysis, you may find a false association between hyperglycemia and mortality. Thus age will become a confounding factor. Severity of illness and other co-morbities can also affect relationship between your outcome variables and predictor variables.
Bias are of various types (some have enumerated 56 types of bias) but you are likely to encounter selection bias; selection of study subjects in such a way that can give incorrect results.
Statistical tests depend on your specific hypotheses. If you plan to control for confounding variables, you may want to consider linear regression (for continous variables) or logistic regression (for dichotomous variables). You can also use propensity scores as an alternative method. I would suggest to keep two groups instead of three and use presence or abscence of diabetes as a dummy variable in your regression model. Comparing three groups at the same time while controlling for confounding variable will be much more difficult and complicated.
You can eliminate bias only after enlisting all possible bias that can arise in your study design and then making every effort to eliminate them. For example you can eliminate selection bias by selecting your sample of patients randomly from the source population.
You can't eliminate confounding variables, you will have to collect data on all possible confounders and adjust your statistical methods accordingly.
Either you need to know (or learn) relevant biostatistics or you will need a biostatistician. It is as important when not to use a statistic as it is to know when to use one.
