I would not rely on correlation measurements (Pearson, Sperman) or logistic regression to infer association between any dependent variable with mortality as outcome. "Fatality rate" is clearly a time-dependent variable, and association of dependent variables (any scale - nominal, ordinal or continuous) and a time-dependent outcome should be analised by survival methods - and Kaplan Meier (univariate) and Cox regression (multivariate) may be more informative here. 

Jia - Yes, but for prediction, or for forecasting, if possible, you want to know what kind of regression (linear, nonlinear, ?) you might be able to use.  Have you looked at any scatterplots, or whatever graphics are appropriate for your data?  They can be very enlightening.  - Cheers. - Jim      

A good suggestion by James. 

Jia- correlations are elementary, consider some survival analysis or regression models. A rough example is: 

Mortality: Yes or No (1 or 0)

Syndromic data: Categories 

Covariates: study population characteristics

 use a logistic regression to predict mortality.

thanks guys...I think I will consider some simple logistic regression here with syndromic data (measurement variable) and mortality (nominal variable).

Just to add my two cents worth to the comments above.  Since you have measurement variables, you might want to modify the syndromic data to see if the relationship is non linear - so you could still use the correlation coefficient to see how much of the modified independent variable explains the variation of the dependent variable (e.g. y=f(x^2)).  Also, I realize you did not mention having a time variable, but since different environmental factors come into play as an epidemic unfolds (more information, earlier detection, better care), if you have different datasets of the same variable that are coming in piecemeal, you might not want to lump them all together, since there may be new factors that are affecting mortality.  Hope this make sense.

thanks Thomas. you right...i did correlation coefficient and CoVar...the syndromic data and mortality have weak linear correlation though

I have a slightly different take on your question.

The kind of correlations you want to evaluate would be used more in initial exploration of your data, to see if certain predictor variables are associated with a higher or lower case fatality rate.

However, this will only be the initial exploration.

Afterwards you will have to decide what your research question is. Do you want to see the effect of certain variables on time to death? ( in which case you will not be dealing with case fatality rates and would need survival analysis). Do you want to see population averages in case fatality rates over time in 2 or more different groups, for which you will need GEE or mixed effects models, and will need to look at the covariate matrix structure. Or do you want to see effect of certain variables on case fatality rates measured at one time period or point only, in which case you can use logistic regression.

Hope this helps. As I an new to this, other researchers must please feel free to correct any errors in this post.

thanks Llewellyn... I am looking at the possible link between syndromic data being collected and number of death....

They can be measures of association, if a relationship between the variables can not be explained by the dependence of these variables on some additional variable. The second - it exists the relationship between them. The third – it is not a feedback of mortality to syndromic data.

Unless there is some governing theory behind it, your correlation coefficient will remain merely a number. Please make sure you authenticate the relationship based on theory not just on statistical fit! Thus, if theory says yes that variable X influences Y and you alpha amount of correlation between the two, then are comfortable to stand for it.

I agree with you. Before starting the research it needs to develop some hypotheses and distinguish variables for a determination a cause-effect relationship. The  positive correlation between variables can be an argument of existence a cause-effect relationship.

I would not rely on correlation measurements (Pearson, Sperman) or logistic regression to infer association between any dependent variable with mortality as outcome. "Fatality rate" is clearly a time-dependent variable, and association of dependent variables (any scale - nominal, ordinal or continuous) and a time-dependent outcome should be analised by survival methods - and Kaplan Meier (univariate) and Cox regression (multivariate) may be more informative here. 

thanks Sandro...I think you are correct...because CFR looks at the probability of dying from a particular disease from a cohort of infected. its worth looking at. how about using logistic regression to determine the association between syndromic data and mortality-number of death ?

If you pick a time-dependent variable (mortality) and transform in dichothomic (Yes/No, nominal scale) in order to perform a logistic regression, for instance, you lose statistical power (you may miss an association otherwise present in your dataset). Instead, you could consider to employ "time on-study" as outcome measure, censoring live subjects, and include relevant dependent variables ("syndromic data", any scale, and demographic data) in a Cox regression model.  

Jia in order to contribute to the discussion, let us know, before, what kind of scale you are using to measure your response variable, in this case, the fatality rate.

Because if you have measured it as "yes/no", one of the best option you have in order to perform the statistical analysis is the Logistic regression, as long as you are interesting in controlling for several other potential confounder.

But if you have measured it as continuous one, I mean as a time elapsed between to different moments, in this case I agree, the best option is Cox proportional model.

Any time one transforms a continuous variable to dichotomy, what one loose is information, not power, power is a study characteristic related to the sample size.

So, let us know what scale you used to measured the fatality rate.