Hello,

since you want to find out risk factors for mortality in my opinion there is not much reasoning in doing a matching of survivors and non-survivors after all. You must be aware, that if you match for certain variables, you can not identify these variables as risk factors in subsequent analyses.

There is no reason to do a matching to accomplish your research question. In fact you would harm the generalizability of your findings, since they will have no meaning for dengue patients in total anymore but will have only meaning for an artificially matched sample with no adequate population that it could represent.

The major problem with your data is the very low number of deceased patients. Since logistic regression would be the method of choice to identify risk factors. The possiblity of logistic regression to handle a set of predictors is restricted by the number of patients in the smaller group.

If you have any possiblity to obtain further data, you should try this. Perhaps achieve at least 30 cases before doing the analyses.

Greetings

Daniel

There is an inherent bias in the  case- control studies.  The matching and the proper selection  could reduce this

Hello Tauqueer,

The limitation of your study is the fact that 8 cases are associated with your study being underpowered in terms of subsequent analyses. Also, if the paper did not provide thorough methodology on how it could be done it is serious limitation of this research.

Best wishes

Your sample cases size is extremely low and at the end you will not get any significant data results, as a rule of thumb your cases sample should be at least 30, in this case your population n = 667 and your sample cases = 8 which equals 1.19% of your population and this will not be statistical significant at all, if you only have 8 cases I would suggest change the epidemiologic model study, probably a case series would be easier, I hope my comment helps. 

Dear Tauqeer Hussain Mallhi

Generally  in  large sample study  no  need to  well  matched the controls. With  more deep  statistics you can  avoid certain  bias.  Moreover you have new findings and new predictors 

@ pablo and witold

Thank you so much for your answers, have you seen one paper that i have attached with question. Authors did same study with same methodology. So according to your suggestion I can not perform this study. Actually My objective is to determine risk factors of mortality.

@Ishaq

I agreed with you. But How can we match cases with controls?

Dear  Tauqeer Hussain Mallhi 

it is not necessary to  be fully  matched e,g age, gender can  be matched easily. Blood groups and others are difficult to  be matched   

So it means we can match manually?

My dear, 

There could be several reasons for those deaths, be the extraneous variables; it seems illogical to do this study for that purpose. Doing a case-control study doesn't seem useful in your case.

Munir

I agree with Mr. Muneer ahmad 

Dear  Tauqeer Hussain Mallhi

It  is possible to  compare the small number ( 8) of the deaths with  the rest or you can 

randomly  select 16  or 24 from  the other using computer generated numbers 

Hello,

since you want to find out risk factors for mortality in my opinion there is not much reasoning in doing a matching of survivors and non-survivors after all. You must be aware, that if you match for certain variables, you can not identify these variables as risk factors in subsequent analyses.

There is no reason to do a matching to accomplish your research question. In fact you would harm the generalizability of your findings, since they will have no meaning for dengue patients in total anymore but will have only meaning for an artificially matched sample with no adequate population that it could represent.

The major problem with your data is the very low number of deceased patients. Since logistic regression would be the method of choice to identify risk factors. The possiblity of logistic regression to handle a set of predictors is restricted by the number of patients in the smaller group.

If you have any possiblity to obtain further data, you should try this. Perhaps achieve at least 30 cases before doing the analyses.

Greetings

Daniel

I agree with you Daniel Schwarzkopf  that  no  need for matching 

@Daniel

Thank you so much for your guidance, So how I select cases. Either I select blindly or what? Kindly guide me here. Because then someone can ask that you biased your result by selecting cases by yourself?

Dear  auqeer Hussain Mallhi ·

You  can  select them  through  this link

http://www.openepi.com/Menu/OE_Menu.htm

Dear Tauqeer, 

I agree with Daniel that matching is not relevant to your study on finding predictive factors. The issue here, in my point of view, is to define independent variables properly such as geographic region, urbanization, monsoon period, sunlight issue, personal hygiene, prior exposure, members in the family,.....I just guess. And then you may use logistic regression for the binary death or alive and publish your paper. 

with best regards,

Victor

cases and control are matched by the investigator and the ratio depends on available of resources and subjects. you can use 1 case for one control when resources are unlimited. to improve on your result and pending resources one case can be matched to more than 1 control...however it the number of controls to match with 1 case should not be more than 4 if not it becomes useless. cases and controls can be matched on every categories studied but it is also not good to over matched cases and control because it will analysis very difficult. hope this proves useful. also take note dead 'cases' don't represent any risk group to your study since they are already dead...and using alive subjects to represent 'dead case' is not public health acceptable since dead people are no longer at risk for the disease.....

I will regard your question in an old fashion way. Although impressive breakthroughs have been achieved using a few cases matched case control design  (Herbst AL, Ulfelder H, Poskanzer DC, “Adenocarcinoma of the vagina: association of maternal stilbestrol therapy with tumor appearance in young women”.  NEJM 1971;284:878-81), general Epidemiology expects a hypothesis - as specific as possible- so as to avoid missing the factor of interest due to overmatching, as is mentioned by Daniel. In such scenario (matched case-control) controls have to be selected “randomly”, (for example, the next in space and time of each case, and not just those pointed out by a randomly made number series picked from a large number of possible controls). In cases when the cases are already available, a “nested” case-control” design looks better: controls, much more available than cases, are being selected whenever a new individual meeting the control definition (for example, the variables to be matched for) appears. Matched case-control designs require matched analysis, not the regular analysis applied to a two group series. This is not evident in the material provided as the example. I really cannot say whether logistic regression may solve for these requirements.

I would regard your question from a clinical point of view: how can we prevent more deaths among those with the living diseased? -  a case-control study is highly relevant (with the dead as the cases and the living controls) but due to the small number of case and the risk of type-1 statistical error I think there are things to do for a start - in order to give the best advice to prevent more death:  

1) Search PubMed with the objective to learn about the comorbidities and age-differences, co-infections, diabetes and other comorbidities and follow-up studies - could be a review article

2) do a descriptive study of your own data of the co-morbidities and age-groups for the dead and the living - could be the next publication 

3) contact other researchers with similar data and propose a case-control study with a larger case sample with the pooled data - then it would be possible to do the logistic regression reasonably - the third publication :) 

What is your research question and why do you want to do a case control study in that particular cohort; be precise about the objectives of your study. Tell us first those and then let us see if that is possible for your cohort.

Munir

Thanks for your reply.

I need to compare fatal and non-fatal cases in order to identify risk factors of mortality.

Perhaps a couple of definitions may help Tauqeer:

The "question" deals with a proposition posed in an interrogative way. It bears an implicit answer (what the authors have in mind) , which is a proposition or statement.  Hypothesis is such proposition -that can true or false. Moreover, the question (the hypothesis) may be theoretical (the conceptual hypothesis), as is the "answer" the author has in mind. An "operational hypothesis" is beyond this: it consists on the statement formulation but with concrete or explicit, metrical traits. The latter gives way to assuming differences, methodology both dealing with the design as well as the statistical methods, expected associations, sample size calculations, etc.

Dear Patricio Herrera

Thanks your comments are valuable for all  of us