Skip to content

[R-meta] Meta-Analisys whit events numbers, RR, OR and HR. It's possibe?

9 messages · Gerta Ruecker, Michael Dewey, Martin Lobo +1 more

Martin Lobo
# 
Hi everyone !!!
I need performed a meta-analisys, but i have a problem whith de data extractions.
In 3 papers I have:
event.e, n.e, event.c, n.c (or  ai=tpos, bi=tneg, ci=cpos, di=cneg)

In another papers I have:

n.e, n.c and HR
In another papers I have:

n.e, n.c and OR
In another papers I have:

n.e, n.c and RR

I have read in the cochraine manual that it could be meta-analyzed, but I don't know how.

Should I transform everything to the same summary measure?

How would these measures be transformed?

How would the meta-analysis with the transformed measurements.

Sorry so many questions but I need your help.

Regards






Lorenzo Mart?n Lobo MTSAC, FACC, FESC
Especialista Jerarquizado en Cardiolog?a
Jefe de Dpto Enf. Cardiovasculares y Cardiometabolismo Hospital Militar Campo de Mayo.
Jefe de Cardiolog?a Hospital Militar Campo de Mayo
Ex Jefe de Unidad Coronaria Hospital Militar Campo de Mayo
Miembro Titular de la Sociedad Argentina de Cardiolog?a
Fellow American College of Cardiology
Fellow European Society of Cardiology
Ex Miembro del Area de Investigaci?n de la SAC
Ex Director del Consejo de Aterosclerosis y Trombosis de la SAC
Miembro Asesor del Consejo de Aterosclerosis y Trombosis de la SAC
Ex Director del Consejo de Epidemiolog?a y Prevenci?n Cardiovascular de la SAC

Miembro Asesor del Consejo de Epidemiolog?a y Prevenci?n Cardiovascular de la SAC


Experto en Lipidos de la Sociedad Argentina de Lipidos.
Miembro de la Sociedad Argentina de Lipidos.
Instructor de ACLS de la American Heart Association
Gerta Ruecker
# 
Dear Martin,

My first question is: What kind of outcome do you have? HR is for a 
time-to-event outcome, for example survival, the others are for a binary 
outcome, for example death within a certain (fixed) time interval. They 
don't describe the same thing.

For extracting data for a time-to-event outcome, there are a number of 
papers, for example

Tierney et al. (providing an excel sheet): 
https://trialsjournal.biomedcentral.com/articles/10.1186/1745-6215-8-16

Parmar et al.: 
https://doi.org/10.1002/(SICI)1097-0258(19981230)17:24<2815::AID-SIM110>3.0.CO;2-8

Best,
Gerta

Am 23.01.2021 um 13:23 schrieb Martin Lobo:

  
  


  


      

    

    

      
      

        


  

        

      Martin Lobo
    

    

      
      #
    

  


  

      
Hi Gerta !!!
yes gerta, it's like you said.
I have seen that with metagen it could be done but I think it would open to pass everything to RR, OR or HR to use the Ln of these, and with the p, or IC95% I could get the standard errors.
But I don't quite understand what it would be like.
Thanks







Lorenzo Mart?n Lobo MTSAC, FACC, FESC
Especialista Jerarquizado en Cardiolog?a
Jefe de Dpto Enf. Cardiovasculares y Cardiometabolismo Hospital Militar Campo de Mayo.
Jefe de Cardiolog?a Hospital Militar Campo de Mayo
Ex Jefe de Unidad Coronaria Hospital Militar Campo de Mayo
Miembro Titular de la Sociedad Argentina de Cardiolog?a
Fellow American College of Cardiology
Fellow European Society of Cardiology
Ex Miembro del Area de Investigaci?n de la SAC
Ex Director del Consejo de Aterosclerosis y Trombosis de la SAC
Miembro Asesor del Consejo de Aterosclerosis y Trombosis de la SAC
Ex Director del Consejo de Epidemiolog?a y Prevenci?n Cardiovascular de la SAC

Miembro Asesor del Consejo de Epidemiolog?a y Prevenci?n Cardiovascular de la SAC


Experto en Lipidos de la Sociedad Argentina de Lipidos.
Miembro de la Sociedad Argentina de Lipidos.
Instructor de ACLS de la American Heart Association

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Gerta Ruecker
    

    

      
      #
    

  


  

      
Dear Martin,


I am not sure whether I understand your last post. So you have a 
time-to-event outcome, right? Then the HR (hazard ratio) would be the 
preferred effect measure. However, the point is that there is no 
easy-to-use formula to simply calculate RRs or ORs as hazard ratios, 
because they describe different aspects. The preferred way to handle 
this is to obtain as many as possible information from the primary 
studies and to use methods as described in the papers I have recommended 
to you.

The papers might present 1-year (or 2-year or 5-year) survival rates, or 
numbers of observed vs expected events, or full Kaplan-Meier curves or 
tables. Sometimes it is possible to obtain almost IPD information from 
figures, thus it could be valuable to take these into account.


Best,

Gerta



Am 23.01.2021 um 19:14 schrieb Martin Lobo:

            

      
      
      
    

        

      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Michael Dewey
    

    

      
      #
    

  


  

      
Dear Martin

As Gerta says HR is different but if you want to try to retrieve the raw 
frequencies from OR and RR there is a package on CRAN which claims to do 
that

https://cran.r-project.org/package=estimraw

There is a reference to a supporting article as well you you can check 
what they are doing.

Michael

        
On 23/01/2021 18:14, Martin Lobo wrote:

            

      
      
      
    

        

      
      
    

        

  
    
  


  


      

    

    
5 days later

    

      
      

        


  

        

      Martin Lobo
    

    

      
      #
    

  


  

      
Dear Gerta and Michael,
Thank you very much for your answers.
I thought it was easier than what is shown in the papers, but maybe I did not explain it well.
I have six studies that report HR and its 95% CI. 2 studies that report number of events in the two branches. 3 studies that report OR and its 95% CI.

In the Cochrane manual, it says that these data can be meta-analyzed with the generic inverse of variance method.
but I don't understand if these measures can really be meta-analyzed together or not.
I will exemplify with less studies what you can find to see what you think

Suppose I have 4 studies that report:

HR (1.05, 0.44, 0.31, 0.77)
CI95% L (0.65, 0.16, 0.09, 0.68)
CI95% U (1.69, 1.22, 1.01, 0.87)
In the META package with the metagen function, this could be done according to the following example:
Conduct meta-analysis using hazard ratios and 95% confidence intervals
#
# Data from Steurer et al. (2006), Analysis 1.1 Overall survival
# https://www.cochranelibrary.com/cdsr/doi/10.1002/14651858.CD004270.pub2/abstract
#
study <- c("FCG on CLL 1996", "Leporrier 2001", "Rai 2000", "Robak 2000")
HR <- c(1.05, 0.44, 0.31, 0.77)
lower.HR <- c(0.65, 0.16, 0.09, 0.68)
upper.HR <- (1.69, 1.22, 1.01, 0.87)
#
# Input must be log hazard ratios, not hazard ratios
#
metagen(log(HR), lower = log(lower.HR), upper = log(upper.HR),
        studlab = study, sm = "HR")

Now, I could add the data of the papers that report OR for example:

OR (0,47 , 0.77, 072)
IC95% L (0.24, 0.52, 0.55)
IC95% U (0.91, 0.81, 0.89)

HR <- c(log(1.05),log(0.44), log(0.31), log(0.77), 0.47,0.77,0.72)
lower.HR <- c(log(0.65), log(0.16), log(0.09), log(0.68),0.24, 0.52, 0.55)
upper.HR <- (log(1.69), log(1.22), log(1.01),log( 0.87),0.91, 0.81, 0.89)


metagen(log(HR), lower = log(lower.HR), upper = log(upper.HR),
        studlab = study, sm = "HR")

Is this correct or should I analyze the HR on the one hand and the OR on the other?

By last. There is a study that reports HR backwards. In other words, not the HR for those who were taking the drug but for those who were not taking it. Is there a way to report it backwards so that you can meta-analyze it with the other HR.

For example, all studies report HR of statin patients. HR 0.78 patients taking statins had 22% fewer events than those not taking it.

One study reports the HR of patients not taking statins. HR2.17 (95% CI 1.04-4.54). In other words, those who did not take statins had twice the number of events.

Best regards














Lorenzo Mart?n Lobo MTSAC, FACC, FESC
Especialista Jerarquizado en Cardiolog?a
Jefe de Dpto Enf. Cardiovasculares y Cardiometabolismo Hospital Militar Campo de Mayo.
Jefe de Cardiolog?a Hospital Militar Campo de Mayo
Ex Jefe de Unidad Coronaria Hospital Militar Campo de Mayo
Miembro Titular de la Sociedad Argentina de Cardiolog?a
Fellow American College of Cardiology
Fellow European Society of Cardiology
Ex Miembro del Area de Investigaci?n de la SAC
Ex Director del Consejo de Aterosclerosis y Trombosis de la SAC
Miembro Asesor del Consejo de Aterosclerosis y Trombosis de la SAC
Ex Director del Consejo de Epidemiolog?a y Prevenci?n Cardiovascular de la SAC

Miembro Asesor del Consejo de Epidemiolog?a y Prevenci?n Cardiovascular de la SAC


Experto en Lipidos de la Sociedad Argentina de Lipidos.
Miembro de la Sociedad Argentina de Lipidos.
Instructor de ACLS de la American Heart Association

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Guido Schwarzer
    

    

      
      #
    

  


  

      
Am 29.01.21 um 16:29 schrieb Martin Lobo:

            

      
      
      
    

        
Dear Martin,

Tierney et al. (2007) describe why using event numbers or odds ratios / 
risk ratios for time-to-event data can be problematic. Accordingly, I 
would recommend to base the main analysis on studies providing hazard 
ratios and to consider the meta-analysis of hazard and odds ratios as a 
sensitivity analysis (with subgroups defined by effect measure).

            

      
      
      
    

        
First, you are taking double logs for hazard ratios. Second, as 
mentioned above, I would conduct a subgroup analysis:

HR?????? <- c(1.05, 0.44, 0.31, 0.77, 0.47, 0.77, 0.72)
lower.HR <- c(0.65, 0.16, 0.09, 0.68, 0.24, 0.52, 0.55)
upper.HR <- c(1.69, 1.22, 1.01, 0.87, 0.91, 0.81, 0.89)
##
effect.measure <- rep(c("HR", "OR"), c(4, 3))
##
metagen(log(HR), lower = log(lower.HR), upper = log(upper.HR),
 ??????? sm = "HR", byvar = effect.measure)

            

      
      
      
    

        
If the hazard ratio for group B vs A is 'x' than the hazard ratio for A 
vs B is '1 / x'. Accordingly, you can calculate the hazard ratio for 
statins vs no statins as 1 / 2.17. The lower confidence limit is 1 / 
4.54 and the upper limit is 1 / 1.04.

Best wishes,
Guido

Reference:

Tierney JF, Stewart LA, Ghersi D, Burdett S, Sydes MR. Practical methods 
for incorporating summary time-to-event data into meta-analysis. Trials. 
2007;8:16.

  
  


  


      

    

    

      
      

        


  

        

      Gerta Ruecker
    

    

      
      #
    

  


  

      
Dear Martin,

To add to Guido's answer, it seems that there is something wrong with 
your odds ratios. You see this when you use Guido's code: For the last 
two studies, the confidence intervals differ markedly from the entered 
values:

Entered:

            

      
      
      
    

        
Result:

[...]
study 5?? 0.47 [0.24; 0.92]?? OR
study 6?? 0.77 [0.62; 0.96] ? OR
study 7?? 0.72 [0.57; 0.92] ? OR

You also see this when looking whether the point estimate is the 
geometric mean (square root of the product) of upper and lower limit of 
the 95% CI - this is not the case for these studies. Please check.

Moreover, as I wrote before, the main problem is not a technical one 
(how to use R to mix these data), but a methodological issue. HR and OR 
don't describe the same thing. The heading is "Hazard ratio", because 
this is what you have written. Nevertheless, the odds ratios remain odds 
ratios and have a different interpretation. I don't know how to 
interpret the final result.

Both Guido and I recommended the Tierney paper to you, so please have a 
look into it.

Best,

Gerta

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Martin Lobo
    

    

      
      #
    

  


  

      
Dear Gerta and Guide.
You are simply two geniuses.
Like you, it didn't sound very logical to me to mix these effects, and the Cochrane manual confused me.
Gerta does not call my attention the error you found in the ORs, since not having the studies at hand I invented them.

thank you very much to both

Regard





Lorenzo Mart?n Lobo MTSAC, FACC, FESC
Especialista Jerarquizado en Cardiolog?a
Jefe de Dpto Enf. Cardiovasculares y Cardiometabolismo Hospital Militar Campo de Mayo.
Jefe de Cardiolog?a Hospital Militar Campo de Mayo
Ex Jefe de Unidad Coronaria Hospital Militar Campo de Mayo
Miembro Titular de la Sociedad Argentina de Cardiolog?a
Fellow American College of Cardiology
Fellow European Society of Cardiology
Ex Miembro del Area de Investigaci?n de la SAC
Ex Director del Consejo de Aterosclerosis y Trombosis de la SAC
Miembro Asesor del Consejo de Aterosclerosis y Trombosis de la SAC
Ex Director del Consejo de Epidemiolog?a y Prevenci?n Cardiovascular de la SAC
Miembro Asesor del Consejo de Epidemiolog?a y Prevenci?n Cardiovascular de la SAC
Experto en Lipidos de la Sociedad Argentina de Lipidos.
Miembro de la Sociedad Argentina de Lipidos.
Instructor de ACLS de la American Heart Association