5
$\begingroup$

I am doing a meta analysis for the first time and have a few basic questions regarding statistical analysis. Let's say I have one study where the primary outcome (thrombosis) in the 2 treatment groups (intervention v. placebo) was compared using the Mantel-Haenszel chi-square test. It does not report df in the article (is this something I can deduce?). Then it tells me that "Treatment effects were expressed as a weighted average of the strata-specific relative risks." Is this something I need to consider for the meta-analysis statistical calculation? If so, what does it mean? In the results we learn "Among the 866 participants who had patency assessed, the primary outcome of fistula thrombosis at 6 weeks occurred in 53 participants (12.2%) in the clopidogrel group compared with 84 participants (19.5%) in the placebo group (relative risk, 0.63; 95% CI, 0.46-0.97;P=.018).

What do I do with all this information? I tried using a computerized program, but I didn't know what to fill all the blanks in with...

In the second study, we have the following: "The hazard ratio was 0.81 (95% CI, 0.47 to 1.40) in favor of aspirin and clopidogrel therapy, but the reduction was not significant (P= 0.45). Although the event rates in the two treatment arms converged, there was no evidence that the proportional-hazards assumption was violated (P 0.53). The annual hazard rate for thrombosis among articipants receiving placebos was 0.59 (95% CI, 0.39 to 0.87), compared with 0.47 (95% CI, 0.31 to 0.71) among participants receiving aspirin and clopidogrel. The absolute risk reduction for the group receiving aspirin and clopidogrel was 0.03 (95% CI,-0.19 to 0.07), and the number needed to treat was 33.3 (95% CI, 14.9 to >/=1000)." Regarding analysis, it states: "The cumulative incidence of the first episode of thrombosis was estimated with the Kaplan-Meier method, and differences in rates between treatment groups were tested with the log rank test. The cumulative incidence of the first bleeding event was analyzed in a similar manner."

What do I do with all this information? Thank you for any help you can provide.

$\endgroup$
  • 3
    $\begingroup$ Your question says "what do I do with all this information". You may get a better response from this site if you split your question into separate more manageable and specific questions. I recommend that you do some more general reading on meta analysis and medical statistics. For what it's worth, here are a few resources that I prepared on meta-analysis: jeromyanglim.blogspot.com/2009/12/… $\endgroup$ – Jeromy Anglim May 23 '11 at 1:38
  • 2
    $\begingroup$ Consider requesting the data from the authors of the original studies... you can always do a better analysis when you have the raw data. $\endgroup$ – Michael Bishop Mar 22 '12 at 15:06
  • $\begingroup$ You might like to look at some of the resources suggested in this question: stats.stackexchange.com/questions/1963/… Also, I second Michael's suggestion, this is well worth the effort. $\endgroup$ – Freya Harrison Mar 22 '12 at 23:32
8
$\begingroup$

Your question indicates, to me, you're not yet ready to embark on the data abstraction portion of your meta-analysis. Your question needs refining, and you need to decide exactly what you're interested in asking. In your examples above, you appear to be interested in the main reported effects of the RCTs, which are found in the following places:

"Among the 866 participants who had patency assessed, the primary outcome of fistula thrombosis at 6 weeks occurred in 53 participants (12.2%) in the clopidogrel group compared with 84 participants (19.5%) in the placebo group (relative risk, 0.63; 95% CI, 0.46-0.97;P=.018)."

and

"The hazard ratio was 0.81 (95% CI, 0.47 to 1.40) in favor of aspirin and clopidogrel therapy".

All the other information is set dressing for specific study-related information, and you can probably discard it unless you think it matters to your question. Generally however, data can be abstracted into a simple Excel spreadsheet, for later analysis in the statistics package of your choice. I'd include the following fields:

  1. Study Authors
  2. Publication Year
  3. Sample Size
  4. Some coding for type of study (Case-Control, RCT, Cohort) - you can ignore this if you are really just extracting RCTs
  5. ln(Effect Estimate). So in the case of the examples above, ln (0.63) and ln(0.81) if you are looking into the main effect of treatment.
  6. Standard error of the estimates, which you can back calculate from the confidence interval.
  7. What kind of estimate was this? (RR vs. HR in your case)
  8. Multiple columns of any study covariates that interest you. What setting, country, etc. the studies were in, were they prospective or retrospective, etc.
| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ +1, eminently sensible answer! I'd add another item to your list of spreadsheet fields: nature of intervention. In the two examples given, the interventions differ (clopidogrel vs. clopidogrel plus aspirin). Esther, I'd advise thinking carefully about how heterogeneous the interventions are across your studies - is it appropriate to treat them all as the same thing? $\endgroup$ – Freya Harrison Mar 22 '12 at 23:29
  • $\begingroup$ Its a good idea to put control / treatment mean, se, and n in the spreadsheet, and then calculate derived effect estimates in the sheet or after reading in to a software for analysis. Otherwise, it may be difficult for readers to track the source of the effect estimate, or for the analyst to evaluate alternative metrics. $\endgroup$ – David LeBauer Sep 21 '14 at 19:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.