0
$\begingroup$

I am conducting a meta-analysis which compares the survival outcomes between patients undergoing different techniques for the same problem. I need to compare the survival of these different groups to test for significant differences. However, I only have time-point survival data from the studies I am reviewing i.e. survival at 1, 3 and 5 years, without the detailed survival data that we would usually use to construct a KM curve.

I do have the absolute numbers for the survival e.g. studyA 1-year survival 5/10 (50%), studyB 1-year survival 10/100(10%) etc

I've looked this up and all I can find is time-point comparison when all survival data is known.

I am using R for stats analysis

Many thanks for any help with this

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

As you note, you can compare proportions at any specific time of interest with standard methods, provided that you have complete data.

One problem is censored data. If, for example, you are evaluating 3-year survival and some event-free cases don't have full 3-year follow up (censored), and particularly if censoring patterns differ between groups, then your meta-analysis might not be interpretable. If each of the Studies reported its own point estimate and confidence intervals of 3-year survival that took censoring into account, however, you could work with those values.

You also might need to consider differences among groups in terms of other variables associated with outcome. For example, if the patient population in Study B was sicker than that in Study A, you would need to account for that.

Whether you will be able to account for censoring and other variables in your meta-analysis will depend on details of how thoroughly the Studies reported their data.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Assuming the study populations are similar in other respects, which test could I use for the comparison? A 2x2 table with OR and p-value? $\endgroup$
    – Keno
    Commented Feb 9, 2021 at 13:04
  • 1
    $\begingroup$ @Keno that depends a lot on the specific details of the information that you have. p-values reported by studies shouldn't be used directly, although information that went into calculating them (survival fractions at a time of interest, numbers of events/cases, etc) would be included. I don't do meta analysis professionally myself, just know some general principles. Look into the R metafor package and its associate website for helpful tools and examples. $\endgroup$
    – EdM
    Commented Feb 9, 2021 at 15:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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