Timeline for Pooling homogenous studies vs. using meta-analysis/bayesian
Current License: CC BY-SA 4.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 1, 2018 at 7:30 | history | tweeted | twitter.com/StackStats/status/1002452351251288064 | ||
| May 31, 2018 at 15:35 | answer | added | half-pass | timeline score: 2 | |
| May 24, 2018 at 5:55 | comment | added | Giuseppe Biondi-Zoccai | Yes, do it both ways. But probably for prediction purposes the two-step strategy (with separate derivation and validation) is more meaningful. However, your sample might be not large enough for precise modeling. Check for instance this: springer.com/us/book/9780387772431 | |
| May 23, 2018 at 18:28 | comment | added | dfife | Thanks for the comment. This is a prediction, not an intervention study. The selection criteria and procedures are the same. I suppose we can analyze it both ways and (hopefully) the results show it doesn't matter. | |
| May 23, 2018 at 13:10 | comment | added | Giuseppe Biondi-Zoccai | The correct answer depends on the type of study you are conducting (eg intervention vs prediction). In addition, you should be wary of any scenario in which the all-in and the 2-step analyses are conflicting. My recommendation would be in any case, if the selection criteria and procedures are the same, to use a single analysis approach. Conversely, if there are even minor differences in selection or you aim for prediction, go first for a 2-step approach (but with individual patient-level meta-analysis). | |
| May 22, 2018 at 16:48 | history | asked | dfife | CC BY-SA 4.0 |