The aritcle DOI: 10.1093/annonc/mdw161 has hazard ratio and 95% CI for three comparisons,
low-grade/homologous vs low-grade/heterologous,
low-grade/homologous vs high-grade/homologous,
and
low-grade/homologous vs high-grade/heterologous.
However, I would like to obtain hazard ratio and 95% CI for
high-grade vs low-grade.
I figured out that by using network meta-analysis, I could get hazard ratios for
low-grade/heterologous vs high-grade/homologous and
low-grade/heterologous vs high-grade/heterologous.
but I am uncertain what to do to collapse the HRs into
low- vs high-grade HR.
Table 2 of the cited paper has HR for low hetero, high homo hetero vs low homo as reference. low hetero is not significantly different to low homo so they could be combines as reference level of 1.
a crude approach could be simply to take the mean of the high homo and high hetero HRs (eg for overall survival univariate (1.83 + 1.49)/2 = 1.68) . Not sure how to deal with CIs but a very conservative approach would be to take the widest interval ie 1.04-2.78. For best estimate you might have to ask authors for raw data.
this would give a high/low HR estimate of 1.68 (1.04- 2.78)
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis