140 Is death-or-hospitalisation in heart failure trials a reliable proxy for mortality itself?

Background Clinical trials in heart failure often target not just mortality but a composite endpoint, frequently encompassing hospitalisations. This increases the number of events, achieving greater statistical power for any given sample size. Often, we find that the resulting hazard ratio is "driven by" the non-mortality events. Does that mean we should not expect a benefit in mortality? Method Aggregating multiple trials by meta-analysis to increase precision, we tested whether the mortality odds ratio is significantly different to the composite endpoint odds ratio. We grouped trials by therapy class, extracted data on death and hospitalisation, and calculated the class effect odds ratios for death, and for death-or-hospitalisation, using inverse variance-weighting and a fixed effect model. Results 54 blinded randomised controlled trials of treatment for heart failure with a total of 115,571 individuals were eligible for this analysis. They spanned 16 treatment classes: beta blockers, angiotensin converting enzyme inhibitors, angiotensin receptor neprilysin inhibitor, mineralocorticoid antagonists, sodium-glucose co-transporter 2 inhibitors, cardiac resynchronisation therapy, ivabradine, digoxin, milrinone, intravenous iron, aliskiren, serelaxin, nesiritide, etanercept, dronderanone and pimobendan. The odds ratios for death and death-or-hospitalisation are strongly correlated (r=0.78, p<0.001), as expected. There is a slight tendency for the odds ratio for death to be more extreme than that for death-or-hospitalisation: the weighted Deming regression relationship between the two odds ratios shows a gradient of 1.26 (95% CI 0.49–2.43)(figure 1). Death-or-hospitalisations were substantially more frequent than deaths by a factor of 2. Conclusion When a blinded trial of heart failure reports an improvement in death-or-hospitalisation that is "driven by hospitalisation" with no significant improvement in mortality, we need not assume that the benefit really is limited to hospitalisation. In fact, the mortality effects seem to be slightly stronger than the death-or-hospitalisation effects. This can only be seen when we use meta-analysis to increase the precision of the odds ratios. Having 2-fold more events means a study can be 4-fold smaller, cheaper and quicker. The use of death-or-hospitalisation as a composite endpoint in blinded randomised controlled trials of heart failure is therefore valid and should be encouraged. Conflict of Interest None