Centrally adjudicated vs. investigator-reported outcomes in randomized heart failure trials

Abstract

Background and aims: Heart failure endpoints in cardiovascular outcome trials are commonly identified through centralized adjudication of investigator-reported events. It remains unclear whether central adjudication improves the accuracy of treatment effect estimates in terms of log[hazard ratios (HR)].

Methods: Data from seven cardiovascular outcome trials with >1000 patients that included centrally adjudicated heart failure outcomes were utilized to assess (i) the concordance between investigator-reported and centrally adjudicated heart failure and cardiovascular death events; (ii) rates of subsequent all-cause mortality following positively vs. negatively adjudicated heart failure events; and (iii) the correlation of log(HR) based on centrally adjudicated vs. investigator-reported events.

Results: Positive adjudication rates for investigator-reported events varied widely across trials, but were generally higher for cardiovascular death (range: 87.9%-99.2%) than for heart failure hospitalization (range: 61.6%-88.0%). The risk for subsequent all-cause death was similar for positively and negatively adjudicated heart failure hospitalizations. Log(HR) correlated well for cardiovascular death [R2 = .80, 95% credible interval (CrI): 0.53-0.93] and the composite of cardiovascular death or heart failure hospitalization (R2 = .79, 95% CrI: 0.46-0.93), but less for heart failure hospitalization (R2 = .57, 95% CrI: 0.10-0.83).

Conclusions: Positive adjudication rates were lower for heart failure events than cardiovascular death, but even negatively adjudicated heart failure events are prognostically important. Central adjudication of events did not alter the results (precision or estimated log(HR)), though some variation was observed, depending on the indication. The results suggest that the decision to pursue centralized adjudication of heart failure events in a specific trial may need to be individualized.

Keywords: Adjudication; Cardiovascular trial; Heart failure; Investigator-reported outcomes.

Structured Graphical Abstract

The graphical abstract figure shows the results of 7 studies for which investigator-reported and adjudicated heart failure outcomes were available, illustrates the agreement of respective log hazard ratios for cardiovascular death, and summarizes the key findings.

Figure 1

Forest plot of HRs using adjudicated and investigator-reported events for (A) CV death, (B) HFH, and (C) composite of CV death and HFH. Adj., adjudication-confirmed events; CV, cardiovascular; HF, heart failure; Inv., investigator-reported events; vs., versus

Figure 1

Forest plot of HRs using adjudicated and investigator-reported events for (A) CV death, (B) HFH, and (C) composite of CV death and HFH. Adj., adjudication-confirmed events; CV, cardiovascular; HF, heart failure; Inv., investigator-reported events; vs., versus

Figure 2

Correlation between HRs using adjudicated and investigator-reported events for (A) CV death, (B) HFH, and (C) composite of CV death and HFH. The dotted line represents the reference line indicating perfect agreement between HR based on adjudicated and investigator-reported endpoints. The line of best fit, with the corresponding 95% prediction interval, is shown in black. The circles reflect each treatment comparison, with the area of the circle being proportional to the amount of information. α: estimated intercept, β: estimated slope. Perfect agreement would occur if α = 0 and β = 1, and CrIs excluding these values would provide evidence against perfect agreement. CV, cardiovascular; HF, heart failure; HFH, heart failure hospitalization; HR, hazard ratios; MI, myocardial infarction; R², coefficient of determination

Figure 2

Correlation between HRs using adjudicated and investigator-reported events for (A) CV death, (B) HFH, and (C) composite of CV death and HFH. The dotted line represents the reference line indicating perfect agreement between HR based on adjudicated and investigator-reported endpoints. The line of best fit, with the corresponding 95% prediction interval, is shown in black. The circles reflect each treatment comparison, with the area of the circle being proportional to the amount of information. α: estimated intercept, β: estimated slope. Perfect agreement would occur if α = 0 and β = 1, and CrIs excluding these values would provide evidence against perfect agreement. CV, cardiovascular; HF, heart failure; HFH, heart failure hospitalization; HR, hazard ratios; MI, myocardial infarction; R², coefficient of determination