Estimating risk differences in patient-important events when studies have focused on surrogate outcomes: Application to systematic reviews and practice guidelines

Article type
Authors
Guyatt G1, Akl E2, Murad M3, Falck-Ytter Y4, Gould M5, Eikelboom J1
1McMaster University, Canada
2State University of New York at Buffalo, USA
3Mayo Clinic, Rochester, USA
4Case Western Reserve University, USA
5Keck School of Medicine of the University of Southern California, USA
Abstract
Background: Four chapters in the ninth iteration of the American College of Chest Physicians antithrombotic guidelines addressed thrombosis prevention. Assessing risk differences (RDs) in symptomatic thrombosis proved problematic because most studies focused on a surrogate outcome, asymptomatic thrombosis.

Objectives: To implement optimal strategies assessing RD.

Methods: Panels considered three approaches: i) using estimates of symptomatic thrombosis from randomized controlled trials (RCTs) for both control group risk (CGR) and relative risk (RR); ii) using estimates of symptomatic thrombosis from RCTs for CGR and estimates of a composite of asymptomatic and symptomatic events from RCTs for RR; and iii) using estimates of symptomatic thrombosis from observational studies for CGR and estimates of a composite of asymptomatic and symptomatic events from RCTs for RR.

Results: All three approaches proved feasible and applicable; examples follow. The chapter dealing with prophylaxis in hospitalized medical patients used approach A and estimated RD −34 symptomatic thrombotic events per 1000 (95% CI -51 to 0), limited by imprecision. The chapter addressing orthopedic surgical patients, using approach B, chose an untreated CGR from the low-molecular-weight heparin (LMWH) arm of recent RCTs after accounting for the effect of LMWH against placebo in earlier studies. Using the RR from RCTs for the combined endpoint of symptomatic and asymptomatic deep venous thromboses (DVTs), they estimated the effect of unfractionated heparin as RD of -10 (95% CI -9 to -12). Using asymptomatic events required rating down for indirectness. The chapter addressing non-orthopedic surgery thromboprophylaxis used approach C for the intermittent pneumatic compression recommendation. An observational study provided an estimate of 26 events per 1,000. Use of RR from RCTs resulted in an RD estimate of -13 per 1000 (95% CI -7 to -20) rated down for indirectness because asymptomatic events were used for RR.

Conclusions: A variety of approaches estimating risk differences when RCTs have focused on surrogate endpoints proved useful.