Effects of Disease Duration and Outcome Definition on Power of Rheumatoid Arthritis (RA) Clinical Trials

[Guest guest]

Effects of Disease Duration and Outcome Definition on Power of Rheumatoid

Arthritis (RA) Clinical Trials

Category:  17 RA‹treatment

J J 1, J A Bolognese2, D T Felson1

1Boston University, Boston, MA;2Merck Research Laboratories, Rahway, NJ

Presentation Number: 1392

Poster Board Number: 347

Keywords: statistical power, RA clinical trial

In RA, those with early disease may respond better to treatment (Rx) than

those with later disease, suggesting that trials in early disease could be

done with fewer subjects. This has not been tested for all types of Rx.

Power of some core set based composite outcomes may be reduced because of

varying response per core set item. All of these considerations may be

relevant in trial planning. We tested whether outcome and Rx affect the

power of RA trials and whether better response in early disease occurs

across different Rx.

We used simulation studies to assess the likely impact on statistical power

for several outcome definitions in 2 RA trial contexts: treatment with a

DMARD (methotrexate vs auranofin for up to 48 weeks), and with a ib

(washout/flare trials of Rofecoxib 25mg vs Placebo for 12 weeks). We used

the trials¹ baseline and change data in simulating Rx comparisons for both

early disease (<4 years) and later disease (4-9 years duration). We

estimated power for 4 outcomes (2 data-driven - outcome definition varies

per patient, and 2 with an a-priori definition of response). The ACR20 and

improvement in Disease Activity Score (DDAS) have predetermined clinically

valid definitions, while nACR (the number of core set items improving by

20%) and O¹Brien¹s test (the most powerful non-parametric test involving all

7 core items) are data-driven.

Baseline levels and changes with Rx varied per case while baseline and

change variance-covariance matrices varied per trial. Each core set item was

standardized to a 0-100 scale, with 100 = worst status. As baseline levels

and changes in items were approximately normally distributed, we generated

corresponding multivariate normal data for 1000 trials per case. Rx effect

was the difference in mean change in a core set item between Rx¹s.

The table shows derived sample sizes (N) per Rx group required for 80% power

in 2-sided tests for each of O¹Brien¹s test, nACR, ACR20, and DDAS per case.

Each case is characterized by the mean and range (min, max) of baseline and

Rx effect means per core set item. Baseline levels and effect means were

more variable across the items in the ib compared with the DMARD cases.

The data-driven outcomes (O¹Brien¹s test, nACR) are uniformly more powerful

than the a priori outcomes. Response in early disease was better in the

DMARD, but not in the ib case, with consequent effects on statistical

power. When drug effects are smaller and inconsistent across core set items,

the power of the a priori outcomes is reduced, even more for DDAS than for

the more broadly based ACR20. In sum, early disease response is Rx

specific,and variability of response across core set items affects the

relative power of different definitions of outcome in RA trials.

Case Core set items baseline means(min, max) Rx effects on items:

mean (min, max) N-O'Brien N-nACR N-ACR20 N-DDAS

DMARD: early 42 (32, 65) 7 (4, 12) 64 80 140 130

DMARD: later 51 (35, 69) 6 (2, 10) 100 115 224 208

ib: early 46 (25, 72) 3 (1, 8) 200 268 652 859

ib: later 53 (26, 71) 5 (-1, 7) 100 130 198 285