ADAPTIVE CONJOINT ANALYSIS AS INDIVIDUAL PREFERENCE ASSESSMENT TOOL: FEASIBILITY THROUGH THE INTERNET AND RELIABILITY OF PREFERENCES

Background
Adaptive conjoint analysis (ACA) has been developed for internet administration, but most healthcare applications have used face-to-face interviews so far. Further, an earlier study showed individual-level test-retest reliability to be limited, possibly due to value change between administrations.

Purpose
To assess a) consistency, internal validity, and subjective evaluation across modes of administration, and b) reliability using two retests.

Methods
Randomly selected disease-free rectal cancer patients were asked to complete a computerized ACA-task three times, either with a non-intervening interviewer present (N=51, 45%) or through the internet (N=62, 55%). The ACA asked for 12 tradeoffs between probabilities for survival, local control, fecal incontinence, and sexual problems. Combinations of outcome probabilities to be traded could vary between tests. Following ACA 3, participants were asked close-ended evaluative questions about ACA. For each test a) relative importances for the four outcomes and b) treatment preferences (surgery versus preoperative radiotherapy followed by surgery) were derived from ACA-utilities.

Results
Ninety-five (84%) participants completed the three tests, at a mean of nine days interval (range, 6-21). The laptop versus internet group was significantly older (66 vs. 61 years, P<0.001) and completed the ACA at a one day shorter interval (P<0.05).    
   Consistency in attribute-level utilities rank order, willingness to trade survival for benefit in local control, incontinence, or sexual problems did not significantly differ between groups, nor did the groups differ in relative outcome importances, ease of understanding the task instructions, doubt about tradeoffs, or in how pleasant or confronting they evaluated the ACA.
   Between ACA 1-2 and 2-3, intraclass correlations were moderate for survival, local control and incontinence (ICC 0.42-0.62), and low for sexual problems (0.13 vs. 0.21); mean differences in importance scores did not significantly differ; and rank order of outcomes was identical in 32% to 40% of participants. ACA-derived treatment preferences did not significantly vary over time.

Conclusions
These results provide evidence for ACA as internally valid and equally acceptable as internet application for healthcare issues. Individual-level importance scores did not seem to converge over time. Variation in importance scores may not result in varying treatment preferences over time.