Purpose: Researchers are increasingly using the Analytic Hierarchy Process (AHP) for measuring patient preferences. Prior research has shown that use of the AHP improves patients' process of decision making by lowering their decisional conflict but the accuracy of some patients' preferences is questionable because of inconsistencies in their pairwise comparisons. When a patient's inconsistency, measured by AHP, is less than 0.10, the preference weights generated by AHP are considered accurate. We tested two different computer interfaces of AHP to determine which format had the lowest inconsistency for weighing risks and benefits related to future childbirth strategies for women with prior cesarean deliveries.

Methods: We recruited 96 women who had just had a cesarean delivery from the postnatal unit of a large university hospital and who had not had a tubal ligation or hysterectomy. After all patients completed a baseline computerized questionnaire, they were randomized to either making pairwise comparisons using a verbally anchored scale with radio buttons (n=48) or a numerical scale with a sliding bar (n=48). For both AHP interfaces, the women received summary information on their preference weights related to having a repeat cesarean or attempting a vaginal delivery. Our analyses focused on measuring decisional conflict (baseline and after using an AHP interface) and internal inconsistency of the two formats.

Results: The two groups were similar in age, health insurance status, obstetric history, and baseline decisional conflict. ANOVA revealed that both groups of patients significantly reduced 4 of 5 decisional conflict measures (knowledge, p<.001; values clarity, p<.001; support, p<.001; and overall decisional conflict, p<.001) to the same extent after using either format. However, for comparisons that included risk to the mother or baby, patients randomized to the numerical format were more consistent in comparisons than patients randomized to the verbal format (mean inconsistency: numerical, 0.05 vs. verbal, 0.09; p=0.015). For comparisons involving only non-risky delivery factors (convenience of scheduled delivery, desire for labor, etc.), inconsistency did not differ between the two formats (numerical, 0.10 vs. verbal, 0.11; p=0.452).

Conclusions: While both formats of AHP reduced decisional conflict, patients were more consistent in decision making involving risk when the format was numerical. This insight may aid in design of future computerized AHP systems to accurately measure patient preferences.