Purpose: To determine whether a conjoint analysis can be designed to obtain valid results from patients with uncorrectable vision impairment.
Methods: Eighty-three presenting patients at the Low Vision and Rehabilitation Service at the Wilmer Eye Institute at Johns Hopkins were consented to participate in a conjoint analysis comparing characteristics of low vision rehabilitation regimens. The five attributes of each choice were out-of-pocket price, number of visits required, satisfaction with reading, satisfaction with television viewing, and difficulty arranging for transportation. Each attribute had three levels. Attributes and levels were described to the patient before any choice sets were described. When describing the choice set and making a choice, the levels of attributes were displayed as a series of horizontal bars: green representing the most favorable level, yellow—the middle level, and red—the least favorable level. Each respondent was presented with 12 choices. We report the proportion of non-responses, summary statistics of the time required to make choices, and the implied tradeoffs. Differences in times required to make choices later in the choice set were analyzed using a linear regression with dichotomous indicators of each question number and clustering at the individual level. Tradeoffs were inferred using a conditional logit with clustering at the individual level and dichotomous indicators of the levels of functioning and a continuous measure of out of pocket price and number of visits.
Results: Subjects responded to all choices in all choice sets. The average time required to respond to the first conjoint question was 50.1 seconds (SD 39.9). The average time for the last question in the choice set was 17.0 seconds (SD 17.8). Based on regression results, even the second question required significantly less time than the first (18.1 seconds, p<0.001) All coefficients on the levels of attributes were of the expected sign; only the middle level of satisfaction with television viewing was not statistically significantly different from zero. The coefficients for somewhat and very dissatisfied with each functional outcome were different from each other.
Conclusion: Presenting patients with uncorrectable visual impairment responded to all choices presented using colored bars to indicate levels of attributes in a way that demonstrated that they understood the choices, made later choices with increasing efficiency, and implied reasonable tradeoffs among levels of the attributes.
Candidate for the Lee B. Lusted Student Prize Competition