Measurement Bias with Mixed-mode Patient-reported Outcome (PRO) Survey Administration: Measurement Equivalence, Cost, and Data Quality
Background: Measurement with mixed-mode administration (i.e., paper-mode versus web-mode) of patient-reported outcome (PRO) health surveys may vary by psychometric measurement equivalence (ME), differential response rates, costs, and data quality.
Purpose: Three data-based papers evaluated ME, cost, and data quality when a mixed-mode approach for self-administration was used.
Methods: Data were derived from the Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) study, a multi-institutional United States based longitudinal prostate cancer registry. ME study used a randomized cross-over design of 209 participants. Cost and data quality analyses used a cross-sectional time frame of 5,008 participants. The Short Form-36 (SF-36) and the UCLA-Prostate Cancer Index (UCLA-PCI) were examined.
Results: ME study, participants were White (97%), college educated (66%), reported an annual income > $75,000 (46%), and a median age of 69 years. Intraclass correlation coefficients were high (ICC = 0.66-0.97). Exact percent agreement was high (> 0.89). For the cost and data quality analyses, 90% opted for paper-mode and 10% for web-mode. Total cost to process 5,008 surveys was $75,216 or on average $15.02 per survey with web-mode costs significantly higher ($18.47/survey) than paper-mode ($14.66/survey). Web-mode surveys had lower error rate (9% versus 14%, NS). Overall response rate was 77%, paper-mode 76% versus 88% web-mode (p <0.01). Predictors of response were older age, being Caucasian, having attained college education, and living in a significant relationship. Paper-mode participants were less likely to respond (OR .69, 95%CI = .68-.70) and had significantly more missing data on the SF-36, the UCLA-PCI, and fewer computable scale scores (all significant p < 0.001). All mean scale scores were lower among paper-mode participants but effect sizes for clinically meaningful differences between modes were small. CaPSURE participants had higher physical and mental function when compared to US based age-stratified norms.
Discussion: The use of a mixed-mode approach found support for ME; costs to administer were lower for paper-mode but web-mode had higher accuracy; and significant but small differences in data quality. Findings suggest that mixed-mode administration did not introduce significant measurement bias but allowed for participation by more diverse participants (e.g., older, poorer health, non-Caucasian).