Abstract
Purpose :
Provide calibrated item estimates for the National Eye Institute Visual Function Questionnaire (NEI VFQ-25) in common retinal diseases.
Methods :
Nine datasets using the NEI VFQ-25 were pooled for analysis. Seven datasets (n=2770) from retina studies using anti-VEGF therapy for AMD, retinal vaso-occlusive disease, and diabetic retinopathy were joined with two datasets (n=572) from low vision studies representing a mix of disorder diagnoses. Data from samples included ratings of patient responses to the NEI VFQ-25 for the subset of items with the same number of response categories. A composite item was created for the conditional questions regarding driving. Eligible patients for all studies were 18 years or older and responses to NEI VFQ-25 items from 3342 patients were pooled for analysis.
Rasch analysis was performed on pre-intervention NEI VFQ-25 responses to estimate item and person measures, reliability coefficients, and response category thresholds. Differential item functioning (DIF) analysis was first performed on the 9 datasets to determine the appropriateness of combining datasets by examining the DIF magnitude both through logit and probability value estimates. After combining the datasets, a DIF analysis was performed to evaluate the effects of covariates, including age and sex.
Results :
Patients ranged in age from 19 to 103 years (median 72, mean 70 years), with males representing 48% of the sample and 83% of responses were from participants initiating anti-VEGF therapy for retinal disease. Mean visual acuity for the better-seeing eye was 20/40.
Minimal amounts of DIF across datasets facilitated data pooling; only 3 out of the 28 items across all datasets showed both clinically meaningful DIF (>0.5 logits) and statistically significant DIF (using a Bonferroni correction). The item related to difficulty shaving/styling hair showed DIF by sex (males endorse less difficulty), while the item related to driving at night showed DIF by both sex and age. Patients 60 and younger found driving at night easier while females and patients 75 and older found the task harder.
Conclusions :
Minimal DIF among datasets allowed us to pool data to maximize precision in these item estimates and provide a set of calibrated item measures that might improve comparative research with the NEI VFQ-25 for any sample size.
This is a 2020 ARVO Annual Meeting abstract.