Background/Purpose: The Stanford Health Assessment Questionnaire Disability Index (HAQ-DI) is a widely used patient reported outcome for functional disability in RA. Minimal clinically important differences (MCIDs) have previously been calculated based on the original ordinal HAQ-DI scores resulting in a dependence of the MCIDs on baseline scores(1), causesing limited generalizability and validity of the proposed MCIDs. The Rasch model provides a transformation of ordinal scores to an underlying latent scale with interval scale properties. This requires adequate fit of HAQ-DI data to the model. We aimed to 1) examine internal construct validity of the Danish version of the HAQ-DI (scored without correction for devices) and 2) determine a MCID of the HAQ-DI in a cohort of RA patients initiating anti-rheumatic treatment in clinical practice based on the transformed linear logit scale.

Methods: RA patients registered in the DANBIO registry at Center for Rheumatology and Spine Diseases, Copenhagen, Denmark were included in the study if HAQ-DI and Patient Global Visual Analogue Scores (PtGl) scores were available at a baseline visit and after > three months of follow-up after initiation of a) first synthetic Disease Modifying Anti Rheumatic Drug (sDMARD) (disease duration < 12 months) or 2) first biological drug (disease duration > 12 months). The Rasch model was fitted to HAQ-DI data at baseline and follow-up and item fit evaluated in separate analyses by comparing observed and expected item-restscore correlations (2). MCID was calculated as the median changes of A) the original and B) logit HAQ-DI score in the group of patients who had improved by 15-30 mm (ie. minimal improvement) on a 0-100 mm PtGl scale. 

Results: 362 RA patients were included in the study. HAQ-DI data showed acceptable fit to the Rasch model as no significant evidence of misfit was disclosed at baseline and only a single misfitting item (the “Grip” sub-item, p=0.01) was identified at follow-up. Consistent item ranking across time indicated instrument invariance. Changes in ordinal and logit HAQ-DI scores were strongly correlated (Spearmans rho = 0.935, p <0.001), but the association between them is not completely linear. Sixty-one patients had an improvement > MCID on the logit scale (0.48), but no improvement on the ordinal scale (MCID 0.250), while no patients had the opposite pattern.

Conclusion: The Danish version of the HAQ-DI scored without correcting for devices showed acceptable internal construct validity and thus a MCID based on a linear transformed scale could be calculated for this study (0.48). Application of the logit MCID classified an additional 17% of patients as having achieved a MCID compared to the MCID calculated on the ordinal scale. This finding has potential implications for the use of ordinal-based MCID and the powering of future studies. References:

1.      Doganay et al.J Rheumatol. 2016 Jan;43(1):194-202

2.      Kreiner S. Appl. Psychol. Meas. 2011;35(7):557-561.