Background: The objective was to solve two problems of an already validated scale measuring inpatient opinion on care: 1) a high non-response rate for some items due to the "not applicable" response option and 2) a skewed score distribution with high ceiling effect.
Methods: The EQS-H scale ("échelle de qualité des soins en hospitalisation") comprised 26 items and 2 sub-scales of 13 items each, 'quality of medical information' (MI) and 'relationships with staff and daily routine' (RS). Three studies were conducted: a first mono-centre study (n = 552, response rate = 83.4%, self-completion of the scale the day before discharge) to construct a shorter version of the scale without the items with high non-response rate and maintaining those useful to ensure good internal validity (construct, convergent and divergent) and reliability; a second mono-centre study (n = 1246, response rate = 77.9%, self-completion of the scale before discharge) to confirm psychometric properties of the new version; a third multi-centre national study (n = 886, response rate 41.7%, self-completion at home 15 days after discharge) to test a new response pattern in order to reduce ceiling effect.
Results: Six items having a non-response rate >20% were deleted, increasing rates of exhaustive response to all items from 15% to 48%. Factorial analysis supported the evidence for removing 4 more items to ensure good internal validity and reliability of the new version. These good results (initial variance explained: 43%; Cronbach's alpha: 0.80 (MI) and 0.81 (RS)) were confirmed by the second study. The new response format produced a normalisation of the 2 scores with a large decrease in ceiling effect (25% to 4% for MI subscale and 61% to 8% for RS). Psychometric properties of the final version were excellent: the 2 subscales (8 items each) explained 66% of the variance in principal component analysis, Cronbach's alpha were respectively 0.92 (MI) and 0.93 (RS).
Conclusion: The new version of the EQS-H has better psychometric properties than the previous one. Rates of missing values are lower, and score distribution is normalized. An English version of this scale focused on quality of medical information delivered and on relationship with staff already exists, and this could be useful to conduct cross-cultural studies of health care service quality.