Our findings show that, despite differences in design and purposes, content, lay-out, length and initial target population of the three questionnaires, none appeared to be obviously better than the others, both when examining their characteristics and the patients' assessments. All three appeared to have advantages and drawbacks.
Response rates did not differ across questionnaires, regardless of their lengths. In previous studies, higher response rates have not been consistently associated with shorter questionnaires [3, 17, 18]. It is possible that the patients' motivation and interest in the topic may be more important than the actual length of the questionnaire.
The high proportion of missing responses found in the Saphora-Psy questionnaire could be at least partly explained by the presence of numerous questions starting with a conditional clause, such as "If you have had ...", that does not offer answer options for patients who do not meet this initial criterion. In addition, Saphora-Psy is supposed to be completed at the end of the hospitalisation, not after discharge. This is however also true for the Perceptions of Care questionnaire, which showed the lowest mean proportion of missing answers. The way questions and answers options are organised in the Saphora-Psy questionnaire may thus not be optimal. Developed for acute care inpatients, the Picker instrument includes a set of questions related to pain and surgery. Interestingly, these items did not get a high proportion of missing data, although pain may not have the same meaning for acute care and psychiatric patients, and surgery is rarely used during psychiatric hospitalisation. This may reflect the appropriateness of skip patterns.
Not surprisingly, the Perceptions of Care questionnaire, which has the lowest mean number of response categories per question, showed the highest ceiling effects. This questionnaire may therefore be less sensitive to changes and less able to discriminate at the high end of the satisfaction spectrum.
Because disease-specific health status or quality-of-life questionnaire often perform differently from generic instruments [19, 20], we were surprised not to detect differences between the psychiatry-specific and generic questionnaires. Examination of their contents suggests that the two psychiatry-specific questionnaires did not differ much from the generic instrument. An alternative hypothesis is that patients hospitalised in mental health facilities have the same basic needs and expectations as any other patients. Another possibility is that we did not consider all the discriminative characteristics allowing to detect true differences between those three questionnaires.
The randomised allocation of questionnaires, the use of three methodically developed satisfaction instruments and the relatively high number of psychiatric inpatients included were important strengths of this study. However, our study does have limitations. First, the participation rate was relatively low  but similar to those reported in other satisfaction studies of psychiatric patients, who are less likely to respond to questionnaires compared to other patients [8, 22]. The effect of selection bias on the measures of interest is unpredictable; however, because the participation rate was similar for all three instruments, the comparison of the questionnaires are likely to be internally valid. Second, the sample size could have prevented the detection of true differences between questionnaires. Third, patients did not make comparative judgements on the three questionnaires, which may have given different evaluation results. They may also not be the best judges to objectively assess questionnaires. In addition, there was a correlation between the ten evaluation questions assessing the patient's opinion about the questionnaires, with a between item correlation which was moderate on average. Even though this statistical association could suggest that these outcome measures should not be considered separately, we analysed each question separately, because we were interested in each outcome variable as such. Moreover, given the lack of differences among the three questionnaires, it would be unlikely that a composite index would show differences. Fourth, we centred our study on patients and did not assess other healthcare stakeholders opinion. Indeed, for quality improvement purposes, their opinion about the usefulness of selected satisfaction questionnaire might be of interest. Finally, we evaluated only three out of several satisfaction questionnaires.
As we did not address all possible aspects related to the selection of a satisfaction questionnaire, further research would be needed to assess, for example, the opinion of other healthcare stakeholders, whether questionnaires perform equally with all types of patients hospitalized in psychiatric wards, and which questionnaire would be best for a specific situation.