We developed this supportive-care patient-reported outcome using multistep methods following COSMIN recommendations. [5]
This project obtained a favorable opinion on September 11, 2019 from the Ethics Committee of the CHU La Pitié Salpêtriere, Paris [CCP Ile de France 6- 2019-A00204-53) and was conducted according to the Declaration of Helsinki. According to French regulations for this type of research [Reference Methodology 003], no written informed consent had to be obtained from participating patients, and an information and nonobjection note was given to them.
A standard methodology for the creation of a patient-reported outcome (PRO) was used, and a multidisciplinary working group of experts (professionals, such as physicians, public health experts, sociologists, supportive care experts, and socioestheticians) was created.
The questionnaire followed the recommendations proposed by Seidenberg et al. and Leidy et al. for the development of a self-report questionnaire. [6]
The same group of experts reached a consensus on how to respond. The choice was made to use a numerical visual scale ranging from 0 for "No, not at all" to 10 for "Yes, always".
To avoid missing data as much as possible, respondents had the option of not expressing themselves if they were not concerned by placing the cursor on zero.
To avoid any confusion regarding timing, all questions began with "currently".
The diversity of the recruiting centers ensured broad recruitment and a consistent diversity of patients in terms of pathologies, geographical location, age and sociological status. The participating subjects, aged at least 18 years, had to have received supportive care, SE in this case.
The construction of this questionnaire followed several steps: elaboration of the questionnaire, measurement properties of the questionnaire, internal and external validation, test-retest validation and translation, cross-cultural adaptation and cognitive debriefing.
Step 1: Elaboration of the questionnaire
We conducted a literature search on PubMed to identify published questionnaires or scoring systems related to SE. We used the algorithm developed by the COSMIN initiative and combined it with the terms “socioesthetic" and the medical subject subheading “aesthetics” as a major topic.
We then conducted a series of 23 individual interviews to elicit expectations and needs of patients in terms of management of physical appearance, self-esteem and relaxation as well as the role of SA in patient support. Using a triangulation approach, we interviewed 10 patients (who were using or used SA) [7] as well as 6 SE practitioners, five medical professionals advising SE and 2 members of the SE association. The verbatim data were analyzed using an itinerary method. [8,9,10,11] Such methods aim at contextualizing and understanding the entire process related to an intervention (when was it proposed, why, what it did, why was it stopped). The result of this analysis was then translated to a series of patients, trying to stay as close as possible to the patients’ words verbatim.
Finally, a multidisciplinary working group involving researchers, methodologists, clinicians (medical, paramedical, sociologist, supportive care expert and SE practitioner) and patients designed a preliminary questionnaire from the qualitative analysis. Each item of the questionnaire was evaluated with a numerical visual scale ranging from 0 for "No, not at all" to 10 for "Yes, always". To avoid missing data as much as possible, the respondents had the option of not expressing themselves if they were not concerned by placing the cursor on zero.
Step 2: Measurement properties of the questionnaire
The measurement properties of the questionnaire were assessed in six steps:
- reduction of the number of items,
- subscale repartition of items and assessment of factorial validity,
- assessment of construct validity,
- assessment of reliability,
- evaluation of the minimal clinically important difference (MCID) and
- transcultural validation in US English.
Reduction of the number of items. We reduced the number of items based on redundancy, which was suspected when the interitem correlation evaluated by the Spearman correlation coefficient was above 0.8 [12]. We wanted our questionnaire to be useful in all contexts of SE interventions. These items can differ greatly, as do benefits for patients. Therefore, we decided to exclude only items presenting a major ceiling or floor effect and did not exclude items because of low interitem correlation.
Subscale repartition of items and assessment of factorial validity. Suitability for factorial analysis was assessed by a Kaiser-Meyer-Olkin (KMO) value above 0.8 for all items [13]. A scree plot with parallel analysis was performed to determine the number of subscales. To allocate each item to a subscale, exploratory factor analysis (EFA) with a cluster rotation was performed. Items presenting a factor loading lower than 0.5 or a cross factor loading higher than 0.2 were excluded from the questionnaire [14]. We assessed internal consistency with Cronbach's alpha [15] and McDonald’s omega coefficients, aiming to have both >0.7 for the entire questionnaire and each subscale. [16 ] Readability of the final questionnaire was assessed with a SMOG index corrected for French. [17,18]
Factorial validity was assessed using a higher-order factor confirmatory analysis. This
analysis confirmed that the subscales could be combined into a single score. The model’s suitability was assessed using three criteria: comparative fit index (CFI) above 0.9, Tucker-Lewis index (TLI) above 0.9 and root mean square error of approximation (RMSEA) under 0.08.
Construct validity. We hypothesized several relationships between the SE score and other constructs. Specifically, there was a moderate negative correlation with the stress evaluated through the Perceived Stress Scale (PSS),[19] a moderate positive correlation with the mental dimension of the SF12 [20] (r = 0.5-0.7) and a low correlation (r=0-0.3) with the physical dimension of the SF12, the satisfaction Visual Analog Scale [VAS] for SE intervention and well-being evaluated by the well-being questionnaire (WBQ12). [21]
Reliability. Fifty participants who did not previously answer the questionnaire were asked to complete it once and then again three weeks later. Reliability was measured by the intraclass correlation coefficient (ICC) for agreement and presented on a Bland and Altman plot. [22,23]
Evaluation of the minimal clinically important difference (MCID [24]). The MCID represents the smallest modification detectable in a clinical setting. It was determined using the standard deviation method and the standard of mean method. Using two methods made our results more robust.
An American English version of the questionnaire was developed by translation and transcultural validation according to the ISPOR task force recommendation. [25]
All data were analyzed using R software version 3.613 for Windows. The significance level was set at p < 0.05.