The diabetes-39 questionnaire

The instrument used was the D-39 questionnaire, a multidimensional scale developed in United States of America [21], which consists of 39 items grouped in five dimensions: Energy and mobility (15 items), diabetes control (12), social burden (5), anxiety and worry (4), and sexual functioning (3). The D-39 is used to assess the QoL of patients with type 1 and type 2 diabetes; regardless of their treatment regimen [16]. Patient themselves could rate their QoL during the last month for each item. Each item can be answered using a seven-point scale ranging from 0.5 (not affected at all) to 7.5 (extremely affected). Each of the five dimensions were summed up, and the resulting raw scores were transformed into scales ranging from 0 to 100 using a linear transformation: (raw score − minimum value)/(maximum value − minimum value) × 100 according to the developer’s instructions. The questionnaire also includes two supplementary “overall ratings”, in which respondents use the same seven-box Likert scale to evaluate their perceived overall QoL (ranging from “lowest quality” to “highest quality”) and the severity of their diabetes (ranging from “not severe at all” to “extremely severe”). The self-rating overall health status was also evaluated with a single item asking participants to rate their overall health using a five-point Likert scale [22]. Participants responded to this item from ‘’one = very poor’’; ‘’two = poor’’; ‘’three = moderate’’; ‘’four = good’’ and ‘’ five = very good).

Translation and cross-cultural adaptation proceduresTranslation of D-39 to Kinyarwanda, back-translation and consensus version in English

Permission to use the D-39 was obtained from the D-39 developers. Then, for the translation of the questionnaire, we used a standard approach [23] coupled with known steps in the process of adaption [24, 25]. The translation was carried out by two native Rwandans. One of the translators possess a university degree in English literature and has twelve years of work experience, while the other is a medical doctor with work experience of seven years, who also has taught English for more than eight years. Both have a certificate of proficiency in English, and they translated the questionnaire into Kinyarwanda independently, following an ‘item intent’ guide. The two translations were synthesised into one, addressing any discrepancies. The Kinyarwanda back-translation of Kinyarwanda questions into English were done by two English native speakers. One of the back-translators has a university degree in International Studies, while the other one has studied linguistics, African studies and computational linguistics. Both back-translators had excellent language skills in Kinyarwanda. They back-translated the Kinyarwanda questions into English, while blinded to the original version. Subsequently, the two backward translations were reconciled into one.

Assessment by expert committee

An expert committee was set up comprised of seven members, including two Rwandan forward translators, one of the back-translators, an epidemiologist, a local bilingual representative, and the two researchers conducting the study. All members of the committee were fluent in English. The aim was to appraise the results of the translations, evaluating their semantic, idiomatic, experiential and conceptual equivalence, and produce a pre-final version. A report was prepared providing an account of these steps, the controversial items, and the ways they were resolved in the consensus translation. The report and the pre-final version were shared with the questionnaire developer, and consent was received.

Pre-testing

The pre-final version was assessed by conducting interviews using a sample of patients (n = 26) with DM. The objective was to evaluate patient comprehension of the translated questions and the answer categories whether respondents could retrieve relevant information from memory, the effort required to answer the degree of interest and social desirability bias. To attain maximum variability of the participants, the interviews were conducted in four different hospitals. After each round, modifications were proposed for some items, based on the interview transcripts and notes. A new iteration of the questionnaire was then prepared and tested in the following round. Lastly, a final version was produced, and a report was made available to the original D-39 developers.

Psychometric properties and statistical analysisStudy participants, data collection and sample size

Evaluating Psychometric Properties of D-39 was part of a cluster-randomised controlled trial (RCT), that aimed at determining the efficacy of an integrated mobile-health and community-health-worker programme for the management of diabetes in primary healthcare in Rwanda. The protocol of this RCT (ClinicalTrials.gov registration: NCT03376607) consists of a mixed-methods study, and has been published elsewhere [26].

For the purpose of conducting the confirmatory factor analysis (CFA), at least 200 participants would be necessary [24, 26,27,28]. The power analysis of the RCT indicated a sample size of 324 participants, which was also adequate for conducting the CFA. Nonetheless, the pre-enrolment screening revealed that a sufficient number of patients living with diabetes could not be recruited in the specific recruitment areas selected for the RCT [26]. Furthermore, logistical challenges impeded the prompt activation of the last two of the nine hospitals (Kabutare and Ruhango).

For these reasons an additional sample was also recruited for the purposes of the evaluation, following the inclusion criteria of the RCT: patients aged 21–80 years and diagnosed with DM at least six months prior to study onset. This supplementary cohort consisted of patients residing in additional zones in the catchment areas of same hospitals, except for the hospitals of Kibungo, Kibuye and Kinihira, where the number of patients was particularly low. The recruitment was carried out between June and December 2019. Exclusion criteria for both samples were illiteracy, severe hearing or visual impairments, severe mental health conditions and pregnancy, or in the post-partum period. The classification of DM type was based on the patients’ clinical records available at the hospitals. As the precise date of the diagnosis of DM was unknown for some participants, only those with at least one year of diagnosis were included so as to limit the effect of the emotional distress linked to recent diagnosis [18, 21]. Reliability analysis and known groups validity To assess internal reliability Cronbach’s α and composite reliability were calculated. Mean differences in total score and in the scales of the model with the closest fit were investigated across socio-demographic and clinical groups with Mann–Whitney U test. Effect sizes were calculated based on z values; r of 0.10, 0.30 and 0.50 were interpreted as small, medium and large effects respectively [29]. For continuous variables, Spearman’s correlation was used to determine which of them were associated with the total score and scales. Correlation coefficients below 0.4 were considered as weak, those between 0.4 and 0.7 as moderate, and those above 0.7 as strong [30, 31].

Construct and discriminant validity analysis To assess construct validity, a CFA was conducted. The five-factor model was fitted to the 39 items of the questionnaire with no cross-loadings or correlated errors terms. The estimation strategy, in line with the standard underlying assumptions of the CFA [32], is reliable, given the relatively different nature of the 5 dimensions (energy and mobility, diabetes control, social burden, anxiety and worry, and sexual functioning) used to group the 39 items under analysis.

The discriminant validity of D-39 was studied by calculating inter-factor correlations. Inter-factor correlations have been included in Table 5, supporting the assumption of a sufficient discriminant validity for all scales. The weighted least square mean and variance adjusted (WLSMV) estimator was used in the CFA. The root mean square error of approximation (RMSEA), the comparative fit index (CFI), and the Tucker Lewis index (TLI) were used to examine the approximate model fit. For RMSEA, values of less than 0.05 were indicative of a close fit and those between 0.05 and 0.08 were interpreted as adequate fit [27, 28]. The 90% confidence intervals of RMSEA were also evaluated, as they should be less than 0.05 for the lower bound and no worse than 0.08 for the upper one [27]. For CFI and TLI, values of 0.90 and above were regarded as acceptable fit [27, 28]. Hu’s and Bentler’s recommendation of raising such cut-offs to 0.95 was also taken into account [27, 33]. The relative χ2 was also calculated and a value of 2 or less was deemed adequate [27]. Finally, although the weighted root mean-square residual (WRMR) was computed and values of 1 or lower were considered a good fit, the experimental nature of this statistic thwarted drawing conclusions based on it [27, 28, 34].

Statistical analyses were performed using Stata version 16. Mplus version 7, and JASP version 16.2.

Ethical and research clearance

The study protocol was developed, and research authorisation was sought from the Rwandan Health and Education Ministries. Ethical approval was obtained from the Rwanda National Ethics Committee and the Ethics Review Panel of the University of Luxembourg.