The effect of a community-based health behaviour intervention on health-related quality of life in people with Type 2 diabetes in Nepal: a Cluster Randomized Controlled Trial

A prospective two-arm cluster randomized controlled trial (RCT) was conducted for a period of six months in two districts of Bagmati province of Nepal—Kavrepalanchowk and Nuwakot. A cluster randomized design was used to minimise contamination and the possibility of selection bias that could have occurred due to preference of the participants or health service providers. The study setting for this study comprised of 30 community health centres in rural and peri-urban areas of these districts. A detailed study protocol has been published elsewhere [11]. The study follows the Consolidated Standards of Reporting Trials (CONSORT) 2010 statement: extension to cluster randomised trials for reporting of its findings [12].

The Co-LID study was conducted among people aged 30–70 years with clinically diagnosed T2DM i.e., HbA1c > 6.5% or fasting plasma glucose (FPG) > 7.0 mmol/l based on the American Diabetes Association and WHO criteria for diagnosis and classification of diabetes [13, 14]. Those below the age of 30 years, clinically diagnosed with Type 1 diabetes mellitus, currently pregnant, or with cognitive impairment who were unable to respond the questions were excluded from the study.

A stratified random sample of 20 healthcare centres from Kavrepalanchowk and 10 from Nuwakot districts were selected. The unit of sampling or clusters for this study were health centres from the selected districts. A detailed sampling process is reported in the study protocol [11]. Out of 4748 people screened from 30 clusters in Nuwakot and Kavrepalanchowk, 619 had a clinical diagnosis of T2DM and were assessed for eligibility. Of them, 481 consented to participate in the study and were enrolled and randomised.

The calculated sample size in the overarching intervention was 414, obtained from 30 clusters in Kavrepalanchowk and Nuwakot districts, selected based on access and feasibility: 15 clusters randomly allocated in the intervention arm and 15 in the usual care (control) arm with an average of 12 participants in each cluster. The sample size in the main study was calculated with the aim to reduce the mean HbA1c levels from 7.5% at baseline to 7.0% at 6 months follow-up among the intervention participants compared with estimated no or minimal changes in participants of usual care arm (HbA1c levels 7.5%), with at least 80% power and at 95% confidence interval assuming an intra-class correlation coefficient (ICC) of 0.050 as noted in a previous study conducted in an Asian setting [15] and an estimated HbA1c mean standard deviation (SD) of 1.3 as reported in previous studies that used diabetes self-management programs [16]. With the adjustment of possible 15% attrition rate from baseline to 6-month follow up, a total of 414 samples were calculated. Hence, the minimum sample size required for this study was 414. All trial participants (n = 481) were included in this study.

The health centre (as a cluster) was considered as a unit of sample for randomization. Thirty health centres were randomly allocated as intervention or usual care clustered groups with the sample allocation ratio of 1:1 using simple randomization technique by a statistician not involved in the study. Due to the nature of the study, neither participants nor those who delivered the interventions were blinded. However, the statistician was blinded to group assignment.

The primary aim of the lifestyle intervention was to achieve reduction in HbA1c level among people with T2DM in the intervention arm at 6-month follow-up compared to no possible change or increment of HbA1c level in people with T2DM in the usual care arm. To ensure participants’ homogeneity, all participants received a two-hour education session on diabetes self-management prior to randomization. After randomization, the intervention group received a culturally tailored and locally based group intervention program consisting of fortnightly sessions that covered 12 modules over six months, accompanied by ongoing support. These modules, each lasting for 60 min, included information on diabetes literacy, diabetes medications, physical activity, dietary modification, management of stress and depression, smoking cessation and reducing harmful alcohol consumption, monitoring blood sugar level, foot care, oral health, quality of life, health care utilization, and social and emotional support (Supplementary File S1). Further, group-based sessions on local practices such as cooking, group exercise, yoga, were conducted to complement the primary intervention modules and enhance participants’ locally acquired skills related to meal preparation and group exercises. Throughout the delivery of the intervention, culturally adapted practices were adopted. This included culturally tailored dietary education respecting traditional eating habits and local recipes; meal preparation using local produces; behavioural education on traditional physical activities such as farming and household chores; culturally relevant approaches such as use of local language, narratives, metaphors, storytelling and knowledge sharing; provision of pictorial book on diabetes management and prevention of complication in Nepali language; empathetic listening and counselling; and family and social support. Furthermore, the intervention program was delivered by trained community health workers (CHWs) and peer supporters, who later facilitated the group-based intervention to the intervention group participants. CHWs are the frontline healthcare workers in government health facilities in Nepal [17] and peer supporters are participants from the intervention group who are committed to managing their own health behaviours while also assisting others. CHWs are the primary point of contact for health-related issues among the local population; therefore, the Training of Trainers model was adopted for local capacity building of CHWs in the two districts (Health Assistant, Auxiliary Nurse Mid-wife or Auxiliary Health Worker) in delivering the intervention sessions. This also helped foster the cultural appropriateness of the intervention. A comprehensive training guide for CHWs and peer supporters is provided in the appendix (Supplementary File S2). Additionally, to ensure regular contact between CHWs and participants and maintenance of self-care behaviours learnt in the sessions, CHWs conducted fortnightly telephone calls for the first three months and monthly calls for the rest of the months, along with online pictorial and audio-visual messages on lifestyle intervention in Nepali language sent to the participants’ phones. The intervention group also received locally available care, which was standard diabetes management and care available at the respective health centres, in accordance with the standard practice and guidelines.

A description of the intervention modules is attached in the appendix (Supplementary File S1) and also published in the study protocol [11]. On the other hand, the participants in the control group received the usual care with no lifestyle intervention by CHWs and the peer supporters. However, they received a pictorial brochure on diabetes education in Nepali language so that they were not deprived of awareness. Follow-up was conducted for the participants of both intervention and usual care groups at six months.

A structured questionnaire was prepared to collect socio-demographic and diabetes-related information from study participants. Health-related quality of life was measured using a Nepali-translated version of EuroQol 5D (EQ5D) tool, which consists of 2 parts—the EQ5D-3L descriptive system and the EuroQol Visual Analogue Scale (EQVAS) [18]. The tool showed a good reliability in this study with a Cronbach’s alpha of 0.75. The descriptive system has 5 dimensions (mobility, self‑care, usual activities, pain/discomfort, and anxiety/depression) and each dimension has 3 response levels (no problems, moderate problems, and extreme problems). The participants’ level of reported problems on each dimension were combined to create a 5-digit number for each participant, known as their health state. EQ5D-3L generates 243 (3⁵) various health states, with 11,111 being the best health state (full health) and 33,333 being the worst health state. If a country has a reference value set, then the health states can be converted to a single utility value (the EQ5D index score). Since there is no EQ5D reference value set for Nepal, the 3L utilities were generated using the reverse crosswalk index value calculator [19] with 5L value sets for a neighbouring country, India, for the purpose of this study. Similarly, an EQVAS tool, a 20 cm wide 100-point scale, was used to allow respondents to rate their health on that day from the worst imaginable health state (0) to the best imaginable health state (100). Measurements of covariates along with the tools used and their psychometric properties are presented in the appendix (Supplementary Table S3). The survey questionnaire was translated into Nepali language to best fit in Nepalese context and pre-tested prior to baseline data collection. The questionnaire was then back translated to English by two independent translators.

The primary outcomes of this study were changes in the EQ5D-3L score and EQVAS score in the intervention group compared to usual care group at the end of six-month intervention against the baseline status. The outcomes were measured at the individual participant level.

The baseline data were analyzed using descriptive summaries such as frequencies and percentages for categorical variables, means and standard deviations for normally distributed continuous variables, and medians and interquartile ranges (IQR) for non-normally distributed continuous variable. The difference between the intervention and usual care groups in terms of socio-demographic and diabetes-related variables were measured at baseline using standardized mean differences (SMDs). An SMD of less than 0.1 was considered to indicate negligible difference [20].

The outcome variables of this study were EQ5D-3L score and EQVAS score. We assessed the outcome of this study using Intention-to-Treat (ITT) analysis in accordance with the study protocol. We handled missing data due to loss of follow-up using a multiple imputation (MI) technique (see below).

First, we computed the proportion of outcomes across various groups, including baseline, follow-up, intervention, and control; and compared the proportions across the groups using a t-test. Then, we employed a difference-in-difference (DID) regression model to assess the effect of the intervention on the study outcomes. This model considered baseline differences in outcomes and variations between the intervention and control groups in terms of the distribution of covariates.

We used a generalized estimating equation (GEE) model to account for within-cluster correlation. First, we fitted the unadjusted model by considering only the time and group variables in the model and got the unadjusted intervention effect. The model was then extended further by adjusting for variables that differed significantly (p-value < 0.200) at baseline between intervention and control group to get the adjusted intervention effect. The effect estimates were reported as coefficients. A p-value < 0.05 was considered statistically significant.

Furthermore, to explore the association between the number of intervention sessions attended and HRQOL among participants who received interventions, a subset of the intervention group was analysed. Both adjusted and unadjusted GEE models with a binary logit function were constructed for this association analysis, where EQ5D-3L and EQVAS served as outcome variables and the number of sessions attended was the primary exposure of interest.

The statistical analyses originally proposed in the trial registration included a random effects regression model and analysis of covariance (ANCOVA). However, we opted for a DID GEE regression model in the final analysis because a) our primary interest was population-averaged effects, which is estimated by GEE, rather than cluster-specific effects, estimated by random effects regression model; b) GEE more effectively accounts for within-subject correlations than ANCOVA, given the longitudinal nature of our data; c) DID controls for baseline differences between groups accounting for differential trends thereby enhancing causal inference; d) GEE estimates robust standard errors ensuring valid inferences even when the true correlation structure is mis-specified versus random effects model, which assumes that random effects (e.g., subject- or cluster-specific variations) follow a normal distribution; and e) GEE addresses missing data providing more robust estimates under a missing at random assumption compared to the random effects model.

All statistical analyses were performed using Stata, version 18.

The mean age of participants at baseline was 54.4 years (SD 9.4). There were no differences in baseline characteristics of participants between the intervention and control groups, as shown in Table 1.

No difference in the EQ5D-3L score was observed within both intervention and control groups (Table 2). The EQVAS score improved slightly at six-month follow-up in the intervention group (baseline: 72.15, 95%CI: 70.20–74.10 vs. follow up: 72.35, 95%CI: 70.15–74.54, p-value 0.89), however this improvement was not statistically significant. A statistically significant (p < 0.03) reduction in EQVAS was observed in the control group (baseline: 71.21, 95%CI: 69.17–73.25 vs. follow up: 68.02, 95%CI: 65.96–70.07). The difference in change between the groups for EQ5D-3L was 0.016 (95%CI: -0.03, 0.07), favouring the intervention group. Similarly, a difference in change between groups of 3.39 (95% CI: − 0.73, 7.50) was observed for EQVAS score, also in favour of the intervention group.

The effect of intervention on EQ5D-3L and EQVAS using a GEE model is presented in \* MERGEFORMAT Table 3. A positive but non-significant intervention effect was seen in the EQ5D-3L index score across the study population in the adjusted model (β₁ = 0.052, 95%CI: − 0.04, 0.14). The intervention had a positive and statistically significant effect on EQVAS score (β₁ = 3.61, 95%CI: 0.05, 7.17) in the adjusted model.

Nearly 55% of participants (n = 130) attended at least seven intervention sessions out of a total of 12. Only 16% (n = 37) attended all 12 intervention sessions while 9.6% (n = 23) did not attend any of the session (Supplementary Table S4). The average number of sessions (SD) attended was seven (SD, 4.19). The predicted EQVAS score declined slightly when intervention group participants attended only one session out of the twelve sessions compared to when they had not attended any session. However, with every additional number of sessions attended, an increase in their EQVAS score was predicted Fig. 2.

Similarly, a slight decline was observed in the predicted EQ5D-3L index score when intervention group participants attended only one session. However, an increase in the index score was predicted for every additional session attended after that as shown in Fig. 3.

Our adjusted GEE model showed a positive and statistically significant increase in EQVAS score by 0.5 per unit session attended by the study participants (β₁ = 0.49, 95% CI: 0.01, 0.98). Further, the EQVAS score of those who attended seven or more intervention sessions was higher by four points (β₁ = 3.86, 95% CI: -0.13, 7.85) than of those who attended six or less sessions, although this difference was not statistically significant (p = 0.06) (Supplementary Table S5). However, we did not find any statistically significant effect of number of sessions attended on the EQ5D-3L index score (β₁ = 0.005, 95% CI: − 0.001, 0.01) ( \* MERGEFORMAT Table 3).

Further subgroup analysis suggested a negative association of HbA1c with EQVAS (β₁ = − 0.31, 95% CI: − 1.67, 1.05) as well as with EQ5D-3L index score (β₁ = − 0.012, 95% CI: − 0.02, 0.003) across the intervention group population suggesting that a decline in HbA1c level was associated with an improvement in HRQOL. However, these associations were not of statistical significance (results not shown in the table).