Task shifting of frontline community health workers for cardiovascular risk reduction: design and rationale of a cluster randomised controlled trial (DISHA study) in India

DISHA is a cluster RCT conducted across 10 sites in India (Fig. 1 and Table 1). At each site, one district was selected based on convenience and from each district 12 clusters were randomly selected using computer generated random numbers (total = 120 clusters in ten sites). A cluster is defined as a small village (ideally including 1–3 Anganwadi Centres) with 250–300 households. The Anganwadi centre is part of the integrated child development service program of Government of India at the village level and typically provides basic health-care including contraceptive counselling and supply, nutrition education and supplementation, and pre-school activities. Over a million such centres are currently active in India. The identified 12 clusters were then randomly allocated to intervention and control clusters in a 1:1 allocation sequence using computer generated random numbers. Random allocation of the clusters was done by a statistician who is unaware of the study settings and cluster locations. If any of the intervention and control clusters were <10 km apart, one was dropped and replaced with another randomly selected cluster from the same district. The iteration continued till we managed to get 6 intervention and 6 control clusters that are at least 10 km apart in each site. The study was initiated in 5 sites in the first phase and in the second phase it has been extended to 5 more additional sites in India. The DISHA study included a baseline cross-sectional survey in all sites involving both the intervention and control clusters with equal representation. The study staff had no information on intervention allocation until the completion of the baseline survey. After completion of the baseline survey in all clusters, frontline health workers (mainly Anganwadi workers and ASHA workers) were trained to deliver CVD risk reduction interventions in the entire intervention clusters for a period of minimum 18 months. A post intervention survey is planned in a representative sample from the same clusters immediately after the intervention phase. The detailed study flow-chart is presented in Fig. 2.

A baseline risk factor survey was carried out initially in the 5 centres included in the first phase (2013–2014) using a structured questionnaire and it is currently on-going in the centres included in the second phase (2014–2015). The survey was conducted among participants from randomly identified 120–150 households in each cluster with a target to obtain a minimum of 300 participants. To account for non-response rate we randomly selected up to 10 % more households from the same cluster initially and in 5 instances up to 20 % more households. All adults over 18 years of age in the randomly selected households were eligible to be included in the survey. The survey questionnaire was translated into local languages (Hindi, Gujarati, Tamil, Oriya, Assamese and Telugu), back translated into English and pilot tested in 30–50 participants at each site (Additional file 1) for internal validity. The survey included assessments of demographic details, general health status, diet, physical activity, tobacco, and alcohol consumption (Table 2). The world health Organization (WHO) STEPS instrument for chronic risk factor surveillance has been modified and adapted to capture local, contextual information. Other components have been adapted from the General Health Questionnaire-12, Fagerstrom Test for Nicotine Dependence (FTND), and WHO AUDIT (Alcohol Use Disorders Identification Test). The general health questionnaire incorporated in the main questionnaire was a measure of current mental health or psychological wellbeing and it is extensively used in different settings and different cultures. A detailed dietary survey was conducted in a sub-sample of 100 random participants at each site. It included, a standardized, semi- quantitative food frequency questionnaire based assessments and two separate 24 h diet recalls within a week time.

The data entry staff under the supervision of site principal investigator completed data entry on a specially designed web-based online data application. The software has been developed on My SQL 7.0 server. The application uploaded the data directly on a server maintained at the Central Coordinating Centre (CCC), New Delhi. The data are collated, cleaned resolved and analysed at the CCC. We double checked the accuracy of data entry by matching the biochemical data received from the central laboratory with the data entry from the sites. If the data errors were more than 3 %, then we independently verified all other records. We found serious problems in data quality from one of the sites participating in the study and dropped that site from all analyses.

A package was designed using intervention mapping to guide development of each component of intervention. The following five main steps were involved in the process of intervention mapping; (1) creation of matrices indicating the proximal program objectives from performance objectives and determinants of behaviour and environmental conditions, (2) selection of theory-based intervention methods and practical strategies, (3) design and organising programs, (4) describe adoption and implementation plans, and (5) generate an evaluation plan [5].

In the first step, each component of the healthy lifestyle behaviour related to CVD risk reduction (diet, physical activity, tobacco, alcohol and adherence to treatment) was broken down into performance objectives. They were then prioritized based on practical feasibility of implementation, resource constraints and the local context followed by subjective selection of important and changeable determinants of the risk behaviour. We then framed a matrix of potential change objectives and the corresponding intervention options at individual, group and mass level for population level CVD risk reduction. These objectives were created at each level of intervention planning by crossing performance objectives with determinants and finally deriving the change objectives.

In the next step, the theoretical domains framework was used, as described by Mitchie et al. to guide the development of the potential interventions [6, 7]. A set of 12 domains covering the main factors influencing behaviour and behaviour change were identified: knowledge; skills; social/professional role and identity; beliefs about capabilities; beliefs about consequences; motivation and goals; memory, attention and decision processes; environmental context and resources; social influences; emotions; behavioural regulations; and nature of the behaviours. Using this theoretical framework the tools were designed such as the information booklet, calendar, posters, and leaflets.

The target community was actively involved and played an important role in the development of the intervention tools. The possible barriers and facilitators of a healthy lifestyle were identified through focus group discussions and key informant interviews with target community members, frontline health workers, primary care physicians and the respective site investigators. A proof was prepared initially for each tool (booklets, posters, leaflets, banners and table-top calendars), pilot tested to ensure contextual relevance, and then modified them in an iterative process based on inputs from the pilot test both with frontline health workers and target community members. The frequency of use and the specific target groups (for example; men, women, smokers, individuals with hypertension etc.) were also determined based on focus group discussions and pilot test inputs. The whole process was independently repeated at each site to account for differences in the local context, cultural issues, and language differences. The final print materials are available in the Additional file 2. In addition to the print materials, a calibrated salt spoon and an oil dispenser was also developed based on feedback from the focus group discussions to help each household to measure and quantify their daily consumption of salt and oil.

In the final step a structured intervention program was developed (Table 5). It includes individual level (for example; individual counselling), household level (for example; household visits by frontline health workers) and community level interventions (for example; display of posters, community level activities and competitions). The intervention program was focussed on; 1) healthy diets including low salt, low fats, low trans-fat and high fibre intake, 2) increasing physical activity of vigorous intensity for at least 30 to 60 min and 3) quitting of tobacco and alcohol. The developed tools were assigned to each task and the frequency of intervention was also prescribed in advance.

Similar to the baseline survey, an independent, representative, cross-sectional end-line survey is planned in both the intervention and control clusters. The end-line survey will follow the same methods as in the baseline survey. However in the end-line survey, a customised tablet computer application is developed for data collection in both the intervention and control clusters.

This multi-site trial is coordinated by a central coordinating team based at the Centre for Chronic Disease Control, New Delhi, India. The team consists of epidemiologists, biostatistician, nutritionists and behaviour communication experts. Each participating site also has at least two investigators, 6 field staff or field attendants recruited for data collection of baseline and post intervention survey, 1 data entry operator for entry of the questionnaires in the software and 1 lab technician for sample collection and processing. Similarly, each study site engaged 6–12 trained frontline health workers for intervention delivery.

The sample size was calculated with the aim to detect an epidemiologically significant difference of 2 mmHg in systolic blood pressure at the population level between the intervention and control arm of the study. This is based on the estimate that, a 2 mmHg decrease in mean blood pressure level at the population level will substantially reduce the incidence of hypertension and future cardiovascular events over a longer period of time [8] and the population approach proposed by Geoffrey Rose as an effective means of prevention in apparently healthy populations [9]. We assumed an intra-class correlation coefficient of 0.002 for blood pressure (based on previous cluster design survey data from India), a type 2 error of 5 % and 90 % power for the study. Based on the aforementioned assumptions and accounting for design effect (1+ [m-1] X ICC, where m is the average cluster size), the sample size was estimated as 1620 in each arm. Accounting for a 10 % drop-out rate, we decided to recruit 300 participants from each cluster, 1800 in each arm of the study and 3600 in each study site.

Summary of continuous variables will be presented as means and standard deviations and compared across intervention and control clusters. In case of skewed data, medians and inter-quartile range will be presented. All categorical variables will be presented as frequencies and percentages. Intra-class correlation coefficients for all important risk factors will be also presented.

After the completion of the study, analyses will be performed using cluster level summaries at each site. Equal weight will be given to each of the six clusters. All analysis will compare the two treatment arms, unless otherwise stated. Binary outcomes will be presented as a difference in proportions. Finally, a meta-analysis will be conducted to estimate the overall effect size.