Study design and participants
Study was a cluster randomised effectiveness-implementation hybrid trial type 2 with two parallel groups and its protocol has been previously published[30]. It is reported according to the Standards for Reporting Implementation Studies (StaRI) and Consolidated Standards of Reporting Trials (CONSORT) extension for cluster trials (Additional file 1).
EIRA study was conducted from January 2017 to December 2018 in 26 PHC centres of seven of the 17 Spanish Autonomous Communities. The Spanish health system is based on universal coverage with free access for all citizens. PHC includes health care, health education and prevention, health promotion and community care and is provided by multidisciplinary teams (physicians, nurses, paediatricians, social workers and dentists).
A PHC centre was eligible to participate if: 1) it had internet access; 2) it could implement community activities; 3) it was not located in areas with significant cultural and linguistic diversity or in tourist areas; and 4) if the management team was pro-actively engaged. All healthcare and administrative staff were invited to participate. The study targeted people aged 45 to 75 years, with at least two of the three following unhealthy behaviours: smoking, insufficient physical activity and low adherence to a Mediterranean dietary pattern. Exclusion criteria comprised advanced serious illness, cognitive impairment, dependence for basic everyday activities, severe mental illness, being in a long-term home health care program, undergoing treatment for cancer or end-of-life care, and planning to move from the area during the intervention.
Intervention
The intervention was based on the Transtheoretical Model and Stages of Change[29,31] and was built on the results of the previous phases of the EIRA study[11,20,29,21–28]. It had a maximum duration of 12 months and was carried out at the individual, group and community level in accordance to the stages of change and unhealthy behaviours (see Table 1). The intervention focused on all target behaviours, and together, participant and PHC professional developed priority actions on one or more of these behaviours.
The individual approach had an average intensity of 2-3 visits; the professionals could add extra visits when appropriate. Depending on the stages of change, the visit included: a) a very brief intervention to raise awareness of the need for MHBC and help with relapse prevention; b) a brief intervention to agree on a plan for MHBC. PHC professionals enhanced their motivational interviewing skills with a 20-hour online training, an in-person group feedback session and an acting patient session. In addition, PHC professionals and participants benefited from web-based tools such as http://proyectoeira.rediapp.org, personalised text messages, a mobile app[32] and other devices such as pedometers and smartwatches.
The group approach consisted of health education workshops which were delivered some weeks after initiating the individual approach and were conducted by healthcare professionals at the PHC centre. These workshops lasted 90-120 minutes, and their main purpose was to strengthen the advice discussed during the individual visits and to provide people with guidelines toward the practice of physical activity and the adoption of a healthy diet, for example through gym sessions, cooking workshops and seasonal menus.
The community approach focused mainly on social prescription of resources and activities offered in the communities of the participants. Previously, the PHC teams had identified the community health assets and selected the most relevant, their accessibility and possibility of referral of participants. These community activities included cooking courses, healthy eating workshops, local walking events, line dances and other physical activity programs.
Usual care
PHC professionals in the control group integrated the recommendations of the Program of Preventive Activities and Health Promotion[33], which incorporates preventive protocols with lifestyle recommendations and activities targeting specific age, sex and risk groups.
Implementation strategy
The implementation strategy was based on the Consolidated Framework for Implementation Research (CFIR)[34] and a set of discrete implementation strategies[35].
It was built on the findings of the previous phases of the EIRA study[11,20,29,21–28] and carried out in three stages (pre-implementation, implementation and post-implementation) (see Table 2).
Assignment of intervention
Participating PHC professionals signed a collaboration commitment to the study before the allocation of the intervention. The PHC centres were computer randomised for the intervention at a central location (IDIAPJGol, Barcelona, Spain). PHC professionals were aware of the study allocation. To minimise bias, an external unit independent of the PHC centre evaluated the intervention at baseline and at the end.
Evaluation
Intervention evaluation. The effectiveness of the intervention compared to usual care at 12 months post-intervention was measured by positive changes in:
- Smoking behaviour: self-reported continuous abstinence[36]. Positive change was defined as smoking at study entry and not smoking at the end of the study. We measured punctual and continuous abstinence at these two times.
- Physical activity behaviour: sufficient physical activity in previously insufficiently active people. The International Physical Activity Questionnaire was used[37]. Positive change was defined as having a low physical activity level at baseline and a moderate or high physical activity level at the end of the study.
- Dietary behaviour: adherence to a Mediterranean dietary pattern in people with low adherence at baseline. The 14-item Questionnaire of Mediterranean Diet Adherence (PREDIMED study) was used[38]. Positive change was defined as obtaining eight or fewer points at study entry and nine or more at the end of the study.
Statistical methods. A sample size of 3640 participants (1820 for each group), allowing for 30% loss to follow-up, was estimated to have 80% power (at 5% significance level, two tailed and with an intracluster correlation of 0.01[39]) to detect an absolute difference in a positive change in one or more of the three behaviours of 8% between groups (EIRA intervention and usual care).
A statistical analysis plan was established before data were available[30] All data were analysed on an intention-to-treat basis. We compared cluster and participant characteristics for all variables of interest by group allocation, using either means (standard deviations) or medians (interquartile ranges) for continuous variables, and numbers (percentages) for categorical variables. To address potential biases due to incomplete follow-up and for nonresponse in surveys, multiple imputation by chained equations (mice function in R software) with 50 imputed datasets were applied to outcomes and covariates[40–42] Estimates from each imputed dataset were combined following the rules outlined by Rubin[43]. We assumed that the variables with missing values were Missing At Random (MAR). The MAR assumption becomes more plausible by collecting more explanatory variables and including them in the analysis, and we included most possible explanatory variables (excluding duplicate variables, very similar variables and highly correlated variables to avoid collinearity)[40].
In order to analyse the effect of the intervention on each outcome measure, Odds Ratios (OR) and their 95% confidence intervals were computed by logistic regression models for clustered data, specifically generalised linear mixed models (using Stata function xtmelogit) with the PHC centre as a random-effects parameter. We analysed the variables associated with smoking cessation, the change in physical activity and in adherence to a Mediterranean dietary pattern, as well as the change in any behaviour and in two or three behaviours, adjusting for possible confounding variables. Final models were chosen in accordance with the study objectives, prior research[10,11] and the nature of the variables (potential confounders, significant and clinically relevant variables). We also calculated an overall impact factor of the intervention on the target population according to an expanded impact formula for MHBC proposed by Prochaska et al.[44]. We used Stata/SE v.15.1 (StataCorp, LP, TX) and SPSS 25.0 (SPSS Inc., Chicago, Illinois) for all analyses.
Implementation evaluation. We assessed implementation outcomes and the determinants of implementation success.
Implementation outcomes. The following implementation outcomes based on the evaluation framework proposed by Proctor et al.[45] were assessed:
Adoption. We calculated the proportion of PHC professionals who pre-implementation indicated their intention to implement the EIRA intervention.
Appropriateness and acceptability (early and final). Both were assessed on PHC professionals and participants. We designed two self-administered questionnaires, one for participants and one for PHC professionals. Two instruments were administered in the pre and the post-implementation stages. The definitions of implementation outcomes[45] constituted the conceptual model to define the items. A set of potentially relevant items was formulated. Questionnaires were pilot-tested in phase II of study. The final questionnaire for PHC professionals included eight items and the participants’ questionnaire included seven. All items in both questionnaires used an 11-point Likert scale with three semantic anchors. In the questionnaire for professionals, appropriateness and acceptability of the intervention were measured according to the type of unhealthy behaviour. In contrast, the items were more generic in the participants’ questionnaire. Additional file 2 includes a copy of both questionnaires. We analysed the structure of questionnaires and factorial analysis found two dimensions in both questionnaires. Goodness-of-fit indices suggested a good model fit in the professionals’ questionnaire (Root Mean Square Error of Approximation (RMSEA)=0.05 and Comparative Fit Index (CFI)=0.99) and adequate fit in the participants’ questionnaire (RMSEA=0.06 and CFI=0.99). Similarly, internal consistency in the scores of the two dimensions was good in both questionnaires (Cronbach’s alpha ≥0.80).
Feasibility. We calculated consent rate (% participants who consented among all invited to participate), recruitment rate (% of participants who were eligible, who accepted and attended the baseline assessment visit among all those invited to participate), intervention uptake rate (% of recruited participants who actually received the intervention) and completion rate (% of recruited participants who completed the study).
Fidelity of the planned intervention. The degree of compliance with planned activities for each intervention (see Table 1) was estimated by analysing the number and kind of activities recorded in the case report form by PHC professionals.
Determinants of implementation success. Ten focus group meetings moderated by an experienced researcher were conducted in the post-implementation stage. A total of 64 PHC professionals (average number per group = 8) from intervention centres participated. We were unable to perform this evaluation in two PHC centres. A structured interview guide based on CFIR constructs was used[46]. Group sessions were recorded and transcribed to create written documents for qualitative coding. All transcripts were reviewed by a group of members of the research team. A thematic content analysis and data coding was performed in accordance with CFIR constructs. Coding was deductive (codes derived from CFIR constructs) and inductive (codes derived from the data). Subsequently, researchers rated each CFIR construct for each PHC centre according to CFIR guidelines. Ratings ranged from -2 to +2, with 0 representing a neutral or mixed influence and M representing missing data. Two researchers independently coded and rated data of each PHC centre and wrote a memo report which was subsequently discussed with the whole team of analysts until an agreement was reached. During all this analysis, researchers were blinded to the intervention and implementation outcomes.
Spearman’s rank correlation coefficients were computed and used to assess the strength of association between construct ratings and fidelity of the planned intervention across PHC centres. Constructs with statistically significant correlations (P<0.05) with fidelity outcomes were believed to strongly distinguish PHC centres with low and high implementation success. Correlations values of rho ≥0.50, but with P values between 0.05 and 0.10 were considered weakly distinguishing.