This paper builds on the findings from the ODHIN trial, which was focused on identifying implementation strategies for implementing brief interventions . Therefore, the methods section presents the main key ingredients of the ODHIN study. More information can be found in another ODHIN paper that presents the trial effects on screening and brief intervention activity. CONSORT guidelines were followed in reporting the trial .
Study design and participants
ODHIN was a cluster randomised 2 × 2 × 2 factorial trial as described in the study protocol (ClinicalTrials.gov. Trial identifier: NCT01501552) . English, Catalan, Polish, Swedish and Dutch PHCUs participated and combined their data to examine the effect of three different implementation strategies (training and support, financial reimbursement and referral opportunities to an internet-based brief intervention programme) on screening and brief intervention activity for risky drinkers identified by screening using the AUDIT-C questionnaire screening tool .
After formal agreement of the PHCUs to take part in the trial, a four-week baseline measurement took place, during which no trial interventions were administered. After a 2- to 6-week gap, the 12-week implementation period occurred with the start date for each country between November 2012 and May 2013. All eight allocation groups received the same input as controls but with additional components added.
Control group: care as usual
Training and support (TS)
Financial reimbursement (FR)
Referral to internet-based brief interventions (e-BI)
TS and FR
TS and e-BI
FR and e-BI
TS, FR and e-BI
More details about the implementation strategies and procedural activities were described in Additional file 1 [19, 21].
Brief intervention proportions
Brief intervention proportions were the primary outcome of the ODHIN study. These were measured during two time frames: during the four-week baseline period and during the 12-week implementation period. The providers completed paper tally sheets (Catalonia applied electronic tally sheets). The tally sheets included AUDIT-C questions, AUDIT-C scores and tick boxes to indicate the type of intervention (oral advice, an advice leaflet, referral to the e-BI programme, or referral for advice to another provider in or outside the PHCU) that was delivered.
The brief intervention proportions were calculated as a number of AUDIT-C positive patients that received one or more of oral advice, an advice leaflet, referral to the e-BI programme, or referral for advice to another provider in or outside the PHCUs, divided by the total number of adult (≥18 years) consultations of the participating providers per PHCU. We added up all the patient consultations in the 12-week implementation period and calculated accordingly the proportion of AUDIT-C positive patients that received a brief intervention. Impact of the implementation strategies on screening and brief advice activity was reported elsewhere .
Role security and therapeutic commitment
Before starting the baseline measurement of brief interventions, providers completed a questionnaire in which they provided their demographical features, including gender, age and occupation (e.g. medical practitioner, nurse, practice assistant).
Role security and therapeutic commitment were measured at baseline by the short version of the Alcohol and Alcohol Problems Perception Questionnaire (SAAPPQ), translated into the native language of each participating country . The SAAPPQ is a validated instrument based on factor analysis  of the original tool as developed and validated by Cartwright . The shortened version of the original attitude scale has been demonstrated to be as representative and discriminatory as the full AAPPQ scale . The SAAPPQ has good test-retest reliability and Cronbach’s alpha in the range of 0.7 and 0.9 . Furthermore, it has been applied in several international contexts and with various disciplines (e.g. [24–27]). All participating providers who provided written informed consent were asked to complete the SAAPPQ. Role security measures role adequacy, for example “I feel I can appropriately advise my patients about drinking and its effects”, and role legitimacy, for example, “I feel I have the right to ask patients questions about their drinking when necessary”. Role security is expressed at the emotional level, whereas therapeutic commitment measures motivation, for example “pessimism is the most realistic attitude to take toward drinkers”; task specific self-esteem, for example “all in all I am inclined to feel I am a failure with drinkers”; and work satisfaction, for example “in general, it is rewarding to work with drinkers”. Role security includes four items on a seven-point Likert scale and summed scores range between 4 and 28. Therapeutic commitment includes six items on a seven-point Likert scale and summed scores range between 6 and 42.
Sample size and randomisation
To achieve sufficient statistical power for significant effects on intervention proportions, it was estimated that 120 PHCUs (15 per eight allocation groups, evenly distributed between countries) would be needed .
Randomisation took place after formal agreement of the PHCUs to take part in the trial. The PHCUs were randomly allocated to one of the eight allocated groups by the ODHIN coordinating centre, using computerised randomisation, stratified by country (i.e. block randomisation), ensuring 15 PHCUs per group (three per group in each country) .
Because of the hierarchical structure (providers nested within PHCU, nested within country), we performed a two-level linear multilevel analysis (mixed model). We performed a model with random intercept for countries and practices, and other variables such as TS, FR, e-BI, and baseline brief intervention proportion fixed. The outcome measure was brief intervention proportion after the implementation period. Multiple imputation was not applied as the percentage of missing cases was 1.5 % .
When examining the impact of baseline role security and therapeutic commitment on the 12-week brief intervention proportions, examination of residuals found them to be not symmetrically distributed around zero, so the data underwent log transformation, which provided a better fit. Prior to log transformation, proportions with a value of zero were assigned a value of 0.001. Coefficients for the combined effects of TS + FR and TS + e-BI were the sum of the individual coefficients. Since the data were log transformed, the contrast coefficients are relative effects. The percentage difference in brief intervention proportions with each implementation strategy as opposed to without, was calculated with the equation: difference (%) = 100 × (exp)2 × coefficient estimate from procedure MIXED minus 1. As this approach creates some issues with outcomes with a zero value, 0.001 was added to each proportion. In order to test the validity of this assumption, a sensitivity analysis was undertaken using the exact proportions but excluding those PHCU with an outcome of zero.
Correlations between the eight variables were tested with Pearson’s correlation test. To test the influence of baseline role security and therapeutic commitment on the delivered brief interventions, the model was run with both provider characteristics, included one by one. When both characteristics significantly correlated, which was expected as they were included in the same questionnaire, they were paired when entered included in the statistical model. Furthermore, for both role security and therapeutic commitment, we added interaction terms in order to identify interactive effects of characteristics with the implementation strategies. We considered a p value <0.05 statistically significant. In case of interaction, subgroups of variables were analysed separately. The statistical analyses were performed with IBM SPSS v20. The datasets are available upon request.