The cost-effectiveness of tailored, postal feedback on general practitioners’ prescribing of pharmacotherapies for alcohol dependence

Drug and Alcohol Dependence 124 (2012) 207–215

Contents lists available at SciVerse ScienceDirect

Drug and Alcohol Dependence journal homepage: www.elsevier.com/locate/drugalcdep

The cost-effectiveness of tailored, postal feedback on general practitioners’ prescribing of pharmacotherapies for alcohol dependence夽,夽夽 Héctor José Navarro a,∗ , Anthony Shakeshaft a , Christopher M. Doran b , Dennis J. Petrie c a b c

National Drug and Alcohol Research Centre, University of New South Wales, Building R3, 22-32 King Street, Randwick Campus, Sydney, NSW 2031, Australia Priority Research Centre for Health Behaviour, University of Newcastle, DMB Room 230K, King and Watt Streets, Newcastle, NSW 2300, Australia University of Dundee, Economic Studies, Nethergate, Dundee DD1 4HN, Scotland, United Kingdom

a r t i c l e

i n f o

Article history: Received 22 October 2011 Received in revised form 13 January 2012 Accepted 13 January 2012 Available online 22 February 2012 Keywords: Postal tailored feedback Alcohol dependence General practitioner Cost-effectiveness

a b s t r a c t Aims: The aims of this study were to conduct a randomised controlled trial to evaluate the costeffectiveness of tailored, postal feedback on general practitioners’ (GPs) prescribing of acamprosate and naltrexone for alcohol dependence relative to current practice and its impact on alcohol dependence morbidity. Methods: Rural communities in New South Wales, Australia, were randomised into experimental (N = 10) and control (N = 10) communities. Tailored feedback on their prescribing of alcohol pharmacotherapies was mailed to GPs from the experimental communities (N = 115). Segmented regression analysis was used to examine within and between group changes in prescribing and alcohol dependence hospitalisation rates compared to the control communities. Incremental cost-effectiveness ratios (ICERs) were estimated per additional prescription of pharmacotherapies and per alcohol dependence hospitalisation(s) averted. Results: Post-intervention changes, relative to the control communities, in GPs’ prescribing rate trends in the experimental communities significantly increased for acamprosate (ˇ = 0.24, 95% CI: 0.13–0.35, p < 0.001), and significantly decreased for naltrexone (ˇ = −0.12, 95% CI: −0.17 to −0.06) per quarter. Quarterly hospitalisation trend rates for alcohol dependence, as principal diagnosis, significantly decreased (ˇ = −0.07, 95% CI: −0.13 to −0.01, p < 0.05), compared to control communities. The median ICER per quarterly hospitalisation(s) averted due to intervention was Dominant (Dominant – $12,750). Conclusion: Postal, tailored feedback to GPs on their prescribing of acamprosate and naltrexone for alcohol dependence was a cost-effective intervention, in rural communities of NSW, to increase the overall prescribing of pharmacotherapies with a plausible effect on incidence reduction of hospitalisations for alcohol dependence as principal diagnosis. © 2012 Elsevier Ireland Ltd. All rights reserved.

1. Introduction Alcohol dependence is a leading cause of disability and a key contributor to the global burden of disease (Samokhvalov et al., 2010; World Health Organization, 2009). In Australia, the prevalence of alcohol dependence is estimated at approximately 3.5% (Hall et al., 1999; Proudfoot and Teesson, 2002). Treatment for alcohol dependence typically comprises a period of detoxification, usually with sedative medications to alleviate and prevent

夽 Supplementary material can be found by accessing the online version of this paper. Please see Appendix A for more information. 夽夽 Trial registration: ANZCTR, http://anzctr.org.au, ACTRN12610000109000. ∗ Corresponding author at: National Drug and Alcohol Research Centre, University of New South Wales, Sydney, NSW 2052, Australia. Tel.: +61 2 9385 0333; fax: +61 2 9385 0222. E-mail addresses: [email protected], [email protected] (H.J. Navarro). 0376-8716/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.drugalcdep.2012.01.007

symptoms of alcohol withdrawal (McKeon et al., 2008), followed by pharmacotherapies to prevent relapse to heavy drinking and to support alcohol abstinence and psychosocial interventions (Haber et al., 2009; Kiefer and Wiedemann, 2004; Rösner et al., 2010a, 2008; Snyder and Bowers, 2008). Current use of pharmacotherapies is limited for a number of reasons: only five are available internationally (Haber et al., 2009; Soyka and Rösner, 2010); evidence for their effectiveness is mixed (Anton et al., 2006; Boothby and Doering, 2005; Bouza et al., 2004; Kranzler and Van Kirk, 2001; Snyder and Bowers, 2008); they have different side-effect profiles for different types of patients (Department of Health and Ageing, 2007, 2009; Haber et al., 2009; Rösner et al., 2010a,b); less than 18% of dependent drinkers seek specialist care (Proudfoot and Teesson, 2002); and addiction specialists, who are most knowledgeable about pharmacotherapies are not readily accessible to patients, especially outside urban areas (Australian Institute of Health and Welfare, 2005; Druss and von Esenwein, 2006; McAvoy, 2008). In Australia, acamprosate and

208

H.J. Navarro et al. / Drug and Alcohol Dependence 124 (2012) 207–215

naltrexone are the two most common types of pharmacotherapies used to treat alcohol dependence due to their listing in the Pharmaceutical Benefits Scheme (PBS) and the Repatriation Pharmaceutical Benefits Scheme (RPBS) (Department of Health and Ageing, 2011a,b). In spite of the subsidised cost, the use of these medications is low, stemming predominantly from low rates of prescribing by GPs (Doran et al., 2003; Haber et al., 2009). For example, of the approximately 70–80% of dependent drinkers who visit their GPs each year in Australia (Proudfoot and Teesson, 2009), only 3% are prescribed a pharmacotherapy (Doran et al., 2003). Strategies to encourage higher rates of GPs’ prescribing, such as written prescribing feedback or feedback in combination with education or clinical guidelines, have yielded mixed results in randomised controlled trials (RCTs) for medications other than alcohol pharmacotherapies (Anderson et al., 1996; Gehlbach et al., 1984; Hux et al., 1999; Lagerløv et al., 2000; Naughton et al., 2009; Nilsson et al., 2001; O’Connell et al., 1999; Schectman et al., 1995; Søndergaard et al., 2002, 2003; Veninga et al., 1999). Indeed, the potential effectiveness and cost-effectiveness of such strategies specific to pharmacotherapies for alcohol dependence is unknown. In addition to improved patient outcomes, increasing rates of alcohol abstinence could also reduce demand for relatively expensive health-care services by reducing hospitalisations for alcohol dependence (Poikolainen et al., 2011). A community-level RCT in Australia, the Alcohol Action in Rural Communities (AARC) project, provided an opportunity to quantify the effect of a primary care intervention for alcohol dependence on both GPs prescribing behaviour and subsequent demand for inpatient hospital services. This study has two specific aims. First, to evaluate the cost-effectiveness of tailored, postal feedback on GPs’ prescribing of acamprosate and naltrexone for alcohol dependence, relative to current practice. Second, to examine the impact of any change in prescribing behaviour of pharmacotherapies on hospitalisations for alcohol dependence. 2. Methods

survey response rate [3017/7580]) from all 20 communities (Breen et al., 2010). The sample consisted of 18 to 62-year olds and, to optimise representativeness, was selected using the age and gender distribution of these communities according to the ABS 2001 Census of Population and Housing (Australian Bureau of Statistics, 2009a): age 18 coincides with the minimum age for voting and legal drinking in Australia, those over 62 contribute relatively little to alcohol-related harm (Breen et al., 2010; Shakeshaft et al., 2002). This survey included the Alcohol Use Disorders Identification Test (AUDIT), a 10-item standardised questionnaire with evidence for its reliability and validity in both clinical (Babor et al., 2001; Reinert and Allen, 2007) and population (Fleming, 1996; Ivis et al., 2000; Selin, 2003) samples. Respondents with an AUDIT score ≥20 were classified as dependent drinkers (Babor et al., 2001; Department of Veterans’ Affairs). To estimate the number of dependent drinkers in the experimental (N = 10) and control communities (N = 10), the respective proportions of survey respondents in the dependent drinking category were multiplied by population data separately for the experimental and control communities, by age and gender, obtained from the 2006 ABS census (Australian Bureau of Statistics, 2009b). 2.3.2. Number of acamprosate and naltrexone prescriptions filled. Prescribing data for acamprosate and naltrexone for the 20 AARC communities was obtained for the time periods 1 October 2000 to 31 December 2004, and 1 October 2005 to 31 December 2009. These data represent claims processed from approved pharmacies, rather than the number of prescriptions actually written by GPs. As conditions on data access, Medicare Australia provided de-identified data aggregated by quarters, separately for experimental and control communities, and would not provide data for the first three quarters of 2005 to limit the time period for continuous data to ensure an individual’s privacy was not breached (Medicare Australia, 2010). 2.3.3. Number of GPs in communities and prescribing rates. The number of GPs was compiled using information obtained through the Divisions of General Practice in NSW and cross-checking the electronic telephone directory for each community. Prescribing rates were converted to a rate per 10 GPs (Fig. 1), because even though the number of GPs in both the experimental and control communities increased by less than 2.5% in the study period (2000–2009), there were fewer GPs in the experimental (N = 115) than control (N = 160) communities. 2.3.4. Alcohol dependence hospitalisations. De-identified unit record data were obtained for all patients with a principal diagnosis of alcohol dependence (International Statistical Classification of Diseases and Related Health Problems, 10th Revision [ICD-10] code F10.2; World Health Organization, 2007), who were admitted to a hospital in one of the 20 AARC communities, between 1 October 2000 and December 31 2009 from the NSW Department of Health’s Admitted Patient Data Collection (including postcode of patient’s residence, length of stay, month and year of discharge). Rates of hospitalisations per 10,000 population, aggregated by quarters, are presented in Fig. 2 separately for experimental and control communities.

2.1. Study sample 2.4. Intervention Communities in New South Wales (NSW), Australia, were invited to participate in AARC if they: had an urban-centre locality population between 5000 and 20,000 (N = 27 communities; Australian Bureau of Statistics, 2009a); were at least 100 kilometres (km) away from a major urban centre, defined as a population of at least 100,000 (N = 24 communities); and were not known to be currently involved in any other large scale project aimed to assess or reduce alcohol-related harm (N = 20 communities). Communities of this population size were selected to ensure they were large enough to have a sufficient number of health resources, such as GPs and hospitals, and a sufficiently high number of alcohol dependent patients to reasonably be able to detect any post-intervention changes as statistically significant. 2.2. Study design This study is a prospective, matched pairs RCT, with whole communities as the unit of randomisation and analyses. Given evidence of disproportionately high levels of alcohol-related harm among males (Stockwell et al., 2002), young people (Stockwell et al., 2002) and in Indigenous communities (Gray, 2000), the proportions of males, people aged 15–24 and Aboriginal and Torres Strait Islanders was obtained for each of the 20 communities, using the Australian Bureau of Statistics (ABS) 2001 Census of Population and Housing data (Australian Bureau of Statistics, 2009a). Since the proportion of males and people aged 15–24 was similar across all communities, communities were ranked, in decreasing order, according to the percentage of the population defined as Aboriginal or Torres Strait Islander and contiguous communities provisionally classified as matched pairs. Each pair was checked to ensure that they were at least 100 km apart, to minimise the risk of cross-contamination of any intervention effects between potential experimental and control communities. One community within each pair was then randomly allocated to the experimental group. 2.3. Data sources 2.3.1. Estimating the number of dependent drinkers in communities to feedback to GPs. In 2005 AARC conducted a community survey, resulting in 3017 responses (40%

The intervention consisted of tailored feedback, in the form of a letter written by the lead AARC researchers, mailed in early September 2006 to all GPs in the experimental communities. Letters were printed on A4 paper with the front page containing a logo of the AARC project and an introductory text. The letter provided clear and concise feedback to each GP, based on calculations from the 2005 AARC survey data, the prescribing data (from October–December 2000 to October–December 2004) and the literature, on: the proportion of his/her community that is alcohol dependent; the percentage of dependent drinkers in Australia likely to be using either acamprosate or naltrexone (Doran et al., 2003); the percentage of dependent drinkers most likely to be using either acamprosate or naltrexone in the experimental communities; information supporting GP’s role in treating dependent drinkers with pharmacotherapies and behavioural interventions for relapse prevention (Anton et al., 2006); the availability of a subsidised cost for acamprosate and naltrexone; and, based on evidence of their underutilisation in Australia (Doran et al., 2003), the recommendation that GPs could increase the prescribing rates of either acamprosate or naltrexone as a means of reducing heavy alcohol consumption and harms for their alcohol dependent patients. 2.5. Costs 2.5.1. Tailored letter. Total direct costs in 2006 Australian dollars (AUD) for generating and mailing tailored letters to GPs were obtained as part of the AARC project and are included in the analysis. 2.5.2. Providing GP services for dependent drinkers in communities. The number of initial consultations and follow-ups for this analysis was estimated based on the number of prescriptions filled and recommended treatment durations. The most common recommended treatment is six months for acamprosate (Department of Health and Ageing, 2007; Haber et al., 2009; Rösner et al., 2010a) and three months for naltrexone (Department of Health and Ageing, 2009; Haber et al., 2009; Rösner et al., 2010b). Since GPs are limited to prescribing one month’s supply plus one repeat, full recommended treatment would require one initial consultation and two

H.J. Navarro et al. / Drug and Alcohol Dependence 124 (2012) 207–215

209

Fig. 1. Trends in mean prescribing rates/10 GPs of acamprosate and naltrexone in experimental and control communities (shaded area indicates data combined from multiple imputation).

follow-ups for acamprosate, and one initial consultation and one follow-up for naltrexone. For both pharmacotherapies, it was assumed that 50% of patients completed treatment (Doran et al., 2003) and the remaining patients were allocated one initial consultation. In Australia, GPs are reimbursed by the Commonwealth Government according to the complexity or both time and complexity of the consultation. Providing an initial consultation to a dependent drinker would allow GPs to claim a level C consultation lasting 20–40 min (Shanahan et al., 2005; e.g., AUD$60.95 in 2006/2007, Department of Health and Ageing, 2006), which permits a detailed history and physical examination, arranging necessary tests and implementing a treatment plan. A follow-up consultation for a dependent drinker would allow GPs to claim a level B consultation lasting less than 20 min (e.g., AUD$32.10 in 2006/2007), which permits reviewing the patient’s progress (Department of Health and Ageing, 2006). Costs for initial and follow-up consultations were estimated in AUD according to fiscal year-specific fees. Since the number of consultations is based on the number of prescriptions claimed by pharmacists, costs of any previous consultations were not considered (e.g., to request laboratory tests prior to initiating treatment).

2.5.3. Prescriptions in the AARC communities. Costs for acamprosate and naltrexone prescriptions, based on the maximum monthly quantity dispensed per pharmacotherapy correspondingly (for acamprosate 180 × 333 mg tablets for patients with a body weight of 60 kg or more (Department of Health and Ageing, 2011a) and for naltrexone 30 × 50 mg tablets (Department of Health and Ageing, 2011b)),

were supplied by Medicare Australia in AUD according to fiscal year-specific contributions by the government (e.g., AUD$152.52 for acamprosate, AUD$149.89 for naltrexone, in 2006–2007), and by the patient (e.g., AUD$12.70 for both pharmacotherapies in 2006–2007). 2.5.4. Hospitalisations for alcohol dependence. The average hospitalisation costs per patient for alcohol dependence (e.g., AUD$3472, in 2006–2007) were estimated using fiscal year-specific NSW cost weights and the Australian Refined Diagnosis Related Groups (AR-DRGs) patient classification system (code V62A; Commonwealth of Australia, 2007). This system adopts ICD-10 codes, linking number and types of patients treated in a hospital with the respective hospital resources required for each condition. 2.6. Statistical analyses Baseline characteristics for experimental and control communities were compared. Data analysis, for prescribing and hospitalisation rates, covered 37 quarters (nine years): 24 quarters pre-intervention and 13 quarters post-intervention. For the prescribing data that Medicare Australia would not provide (less than 8.2% of longitudinal sample), values were estimated using Rubin’s multiple imputation method (MIM; Little and Rubin, 1987; Rubin, 1987; Schafer, 1999). This Monte Carlo technique requires three steps: imputation, completed-data analysis and pooling of the results (Schafer, 1999; StataCorp, 2009a; Wayman, 2003). Each missing value was

Fig. 2. Trends in mean alcohol dependence hospitalisation rates/10,000 population in experimental and control communities.

210

H.J. Navarro et al. / Drug and Alcohol Dependence 124 (2012) 207–215

Table 1 Selected demographic characteristics of experimental and control communities. Communities Experimental N (%) Population 18–62 yrsa Male population 18–62 yrsa Female population 18–62 yrsa Potential dependent drinkers 18–62 yrsb,e Potential male dependent drinkers 18–62 yrsb,e Potential female dependent drinkers 18–62 yrsb,e GPsc Mean rate per dependent drinker 18–62 yrs per quarterd Acamprosate Naltrexone a b c d e

79,177 (100%) 39,621 (50.04%) 39,556 (49.96%) 2,772 (3.50%) 1,981 (5.00%) 791 (2.00%) 115 (100%) 0.0096 0.0082

Control N (%) 75,993 (100%) 37,834 (49.79%) 38,159 (50.21%) 2,757 (3.63%) 2,081 (5.50%) 676 (1.77%) 160 (100%) 0.016 0.007

Australian Bureau of Statistics’ 2006 Census of Population. Alcohol Action in Rural Communities project’s 2005 Community Survey AUDIT scores ≥20. Compiled from information obtained through Divisions of General Practice. Five (5) imputations, number of observations = 37, complete degrees of freedom = 33. Note: Series corrected for first-order autocorrelation. Principal diagnosis ICD 10 code F-102 (World Health Organization, 2007).

replaced by a set of five imputations (m) in the multivariate model, generating five plausible complete data sets (first step; Schafer, 1999; Sinharay et al., 2001). Each imputed dataset is then analysed (second step) with standard methods which are then combined (third step) to obtain final pooled parameters and confidence intervals incorporating missing-data uncertainty (Little and Rubin, 1987; Rubin, 1987; Schafer, 1999; Sinharay et al., 2001). Neither the prescribing nor hospitalisations data sets demonstrated non-stationarity or seasonality effects and for the presence of autocorrelation the Prais–Winsten method was used to estimate the regressions (Judge et al., 1985; Lagarde, 2011). Segmented regression analysis of interrupted time series, with a change point at the time of intervention, was used (Wagner et al., 2002). The ˇ coefficients from the regression model were estimated, separately for the prescribing rates for each pharmacotherapy and the hospitalisation rates. The regression model first accounted for the mean quarterly level both at baseline (ˇ0 ) and the change immediately postintervention (ˇ2 ); the mean quarterly trend for the pre-intervention (ˇ1 ) period and the change in the trend for the post-intervention (ˇ3 ) period within the experimental communities. Subsequently, data from both experimental and control communities were used and a bivariate interaction term with a variable named group, where 1 was the experimental communities and 0 was the control communities (Hartung et al., 2010), was added to estimate the differences between experimental and control communities, for both the prescribing rates for each pharmacotherapy and hospitalisation rates. These included the pre-intervention baseline level (group), the change in level immediately after the intervention (group * intervention1 ), the trend before (group * time0 ) and the change in the trend after the intervention (group * time1 ) between experimental and control communities. All data were analysed using Stata 11.1 (StataCorp, 2009b). 2.6.1. Cost effectiveness analysis. The cost-effectiveness analysis was conducted using a health care perspective. Incremental cost-effectiveness ratios (ICERs) were estimated to compare additional costs for acamprosate and naltrexone prescriptions to the change in costs of alcohol dependence hospitalisations as a result of the intervention. 2.6.2. Sensitivity analyses. Small numbers of m might be considered insufficient when using the MIM, because statistical power for small effect sizes decreases, and this reduction in statistical power is greater than predicted changes in relative efficiency (Graham et al., 2007). One-way sensitivity analyses were carried out to explore the robustness of effectiveness results by varying the numbers of m to 10 and 20. 2.6.3. Uncertainty analyses. Uncertainty in cost-effectiveness outputs were evaluated by Monte Carlo simulation (2000 iterations) using Ersatz version 1.13 (EpiGear International Pty Ltd, 2011). A 95% uncertainty interval (UI) was calculated from the values resulting from the iterations (Briggs et al., 2002).

18–62 in the experimental and control communities. There was approximately one GP for every 24 dependent drinkers aged 18–62 in the experimental communities, compared to one GP for every 17 dependent drinkers in the control communities. 3.2. Prescribing rates The impact of the intervention on prescribing rates of acamprosate and naltrexone are summarised in Table 2. For the experimental communities, the mean prescribing rate for acamprosate significantly increased immediately after implementation of the intervention (Box 1: 2.31–3.47; 95% CI 1.92–5.02; p < 0.001), but significantly decreased for naltrexone (Box 1: 1.97–1.85; 95% CI 1.26–2.45; p < 0.001), relative to the baseline period. Compared to the control communities, the mean prescribing rate for acamprosate immediately after implementation of the intervention in the experimental communities also increased significantly (Box 2: 1.57; 95% CI 0.68–2.46; p < 0.01), but decreased significantly for naltrexone (Box 2: −0.79; 95% CI −1.23 to −0.35; p < 0.01). From pre- to post-intervention, changes in mean quarterly prescribing rate trends relative to the control communities, significantly increased for acamprosate (Box 2: 0.24; 95% CI 0.13–0.35; p < 0.001), but significantly decreased for naltrexone (Box 2: −0.12; 95% CI −0.17 to −0.06; p < 0.001). 3.3. Alcohol dependence hospitalisation rates The impact of the intervention on hospitalisation rates of alcohol dependence are summarised in Table 2. The mean hospitalisation rates for the experimental communities significantly increased from baseline to post-test immediately after implementation of the intervention (Box 1: 1.30–1.37; 95% CI 0.91–1.83; p < 0.001), although the immediate change was not significant relative to the control communities (Box 2: −0.33; 95% CI −0.79 to 0.14; p = 0.166). The pre- to post-test change in the mean quarterly hospitalisation rate trends in the experimental communities was statistically significantly less than in control communities (Box 2: −0.07; 95% CI −0.13 to −0.01; p < 0.05).

3. Results 3.4. Cost effectiveness 3.1. Demographics Table 1 shows that the demographic characteristics of the experimental and control communities at baseline were comparable. There were similar predicted numbers of dependent drinkers aged

Table 3 summarises the cost components taken into account for the postal, tailored letters (average cost AUD$24 per GP), and the average differences in costs for post-intervention changes in the immediate quarter (level) and trend per quarter for additional

H.J. Navarro et al. / Drug and Alcohol Dependence 124 (2012) 207–215

211

Table 2 Trends in prescribing rate/10 GPs of acamprosate and naltrexone for dependent drinkers and for alcohol dependence hospitalisations/10,000 population pre- and postintervention in experimental communities alone and relative to control.

Acamprosatea

Naltrexonea

Hospitalisationsb

Box 1

Box 2

Experimental communities

Relative to control communities

Estimate

95% CI

p-Value

Baseline level ˇ0 Pre trend ˇ1 Level change ˇ2 Post trend ˇ3

2.31 −0.02 3.47 0.17

0.47 to 4.15 −0.13 to 0.09 1.92 to 5.02 −0.04 to 0.38

<0.05 0.742 <0.001 0.115

Estimate

95% CI

p-Value

0.04 1.57 0.24

−0.02 to 0.09 0.68 to 2.46 0.13 to 0.35

0.173 <0.01 <0.001

Baseline level ˇ0 Pre trend ˇ1 Level change ˇ2 Post trend ˇ3

1.97 0.01 1.85 −0.01

1.38 to 2.56 −0.03 to 0.05 1.26 to 2.45 −0.09 to 0.08

<0.001 0.633 <0.001 0.899

−0.03 −0.79 −0.12

−0.06 to −0.01 −1.23 to −0.35 −0.17 to −0.06

<0.01 <0.01 <0.001

Baseline level ˇ0 Pre trend ˇ1 Level change ˇ2 Post trend ˇ3

1.30 0.01 1.37 −0.02

0.92 to 1.68 −0.02 to 0.04 0.91 to 1.83 −0.08 to 0.05

<0.001 0.512 <0.001 0.702

−0.005 −0.33 −0.07

−0.03 to 0.02 −0.79 to 0.14 −0.13 to −0.01

0.723 0.166 <0.05

Note: Baseline level ˇ0 : prescribing rate or hospitalisation at quarter 1 (October–December 2000); pre trend ˇ1 : quarterly change in prescribing rate or hospitalisations in period before intervention; level change ˇ2 : change in prescribing or hospitalisation rates immediately after intervention (October–December 2006); post trend ˇ3 : quarterly change in prescribing rate or hospitalisations in period after intervention. Prescribing and hospitalisation rate estimates relative to control are adjusted for respective trends in the control communities and interpreted likewise. CI: confidence interval. Bold values highlight significant changes in prescribing and hospitalisation rates. a Overall model, pooled estimates for five (5) imputation sets, number of observations = 37, complete degrees of freedom = 33. Series corrected for first-order autocorrelation. b Alcohol dependence as principal diagnosis. Number of observations = 37, complete degrees of freedom = 33.

Table 3 Summary of costs. Postal, tailored lettersa

Units

Quantity/GP

Generating letter template and tailored letters to GPsb Paper and envelope Colour printing Local postage

69.14 per hour 0.06 per envelope 0.28 per page 0.35 per item

20 min 1 1 1

$2,645 $7 $32 $40

10 min

$2,724 $24 $16

Subtotal Cost per GP Average time spent per GP reading tailored letterc Total number of GPs Number of GPs reading the letter

94.35 per hour 115 104

Costs (2006 AUD prices)

$1,625 $4,349

Total Additional prescriptions Average number of additional prescriptions Level Trend Average cost per GP services 2006–2009d Level Trend Average cost per pharmacotherapy 2006–2009d Level Trend Total average treatment costs per quarter Level Trend Additional hospitalisations avertede Level Trend Total average cost per hospitalisation 2006–2009 Total average costs per quarter Level Trend a

Acamprosate

Naltrexone

18.06 2.76 $63 $1,137 $174 $166 $2,995 $458

−9.09 −1.38 $63 −$572 −$87 $150 −$1,367 −$208

$4,132 $643

−$1,939 −$289 Alcohol dependence

Costs (2006–2009 AUD prices)

$2,193 $354 Costs (2006–2009 AUD prices)

5 1 $4,410 $26,433 $5,420

$26,433 $5,420

Source: Alcohol Action in Rural Communities project; AUD: Australian dollars; GP: general practitioner. Source: Medical Benefits Schedule 2006–2009. c Source: Medicare Australia 2006–2009, for a one month’s supply. d Source: New South Wales cost weights & the Australian Refined Diagnosis Related Groups (AR-DRGs) patient classification system 2006–2009, for alcohol dependence as principal diagnosis. e Level C Step 1; 35 h/week +DOE 2.5%. b

212

H.J. Navarro et al. / Drug and Alcohol Dependence 124 (2012) 207–215

Table 4 Cost-effectiveness for average quarterly additional prescriptions and hospitalisation(s) averted compared to control communities (1 October 2006 to 31 December 2009) in AUD (2006–2009). Change

Outcomes

Additional prescriptions Trend Mean (95% UI) 1.45 (0.95–1.97) Hospitalisations averted Mean (95% UI) Trend 1.02 (0.36–1.70)

Cost offsetsa

Interventionb + treatment costsa

Net costsa

ICER

Mean (95% UI) –

Mean (95% UI) $4,703 (4,585–4,821)

Mean (95% UI) $4,703 (4,585–4,821)

Median (95% UI) $3,243 ($2,385 - $4,860)

$5,420 (1,875–8,965)

$4,703 (4,585–4,821)

-$717 (−4,175–2,740)

Dominant (Dominant – $12,750)

AUD: Australian dollars; UI: uncertainty interval. a 2006–2009 AUD. b 2006 AUD.

Fig. 3. Cost-effectiveness of postal, tailored prescribing feedback on pharmacotherapies (acamprosate and naltrexone) in experimental communities compared to current practice (without prescribing feedback) in control communities. Note: Net cost = intervention and treatment costs.

prescriptions (which include GP services and pharmacotherapies costs). The average potential cost savings per quarter per hospitalisation(s) for alcohol dependence averted, in the 10 experimental communities, was AUD$5420, or AUD$21,680 per annum. Table 4 shows the median ICERs per additional prescriptions of acamprosate and naltrexone combined, per quarter, after implementation of the intervention was AUD$3243 ($4703/1.45). For alcohol dependence hospitalisations, the median ICER per hospitalisation averted was Dominant (Dominant – $12,750) per quarter. 3.5. Sensitivity analyses Overall, increasing the number of imputations to 10 or 20 did not vary the trajectory of the ˇ coefficients1 nor their statistical significance. For acamprosate, there were minimal changes for the ˇ2 coefficient (3.53 with m = 10 and 3.48 with m = 20 compared to 3.47 with m = 5) and no change in the significance of any parameter with 10 or 20 imputations compared to that with five. For naltrexone, there were no changes either in the ˇ coefficients or in the statistical significance of any parameter with additional imputations. 3.6. Uncertainty analyses Fig. 3 shows the incremental costs and effectiveness for quarterly average additional prescriptions of pharmacotherapies combined, with the intervention falling in the north-east quadrant of the cost-effectiveness plane, indicating an increase in the number of prescriptions processed by a pharmacy following the GP intervention, but at an additional cost to the health care system.

1 Results of the sensitivity analyses can be found in supplementary materials by accessing the online version of this paper. Please see Appendix A for more information.

Fig. 4. Cost-effectiveness of postal, tailored prescribing feedback on hospitalisations averted in experimental communities compared to current practice (without prescribing feedback) in control communities. Note: Net cost = intervention and treatment costs, and cost savings.

In relation to quarterly average hospitalisations averted, Fig. 4 shows that the intervention falls primarily in the north-east and south-east quadrants of the cost-effectiveness plane, indicating that in approximately 60% of cases the reduction in hospitalisations following the intervention resulted in net cost-savings. 4. Discussion Subsequent to receiving the tailored feedback, GPs in the experimental communities prescribed statistically significantly more acamprosate and statistically significantly less naltrexone. Relative to GPs in the control communities, GPs in the experimental communities also prescribed statistically significantly more acamprosate and less naltrexone after the intervention, both on average and over time (trend analysis). The sensitivity analyses did not change these results, indicating that this is a robust finding. As a possible consequence of the intervention, the observed quarterly hospitalisation rate trend was statistically significantly less in the experimental communities, compared to the controls. The cost-effectiveness analysis showed that the median ICER for prescription costs per quarter was AUD$3243 and the median ICER for hospitalisation costs averted per quarter was AUD$12,750. Before considering the interpretation of these findings, there are a number of caveats that need to be mentioned. First, although prescribing data for the three quarters before the intervention were not supplied by Medicare Australia, this represents less than 8.2% of the longitudinal sample, well within recommended limits of 10–20% missing data to justify imputation (Velicer and Colby, 2005). The MIM used is appropriate because it generates repeated, randomly drawn imputations resulting in accurate estimates for parameters such as level, error variance, slope and/or dependency, and is applicable to a range of statistical analysis models (Croiseau et al., 2007; Rubin, 1987; Tufis¸, 2008). In addition, the MIM allows imputations that can be used irrespective

H.J. Navarro et al. / Drug and Alcohol Dependence 124 (2012) 207–215

of whether the data are randomly missing (Rubin, 1987). Moreover, increasing the number of imputations from m = 5 to m = 10 and m = 20 resulted in similar changes and trajectories in level and slope, similar statistical significance, and the same mean differences in all of the parameters analysed, relative to the control communities, thereby confirming the robustness of the results. Second, the prescribing data supplied by Medicare Australia represent prescriptions dispensed at a pharmacy, rather than all prescriptions actually written by GPs. Although these data comprise an estimated 90% of all prescriptions dispensed in Australia (Edmonds et al., 1993), they may under-represent either naltrexone or acamprosate prescribing by GPs if one or other of these pharmacotherapies is disproportionately less likely to be filled: given that acamprosate requires three times a day dosing or its main side effect may hold back adherence for some patients (i.e., diarrhea), it is arguable that fewer acamprosate prescriptions could be filled, relative to naltrexone, particularly since dependent drinkers are known to have high drop-out rates from treatment (McNeely and Sherman, 2011). Conversely, however, it is possible that naltrexone’s main side effect (i.e., nausea) could have the opposite effect on the rate with which prescriptions are actually filled. In any case, there is little reason to conclude that patients of GPs in the experimental communities are less likely to fill a prescription for alcohol pharmacotherapies than patients of GPs in the control communities. Therefore, although some potential prescription costs may have been missed, these costs are assumed to be minimal. Third, the potential number of dependent drinkers was estimated using self-reported alcohol consumption on the AUDIT questionnaire in the AARC community survey. Given a response rate of 40% to this survey respondents may not be representative of their communities, since females and older people are overrepresented in the sample (Petrie et al., 2008), and those who completed the survey may under-report their drinking. Nevertheless, previous surveys have reported similar prevalence rates of alcohol dependence for males and females in Australia (Hall et al., 1999; Proudfoot and Teesson, 2002) and these survey data were not used to evaluate the effectiveness or cost-effectiveness of the intervention. Notwithstanding its limitations, this evaluation shows that, overall, the main objective of significantly increasing the prescribing rate of pharmacotherapies for alcohol dependence by GPs in the experimental communities was achieved both for the immediate quarter after the tailored feedback to GPs, as well as for the overall study period. The statistically significant increase over time in acamprosate prescriptions in the experimental communities outweighed the statistically significant decrease in naltrexone prescriptions. In practical terms, this suggests the effect of the intervention was twofold: that some GPs who were prescribing naltrexone switched to prescribing acamprosate; and other GPs increased the frequency with which they prescribed acamprosate. The possible explanation for GPs’ apparent decrease in naltrexone prescriptions and preference for acamprosate is their increased perception that acamprosate has been shown to be associated with longer periods of alcohol abstinence, a likely lower risk for early drop-outs from treatment due to adverse events and the lack of a dose-related hepatotoxicity (Rösner et al., 2010a). In terms of hospital morbidity, the quarterly trend in incidence of hospitalisations for alcohol dependence as principal diagnosis showed a non-significant decrease within the experimental communities over time, but the rate of these hospitalisations at post-test in the experimental communities was statistically significantly less than in the control communities. This statistically significant difference may be attributable to the increase in prescribing of acamprosate by GPs in the experimental communities: it is likely that the RCT design randomly allocated any extraneous

213

differences equitably between the intervention and control communities. Although the validity of using the ICD-10 diagnosis to detect all alcohol-related hospitalisations for alcohol dependence is known to be less than optimal, this study was only concerned with relative changes in hospitalisations in the experimental communities, relative to the control communities. The reliability of this measure is similarly unknown, although it is unlikely that hospitalisations for alcohol dependence in the experimental communities would significantly change independently of a change in GPs treatment of their alcohol dependent patients. The generalisability of results is acknowledged as a major concern in the alcohol literature and is especially pertinent in this case given that previous results from RCTs on the effectiveness of written feedback on prescribing practices in primary care have been mixed (Anderson et al., 1996; Gehlbach et al., 1984; Hux et al., 1999; Lagerløv et al., 2000; Naughton et al., 2009; Nilsson et al., 2001; O’Connell et al., 1999; Schectman et al., 1995; Søndergaard et al., 2002, 2003; Veninga et al., 1999). There are, however, a number of reasons that could explain why the intervention was effective in this study compared to some previous efforts. First, the feedback was framed as an opportunity for GPs to improve the quality of health for their alcohol dependent patients by increasing their prescribing of pharmacotherapies, rather than emphasising the limitations of their current practice using aggregated data, or by invading their privacy, in relation to their prior knowledge and their criteria to prescribe or not a particular medication. Second, the message delivered was succinct, clear and tailored to their specific circumstances, such as GPs’ current prescribing pattern of acamprosate and naltrexone in their communities and the proportion of dependent drinkers in their specific community that could potentially benefit from their increase in prescribing of treatment. Third, the feedback was not provided by the pharmaceutical industry, the government or an administrative authority, such that it clearly had no commercial or accountability purpose (O’Connell et al., 1999; Søndergaard et al., 2002). Finally, since organisations with a specific role to support GPs, such as the Royal Australian College of General Practitioners, have been increasingly providing their GPs with support and training in addiction medicine, it may be that the timing of this written feedback reinforced these broader training programs (McAvoy, 2008). 5. Conclusion Postal, tailored feedback to GPs on their prescribing patterns of acamproaste and naltrexone for alcohol dependence was costeffective compared to current practice, in rural communities of NSW, increasing the overall prescribing of pharmacotherapies, with an effect on tertiary prevention by reducing alcohol dependence hospitalisations. Although there was a significant reduction in the prescribing of naltrexone, prescriptions for acamprosate increased significantly, as did prescriptions of these pharmacotherapies combined with a median ICER of AUD$3243 per additional quarterly prescriptions compared to the control communities. In addition, the intervention was dominant in at least 60% of cases for reductions of hospitalisations of alcohol dependence, generating net-savings in the experimental communities compared to the control. Role of funding source This study was supported by funding from the Alcohol Education and Rehabilitation Foundation of Australia (AERF). The AERF had no further role in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the paper for publication.

214

H.J. Navarro et al. / Drug and Alcohol Dependence 124 (2012) 207–215

Contributors HJN wrote the drafts of the manuscript. All authors took an active role in commenting and analyzing the drafting of the manuscript. All authors commented on data analyses and CMD and DJP commented on the costing. All authors have seen and approved the final version of the manuscript. Ethics approval All study procedures were approved by the Human Research Ethics Committee’s of the University of Newcastle and the University of New South Wales. Conflict of interest No conflict declared. Acknowledgement Grateful appreciation is expressed to all GPs of the experimental AARC communities. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.drugalcdep.2012.01.007. References Anderson, J.F., McEwan, K.L., Hrudey, W.P., 1996. Effectiveness of notification and group education in modifying prescribing of regulated analgesics. CMAJ 154, 31–39. Anton, R.F., O’Malley, S.S., Ciraulo, D.A., Cisler, R.A., Couper, D., Donovan, D.M., Gastfriend, D.R., Hosking, J.D., Johnson, B.A., LoCastro, J.S., Longabaugh, R., Mason, B.J., Mattson, M.E., Miller, W.R., Pettinati, H.M., Randall, C.L., Swift, R., Weiss, R.D., Williams, L.D., Zweben, A., COMBINE Study Research Group, 2006. Combined pharmacotherapies and behavioral interventions for alcohol dependence: the COMBINE study: a randomized controlled trial. JAMA 295, 2003–2017. Australian Bureau of Statistics, 2009a. 2001 Census Data Online, http://www. abs.gov.au/websitedbs/censushome.nsf/home/Data (accessed on 14.02.10). Australian Bureau of Statistics, 2009b. 2006 Census Data Online, http://www.abs. gov.au/websitedbs/censushome.nsf/home/Census (accessed on 14.02.10). Australian Institute of Health and Welfare, 2005. In: AIHW (Ed.), Rural, Regional and Remote Health: Indicators of Health. AIHW, Canberra, ACT. Babor, T.F., Higgins-Biddle, J.C., Saunders, J.B., Monteiro, M.G., 2001. AUDIT: The Alcohol Use Disorder Identification Test: Guidelines for Use in Primary Health Care. World Health Organization, Geneva, Switzerland. Boothby, L.A., Doering, P.L., 2005. Acamprosate for the treatment of alcohol dependence. Clin. Ther. 27, 695–714. ˜ A., Amate, J.M., 2004. Efficacy and safety of naltrexone Bouza, C., Angeles, M., Munoz, and acamprosate in the treatment of alcohol dependence: a systematic review. Addiction 99, 811–828. Breen, C.L., Shakeshaft, A.P., Doran, C.M., Sanson-Fisher, R.W., Mattick, R.P., 2010. Cost-effectiveness of follow-up contact for a postal survey: a randomised controlled trial. Aust. N. Z. J. Public Health 34, 508–512. Briggs, A.H., O’Brien, B.J., Blackhouse, G., 2002. Thinking outside the box: recent advances in the analysis and presentation of uncertainty in cost-effectiveness studies. Annu. Rev. Public Health 23, 377–401. Commonwealth of Australia, 2007. National Hospital Cost Data Collection, http://www.health.gov.au/internet/main/publishing.nsf/Content/healthcasemix-data-collections-NHCDCReports (accessed on February 2011). Croiseau, P., Génin, E., Cordell, H.J., 2007. Dealing with missing data in family-based association studies: a multiple imputation approach. Hum. Hered. 63, 229–238. Department of Health and Ageing, 2006. Medicare Benefits Schedule Book. Commonwealth of Australia, Canberra, ACT. Department of Health and Ageing, 2007. Product Information (Acamprosate http://www.pbs.gov.au/pbs/pdf-viewer?pdf=%2Fmeds%2Fpi%2 Calcium), Fafpcampr10507.pdf (accessed on January 2011). Department of Health and Ageing, 2009. Product Information (Naltexone hydrochloride), http://www.pbs.gov.au/pbs/pdf-viewer?pdf=%2Fmeds%2Fpi%2 Fbqprevia11109.pdf (accessed on January 2011). Department of Health and Ageing, 2011a. PBS Schedule Acamprosate Calcium Item Code 8357W, http://www.pbs.gov.au/medicine/item/8357W (accessed on January 2011).

Department of Health and Ageing, 2011b. PBS Schedule Naltrexone Hydrochloride Item Code 8370M, http://www.pbs.gov.au/medicine/item/8370M (accessed on January 2011). Department of Veterans’ Affairs, Alcohol Screen (AUDIT). http://www.therightmix. gov.au/resources/documents/D718 - Alcohol Screen AUDIT - Health Professionals1.pdf (accessed on September 2010). Doran, C.M., Fawcett, J.E., Shakeshaft, A.P., Shanahan, M.D., Mattick, R.P., 2003. New pharmacotherapies for alcohol dependence: are they being used and what do they cost? Med. J. Aust. 179, 218. Druss, B.G., von Esenwein, S.A., 2006. Improving general medical care for persons with mental and addictive disorders: systematic review. Gen. Hosp. Psychiatry 28, 145–153. Edmonds, D.J., Dumbrell, D.M., Primrose, J.G., McManus, P., Birkett, D.J., Demirian, V., 1993. Development of an Australian drug utilisation database: a report from the Drug Utilization Subcommittee of the Pharmaceutical Benefits Advisory Committee. Pharmacoeconomics 3, 427–432. EpiGear International Pty Ltd, 2011. Ersatz 1.1. Uncertainty Analysis Software. In: EpiGear International Pty Wilston, QLD, Australia. Fleming, J., 1996. The epidemiology of alcohol use in Australian women: findings from a national survey of women’s drinking. Addiction 91, 1325–1334. Gehlbach, S.H., Wilkinson, W.E., Hammond, W.E., Clapp, N.E., Finn, A.L., Taylor, W.J., Rodell, M.S., 1984. Improving drug prescribing in a primary care practice. Med. Care 22, 193–201. Graham, J.W., Olchowski, A.E., Gilreath, T.D., 2007. How many imputations are really needed? Some practical clarifications of multiple imputation theory. Prev. Sci. 8, 206–213. Gray, D., 2000. Indigenous Australians and liquor licensing restrictions. Addiction 95, 1469–1472. Haber, P., Lintzeris, N., Proude, E., Lopatko, O., 2009. Guidelines for the Treatment of Alcohol Problems. Australian Government Department of Health and Ageing, Canberra. Hall, W., Teesson, M., Lynskey, M., Degenhardt, L., 1999. The 12-month prevalence of substance use and ICD-10 substance use disorders in Australian adults: findings from the National Survey of Mental Health and Well-Being. Addiction 94, 1541–1550. Hartung, D.M., Evans, D., Haxby, D.G., Kraemer, D.F., Andeen, G., Fagnan, L.J., 2010. Effect of drug sample removal on prescribing in a family practice clinic. Ann. Fam. Med. 8, 402–409. Hux, J.E., Melady, M.P., DeBoer, D., 1999. Confidential prescriber feedback and education to improve antibiotic use in primary care: a controlled trial. CMAJ 161, 388–392. Ivis, F.J., Adlaf, E.M., Rehm, J., 2000. Incorporating the AUDIT into a general population telephone survey: a methodological experiment. Drug Alcohol Depend. 60, 97–104. Judge, G.G., Griffiths, W.E., Hill, R.C., Lutkepohl, H., Lee, T.C. (Eds.), 1985. The Theory and Practice of Econometrics. John Wiley and Sons, New York. Kiefer, F., Wiedemann, K., 2004. Combined therapy: what does acamprosate and naltrexone combination tell us? Alcohol Alcohol. 39, 542–547. Kranzler, H.R., Van Kirk, J., 2001. Efficacy of naltrexone and acamprosate for alcoholism treatment: a meta-analysis. Alcohol. Clin. Exp. Res. 25, 1335– 1341. Lagarde, M., 2011. How to do (or not to do). Assessing the impact of a policy change with routine longitudinal data. Health Policy Plan. 27, 76–83. Lagerløv, P., Loeb, M., Andrew, M., Hjortdahl, P., 2000. Improving doctors’ prescribing behaviour through reflection on guidelines and prescription feedback: a randomised controlled study. Qual. Health Care 9, 159–165. Little, R.J.A., Rubin, D.B., 1987. Statistical Analysis with Missing Data. John Wiley and Sons, New York. McAvoy, B.R., 2008. Addiction and addiction medicine: exploring opportunities for the general practitioner. Med. J. Aust. 189, 115–117. McKeon, A., Frye, M.A., Delanty, N., 2008. The alcohol withdrawal syndrome. J. Neurol. Neurosurg. Psychiatry 79, 854–862. McNeely, J., Sherman, S., 2011. Review: acamprosate increases abstinence in patients with alcohol dependence. Ann. Int. Med. 154, JC1–JC10. Medicare Australia, 2010. Clarification Document for Pharmaceutical Benefits Schedule Statistics, https://www.medicareaustralia.gov.au/statistics/ pbs item.shtml (accessed on May 2010). Naughton, C., Feely, J., Bennett, K., 2009. A RCT evaluating the effectiveness and cost-effectiveness of academic detailing versus postal prescribing feedback in changing GP antibiotic prescribing. J. Eval. Clin. Pract. 15, 807–812. Nilsson, G., Hjemdahl, P., Hässler, A., Vitols, S., Wallén, N.H., Krakau, I., 2001. Feedback on prescribing rate combined with problem-oriented pharmacotherapy education as a model to improve prescribing behaviour among general practitioners. Eur. J. Clin. Pharmacol. 56, 843–848. O’Connell, D.L., Henry, D., Tomlins, R., 1999. Randomised controlled trial of effect of feedback on general practitioners’ prescribing in Australia. BMJ 318, 507– 511. Petrie, D., Doran, C., Shakeshaft, A., Sanson-Fisher, R., 2008. The relationship between alcohol consumption and self-reported health status using the EQ5D: evidence from rural Australia. Soc. Sci. Med. 67, 1717–1726. Poikolainen, K., Paljärvi, T., Mäkelä, P., 2011. Risk Factors for alcohol-specific hospitalizations and deaths: prospective cohort study. Alcohol Alcohol. 46, 342–348. Proudfoot, H., Teesson, M., 2002. Who seeks treatment for alcohol dependence? Findings from the Australian National Survey of Mental Health and Wellbeing. Soc. Psychiatry Psychiatr. Epidemiol. 37, 451–456.

H.J. Navarro et al. / Drug and Alcohol Dependence 124 (2012) 207–215 Proudfoot, H., Teesson, M., 2009. The association of alcohol dependence with general practice attendance. Drug Alcohol Rev. 28, 154–159. Reinert, D.F., Allen, J.P., 2007. The alcohol use disorders identification test: an update of research findings. Alcohol. Clin. Exp. Res. 31, 185–199. Rösner, S., Hackl-Herrwerth, A., Leucht, S., Lehert, P., Vecchi, S., Soyka, M., 2010a. Acamprosate for alcohol dependence. Cochrane Database Syst. Rev., 9. Rösner, S., Hackl-Herrwerth, A., Leucht, S., Vecchi, S., Srisurapanont, M., Soyka, M., 2010b. Opioid antagonists for alcohol dependence. Cochrane Database Syst. Rev., 12. Rösner, S., Leucht, S., Lehert, P., Soyka, M., 2008. Acamprosate supports abstinence, naltrexone prevents excessive drinking: evidence from a meta-analysis with unreported outcomes. J. Psychopharmacol. 22, 11–23. Rubin, D.B., 1987. Multiple Imputation for Nonresponse in Surveys. John Wiley and Sons, New York. Samokhvalov, A.V., Popova, S., Room, R., Ramonas, M., Rehm, J., 2010. Disability associated with alcohol abuse and dependence. Alcohol. Clin. Exp. Res. 34, 1871–1878. Schafer, J.L., 1999. Multiple imputation: a primer. Stat. Methods Med. Res. 8, 3–15. Schectman, J.M., Kanwal, N.K., Schroth, W.S., Elinsky, E.G., 1995. The effect of an education and feedback intervention on group-model and network-model health maintenance organization physician prescribing behavior. Med. Care 33, 139–144. Selin, K.H., 2003. Test–retest reliability of the alcohol use disorder identification test in a general population sample. Alcohol. Clin. Exp. Res. 27, 1428–1435. Shakeshaft, A.P., Bowman, J.A., Sanson-Fisher, R.W., 2002. Community-based drug and alcohol counselling: who attends and why? Drug Alcohol Rev. 21, 153– 162. Shanahan, M., Doran, C., Gibson, A., Stafford, J., Mattick, R.P., 2005. Interventions for excessive alcohol use: costs and outcomes of pharmacotherapies and other treatments. NDARC Monograph No 56. NDARC, Sydney. Sinharay, S., Stern, H.S., Russell, D., 2001. The use of multiple imputation for the analysis of missing data. Psychol. Methods 6, 317–329. Snyder, J.L., Bowers, T.G., 2008. The efficacy of acamprosate and naltrexone in the treatment of alcohol dependence: a relative benefits analysis of randomized controlled trials. Am. J. Drug Alcohol Abuse 34, 449–461.

215

Søndergaard, J., Andersen, M., Støvring, H., Kragstrup, J., 2003. Mailed prescriber feedback in addition to a clinical guideline has no impact: a randomised, controlled trial. Scand. J. Prim. Health Care 21, 47–51. Søndergaard, J., Andersen, M., Vach, K., Kragstrup, J., Maclure, M., Gram, L.F., 2002. Detailed postal feedback about prescribing to asthma patients combined with a guideline statement showed no impact: a randomised controlled trial. Eur. J. Clin. Pharmacol. 58, 127–132. Soyka, M., Rösner, S., 2010. Emerging drugs to treat alcoholism. Expert Opin. Emerg. Drugs 15, 695–711. StataCorp, 2009a. STATA Multiple-Imputation Reference Manual Release 11. StataCorp LP, College Station, TX. StataCorp, 2009b. STATA Statistics Data Analysis 11.1. Statistical Software. StataCorp LP, College Station, TX. Stockwell, T.R., Heale, P., Chikritzhs, T.N., Dietze, P., Catalano, P., 2002. How much alcohol is drunk in Australia in excess of the new Australian alcohol guidelines? Med. J. Aust. 176, 91–92. Tufis¸, C.D., 2008. Multiple imputation as a solution to the missing data problem in social sciences. Calitatea Viet¸ti, 19. Velicer, W.F., Colby, S., 2005. A comparison of missing-data procedures for arima time-series analysis. Educ. Psychol. Meas. 65, 596–615. Veninga, C.C., Lagerløv, P., Wahlström, R., Muskova, M., Denig, P., Berkhof, J., Kochen, M.M., Haaijer-Ruskamp, F.M., 1999. Evaluating an educational intervention to improve the treatment of asthma in four European countries. Drug Education Project Group. Am. J. Respir. Crit. Care Med. 160, 1254–1262. Wagner, A.K., Soumerai, S.B., Zhang, F., Ross-Degnan, D., 2002. Segmented regression analysis of interrupted time series studies in medication use research. J. Clin. Pharm. Ther. 27, 299–309. Wayman, J.C., 2003. Multiple Imputation For Missing Data: What Is It And How Can I Use It? Center for Social Organization of Schools, Johns Hopkins University, Baltimore, MD. World Health Organization, 2007. International Statistical Classification of Diseases and Related Health Problems, 10th Revision, http://apps.who.int/ classifications/apps/icd/icd10online/ (accessed on 12.01.11). World Health Organization, 2009. Global Health Risks: Mortality and Burden of Disease Attributable to Selected Major Risks. WHO, Geneva, Switzerland.