Cost-effectiveness of a programme of screening and brief interventions for alcohol in primary care in Italy

We obtained consumption data for the Italian population from the Aspects of Daily Life survey 2008 [18, 19], conducted by the Italian national statistics institute (ISTAT). This nationally representative survey (N = 48,861) records demographic data on each respondent as well as asking a series of quantity-frequency questions regarding their usual alcohol consumption. These responses were converted to a mean weekly consumption in grams of alcohol. The survey also asks respondents how many times in the preceding year they have drunk 6 or more glasses of alcoholic beverage (1 glass = 12 grams of pure alcohol). We used this as a measure of the risk of harm for health conditions associated with acute, rather than chronic, alcohol consumption.

We modelled 42 alcohol-attributable health conditions based on the work of the WHO on the global burden of disease due to alcohol [20, 21] (see Table 1). Absolute mortality data from 2008 for the 42 health conditions was obtained from ISTAT for each age and gender subgroup in the model. Morbidity data was derived from the Italian database of hospital admissions for 2008, which contained a total of 1,559,310 admissions for the 42 alcohol-related health conditions. In order to account for repeat admissions in the same year for the same individual, the absolute number of admissions was divided by the adjustment coefficients shown in Table 1. These coefficients were calculated from hospital admissions data for the Netherlands as no Italian data was available and were obtained from the Dutch Hospital Data Foundation (DHD) [22]. UK coefficients were used in previous versions of the Sheffield Model [15], however these were estimated indirectly from several data sources, whereas the DHD data allowed them to be calculated directly. The coefficients represent the mean number of admissions in a year for an individual with each health condition and full details on their derivation are contained within the Additional file 1. Hospital admissions data for Italy was not available for any of the partially-attributable acute conditions, so these were estimated from UK morbidity data using the relationship between Italian and UK mortality rates for each relevant condition.

Risk functions relating alcohol consumption (either mean consumption level or frequency of acute drinking) to risk are required to implement the potential impact fraction method. Following a recent review by the Sheffield Alcohol Research Group, updated risk functions were identified for malignant neoplasm of the lip, oral cavity and pharynx [23], malignant neoplasm of the liver and intrahepatic bile ducts [24], malignant neoplasm of the larynx [25], malignant neoplasm of the breast [26], epilepsy [27] and cardiac arrhythmias [28]. For all other partially-attributable chronic conditions the same risk functions as have been implemented previously [29] were used. For partially attributable acute conditions (e.g. falls and assaults) we used Italy-specific alcohol-attributable fractions [19] where available, or otherwise those from England [30], and assumed a linear relationship between frequency of drinking >6 glasses and risk to calculate risk functions for each age-gender subgroup. For wholly-attributable partial and chronic conditions, the risk functions were calibrated to the actual absolute levels of morbidity and mortality in the population, assuming a linear relationship between alcohol consumption/binge frequency and risk. Further details of the calibration methodology and risk functions used are included in the Additional file 1.

The model incorporates all healthcare costs to the INHS for each health condition, including inpatient, outpatient and accident and emergency visits, ambulance costs, GP consultations, nurse visits and other costs. For each cost category identified, we aimed to collect context-specific costs in line with the study perspective. In Italy, hospitals are funded through reimbursement tariffs that vary across 21 regions and between different types of providers. The reimbursement tariffs are estimated on the basis of full cost of hospitalisation and are inclusive of all inpatient services in addition to emergency visits if these led to hospitalisation [31]. Since we aimed to obtain nationally representative estimates, we assigned hospital costs to each admission in the database based on the nationally defined reimbursement tariff which applies when a patient is treated outside their region of residence. Out of 42 identified alcohol-attributable conditions, Italian data were available for 33. Mean costs were calculated for each condition by gender and tested for significant differences between sexes using student t-tests in Stata 11 [32]. Costs were significantly different (p < 0.01 in all cases) for 11 of the conditions for which data was available. Where there was no significant difference (p > 0.23 in all cases) the overall population mean was used. For conditions where no cost data was available in Italy, English costs [15] were adjusted by the mean ratio between Italian and English costs for all other conditions. This assumption was tested with a sensitivity analysis assuming, conservatively, that these costs were 25% lower than the baseline estimates. Costs not covered within the hospital reimbursement tariffs (e.g. ambulance and GP costs) were estimated assuming the ratio of hospital admission to other costs for each condition was the same as in England.

The costs associated with implementing a screening and brief intervention programme were separated into the cost of briefing materials provided to the patient and the cost of the GP’s time. The former were taken from a UK study by Lock et al. [33], converted into euros using OECD purchasing power parities [34] and inflated to 2008 prices using the consumer prices index for Italy [35]. To obtain an estimate of the GP’s time, we first estimated the annual salaries of GPs of different levels of seniority, using data from the Friuli-Venezia-Giulia (FVG) region of Italy. We took an average of these, weighted by the proportion of GPs at each level in the province of Udine (in the FVG region) to give us an estimate for the mean annual salary of a GP of €79900 before tax. Italian GPs spend an average of 15 hours contact time in surgery with patients [36] and an estimated 12.5 hours on home visits, giving us an average direct cost per minute to the INHS of €0.87 for this contact time (after adjusting to 2008 prices). As no data exists on the costs of overheads and other related costs (such as ongoing training) these were estimated relatively from UK figures [37], giving a total cost per minute to the INHS of €1.07. Owing to the uncertainty around this figure, an alternative estimate of €1.58, derived using the absolute UK costs, was used as a sensitivity analysis.

Two alternative intervention settings were examined: the screening of patients when they next registered with a new GP and screening when they have their next standard GP consultation. Data on between-GP migration by patient age and gender was obtained for the Friuli-Venezia-Giulia region for a 10 year period from 2000–2009 (personal communication from Roberto Maffetone at INSIEL) together with regional population demographics for the same period from ISTAT. These were used to derive a gender and age-group specific ‘arrival profile’, after adjusting for long-term trends in migration, giving the probability of being screened in each year of a 10 year screening programme, assuming that the probability of registering with a new GP was independent from year to year. Data on the mean frequency of GP consultations by age and gender was obtained for both Italy [38] and England [39] and used to estimate the proportion of patients in each subgroup who would visit their GP in each year of the programme. Full details of the methodology used in this estimation are given in an additional file [Additional file 1].

In line with current Italian guidelines, screening is modelled using the AUDIT-C questionnaire with a threshold of 4 for women and 5 for men [40] with a brief intervention lasting 10 minutes for all patients who screen positive. The probability of a patient screening positively, given their alcohol consumption, was estimated using screening models derived from the UK Psychiatric Morbidity Surveys from 2000 and 2007. Evidence on the relationship between duration of intervention and effectiveness is inconclusive [7] and therefore the mean effect estimate of a 12.3% reduction in consumption after a 24.9 minute intervention, taken from the Cochrane review of Kaner et al. [6], is applied to all patients who receive the intervention. This reduction is assumed to rebound linearly towards the patient’s baseline consumption over a period of 7 years in line with the findings of Fleming et al. [41].

In order to investigate the impact of uncertainty around the hospital costs estimated indirectly from UK costs, the cost of GP’s time, the duration and magnitude of effect of the intervention on patients’ drinking and the relationship between these and the duration of the brief intervention itself, we conducted a number of sensitivity analyses using more pessimistic assumptions for these parameters. A number of alternative scenarios were also examined to explore the estimated impact of using different screening tools, such as the full AUDIT questionnaire, FAST, which has been recommended for use in primary care [42], or combinations of several tools (e.g. full AUDIT with an AUDIT-C pre-screen).