Improving screening and brief intervention activities in Primary Health Care: secondary analysis of

39 Introduction and objective The ODHIN trial found that training and support and financial 40 reimbursement increased the proportion of patients that were screened and given advice for their 41 heavy drinking in primary health care. However the impact of these strategies on professional 42 accuracy in delivering screening and brief advice is under-researched and is the focus of this paper. 43 Method From 120 primary health-care units (24 in each jurisdiction: Catalonia, England, the 44 Netherlands, Poland and Sweden), 746 providers participated in the baseline and the 12-week 45 implementation periods. Accuracy was measured in two ways: correctness in completing and scoring 46 the screening instrument, AUDIT-C; the proportion of screen negative patients given advice, and the 47 proportion of screen positive patients not given advice. Odds ratios of accuracy were calculated for 48 type of profession, and for intervention group: training & support; financial reimbursement; and, 49 internet-based counselling. Results. 32 of 36,711 questionnaires were incorrectly completed, and 50 65 of 29,641 screen negative patients were falsely classified. At baseline, 27% of screen negative 51 patients were given advice, and 22.5% screen positive patients were not given advice. These 52 proportions halved during the 12-week implementation period, unaffected by training. Financial 53 reimbursement reduced the proportion of screen positive patients not given advice (OR = 0.56, 95% 54 CI=0.31 to 0.99, p<0.05). Conclusion. Although the use of AUDIT-C as a screening tool was 55 accurate, a considerable proportion of risky drinkers did not receive advice, which was reduced with 56 financial incentives. 57

Organization (WHO) as a screening instrument for use in primary health care [5]. The AUDIT contains 67 ten questions and can be used to identify individuals drinking at hazardous and harmful levels 68 (identified as an alcohol use disorder). A shorter form of AUDIT is the AUDIT-C, which includes only 69 the three alcohol questions of the full AUDIT, has been validated for use in primary health care in the 70 (AUDIT-C) as a screening tool to promote early identification of hazardous and harmful drinking and 76 tested three strategies alone, and in combination, to encourage clinicians to give brief alcohol advice 77 to patients as follows: training and support (TS), financial incentives (FR) and internet-based 78 counselling (eBI). While the most commonly used cut off points in the AUDIT-C are ≥5 for men and ≥4 79 for women [5], the ODHIN trial used cut off points of ≥5 for men and women in Catalonia and 80 England. These cut offs avoid the risk of excessive false positives among women [15], where a score 81 of 5 is equivalent to a consumption level of about 20 grams of alcohol per day [16]. 82 83 Further, despite its validity as a screening instrument for use in primary health care, the use of 84 AUDIT-C has shown some inconsistencies between the final classification result of either a positive or 85 negative score. One study showed that up to 21% of men and women were misclassified, because of 86 either an underestimation of alcohol consumption, stigma, or a previous alcohol use disorder ( classifications as risky or non-risky drinkers according to the AUDIT-C and reported drinking limits as 96 reported by patients, but none have assessed inconsistencies in professionals' performance. We 97 collected nearly 36,000 screening questionnaires during the ODHIN baseline and 12-week 98 implementation periods from the included questionnaires. All questionnaires included completed 99 AUDIT-C questions as well as information relating to whether or not brief advice was delivered. Our 100 main objective was to assess the accuracy of screening tool completion, errors in its scoring, and the 101 incorrect provision of brief advice at both baseline and 12-week ODHIN implementation periods.

Accuracy of completing AUDIT-C 117
The accuracy of completing AUDIT-C was assessed by two different indicators: the accuracy of the 118 AUDIT-C scoring, in which any noted/recorded value other than between 0 and 4 (correct response 119 categories for AUDIT-C) for any of the three AUDIT-C questions was considered incorrect; and, the 120 accuracy of the professionals' scoring of the AUDIT-C for each of the three separate AUDIT-C 121 questions, compared to the authors' scoring, with any deviation considered wrong. In both cases, the 122 proportion of patient questionnaires with an error was calculated. 123 124

Accuracy of advice 125
The accuracy of advice was assessed by calculating the proportion of screen negative patients that 126 received advice, and the proportion of screen positive patients that did not receive advice. 127 128

Statistical methods 129
The original trial was conceived and analysed as a factorial design. A generalised linear model 130 utilizing logistic models for binary data was used employing a multi-level approach using country and 131 PHCU with random intercepts and slopes. Analysis was conducted using IBM SPSS V23, procedure 132 GENLIN. 133

135
During the study, 746 providers from 120 primary health care units (24 per each of the five 136 jurisdictions) participated in the study. During the four-week baseline measurement period, 6,091 137 questionnaires were available for analysis, and during the 12-week implementation period, 30,623. 138 Two-thirds of questionnaires were completed by doctors, and one third by non-doctors (nurses and 139 practice assistants). Table 1  During the 12-week implementation period, the proportion of screen negative patients given advice 199 was 13% amongst patients whose providers had received financial reimbursement compared with 200 18% amongst patients whose providers had not received financial reimbursement (OR in favour of 201 financial reimbursement = 0.66, 95% CI=0.34 to 1.28, ns); the proportion of screen positive patients 202 not given advice was 10% amongst patients whose providers had received financial reimbursement 203 compared with 17% amongst patients whose providers had not received financial reimbursement 204 (OR in favour of financial reimbursement = 0.56, 95% CI=0.31 to 0.99, p<0.05). 205 206 During the 12-week implementation period, the proportion of screen negative patients given advice 207 was 15% amongst patients whose providers had the option of e-BI compared with 16% amongst 208 patients whose providers did not have the option of eBI (OR in favour of e-BI = 0.91, 95% CI=0.40 to 209 2.09, ns); the proportion of screen positive patients not given advice was 16% amongst patients 210 whose providers had the option of eBI compared with 11% amongst patients with providers who did 211 not have the option of eBI (OR in favour of eBI = 1.60, 95% CI=0.89 to 2.85, ns). practitioners are asked to screen and intervene for alcohol in all primary care patients, some 242 professional and patient variables modified the provision of advice with only 50% of those 243 categorized as risky drinkers receiving a brief intervention [19]. No patient variables were included in 244 our analysis as predictors of accurate provision of advice, but when professionals received financial 245 reimbursement, their accuracy in the provision of advice was higher than those that did not receive 246 this incentive. 247 248

Strengths and weaknesses 249
There are some strengths and weaknesses in our study. To our knowledge, our study is the first to 250 analyse some aspects of the fidelity to alcohol SBI guidelines in PHC services. Furthermore the study 251 benefits from using an experimental design, consisting of the implementation of different types of 252 strategies and using a large multi-centric design. In addition, it included a large number of practices, 253 providers, and patients, giving confidence in the findings across five different European jurisdictions. 254 The study does however have some weaknesses; firstly, there is no information about the reasons 255 why professionals did not provide advice to those patients that screened positive or why they did 256 provide advice to those who screened negative. Non-controlled factors may have played an 257 important role in the professional decision-making, such as patients' characteristics, including 258 gender, employment status and level of education as described in previous studies [19]. Secondly, we 259 did not perform a validation of AUDIT-C against any other tools. In previous European studies, 260 researchers have demonstrated discrepancies between the use of two screening and diagnostic tools 261 with fewer than one-fifth of alcohol-dependent cases being identified by two different methods [21]. 262 Finally, PHC centres that took part in the RCT were volunteers and no information is available from 263 those that refused to participate. This might have added a bias in the form of inclusion of PHC 264 centres whose professionals are more motivated in working with drinkers. The challenge is finding strategies that result in high rates of SBI implementation, whilst ensuring that 280 accuracy of screening and advice is also high. The fact that financial incentive was associated with the 281 proper provision of advice to risky drinkers could be significant from a policy perspective as a way to 282 promote the reduction of alcohol consumption and implement public health measures aimed at 283 these professionals. 284 285

Declaration of interest 286
Antoni Gual has received honoraria, research grants and travel grants from Lundbeck, Abbvie and 287 D&A Pharma, outside the submitted work. Other co-authors do not declare conflicts of interest. 288