Specialist feed-back in addition to a self-audit facilitated by peer-led academic detailing did not increase anticoagulant prescribing in Australian general practice. We addressed GPs’ concerns about specific cases and detailed medical histories informed specialist feedback. This is in contrast to traditional methods for providing feedback using aggregated statistics and benchmark targets. Yet this novel method was not effective in promoting anticoagulant prescribing.
Comparison with other studies
Audit and feedback and academic detailing do not consistently promote evidence-based practice [13, 14] and our findings are in step with other evaluations of these strategies undertaken in a variety of contexts. Results are also consistent with findings from another study evaluating a multifaceted intervention for anticoagulation prescribing in AF, incorporating academic detailing and audit with feedback as we have done . Moreover, three secondary stroke prevention trials did not report increased rates of anticoagulant prescribing for patients in AF in response to structured discharge care plans or nurse-led care [19, 31, 32].
Frequent feedback, using more than one delivery format, implementing action plans and providing targets enhance the effectiveness of audit and feedback . While feedback here was delivered twice, verbally and in writing, we did not include action plans or specific targets and this may have underpinned the null findings. While feedback delivered by stroke specialists was hypothesised to be persuasive, the influence of experts may have been limited because they were not otherwise known to GPs nor involved in the care of their patients. This finding is consistent with results from another community-based study evaluating written, evidence-based summaries provided by opinion leaders, showing no improvement in the management of heart failure or ischaemic heart disease .
The intervention focussed on GPs while patients were not engaged. A recent study reported a small (3%) but significant increase in anticoagulant prescribing attributable to a computerised decision support tool activated during interactions between physicians and patients , consistent with results from an earlier study employing a computerised risk assessment tool to guide real-time decision making . In contrast, another recent study found no effect of an intervention using email notifications reporting patient-specific stroke risk and guideline recommendations for antithrombotic treatment . Evidence from other randomised and non-randomised trials suggests simply having access to support tools, such as risk calculators, electronically generated feedback reports and automated prompts in anticipation of patient visits does not impact prescribing rates [37,38,39]. Taken together, findings from this and our own research imply that interventions delivered outside the context of the consultation are ineffective, yet providing opportunities for GPs to engage patients in discussions during real-time decision making may drive changes in prescribing behaviour.
The selection of patients by GPs in whom anticoagulation has been challenging may have made change especially difficult to achieve as there was no meaningful differences in anticoagulation rates from baseline to post-test in either group. The intervention therefore can be considered ineffective for patients selected in this way and the results may not generalise to all AF patients in general practice, for example, patients newly diagnosed with AF who are naïve to anticoagulants or those with less challenging needs. The impact of expert decisional support did not differ according to baseline patient characteristics, including baseline stroke risk. Therefore, there was no evidence that the impact of the intervention would have differed in patients with varying stroke risk. We acknowledge, however, that complex needs not reflected in the CHADS2 score, such as a patient’s social and psychological circumstances, their cognitive capacity and adherence to medication may work against efforts to promote anticoagulation. While we did not systematically assess such barriers, these factors may have been relevant for the GPs. Moreover, there was no change in either group from baseline to post-test.
Implications for current practice
Our study was carried out as guidelines were evolving away from use of the CHADS2 score to the CHA2DS2-vasc score for risk stratification although US guidelines continued to recommend the use of CHADS2 sore in 2012 . We had incorporated the CHA2DS2-vasc scheme to ensure GPs were aware of impending changes to local guidelines and feedback from experts also utilised this scheme where relevant. According to current recommendations , more AF patients are identified as suitable for anticoagulation, including those younger than 65 years of age. In this study, around 90% of patients were eligible for anticoagulation using the CHADS2 risk stratification scheme and therefore would have also been eligible under the CHA2DS2-vasc scheme. The main impact of the change in guidelines would have related to the 57 patients under 65 years of age in our study.
Antiplatelet medication has been largely superseded for the management of AF and recommendations now advise anticoagulation is to be preferred even in relatively low risk patients . Real-world data indicates that a significant proportion (~ 15%) of patients with AF are prescribed antiplatelets,  and a recent study demonstrates no change in the proportion of patients prescribed aspirin since the change in guidelines , highlighting that current day treatment for AF remains sub-optimal for a sizeable proportion of patients.
The study was carried out before DOACs became widely available and at a time when their subsidised release under Australia’s universal medical insurance scheme was imminent. Reluctance to prescribe warfarin may be intransigent. DOACs are perceived to be more convenient  with a proven lower risk of intracerebral haemorrhage  and these characteristics may leverage wider uptake of anticoagulation. Expert decisional support may have greater traction in the current post-DOAC era as DOACs may be an “easier sell” if GPs are more pre-disposed to using these drugs over warfarin. However, our results are consistent with several randomised and non-randomised evaluations of implementation strategies for reducing stroke risk in AF which mainly report null findings or small differences in the uptake of anticoagulants in the pre and post-DOAC era [32, 36,37,38,39]. Taken together, these findings indicate that shifting prescribing practice remains challenging.
The recently published multicentre, multi-national IMPACT-AF study conducted in low to middle-income countries, incorporated both patient and GP education and closely monitored individual patients over a 12-month period . The study reported a 12% increase in anticoagulant prescribing over one-year compared with a 3% change in the usual care control arm. The results of IMPACT-AF may owe its success to persistent and sustained contact with both GPs and patients and the availability of DOACs. However, almost 80% of patients on anticoagulants were receiving warfarin at post-test, indicating a continued role for warfarin.
The study findings can be applied to current practice where warfarin is considered and continues to be used and, more generally, if the barriers to warfarin use also impede the uptake of DOACs. In Australia and elsewhere, DOAC dispensing has increased concomitantly with decreasing warfarin, although warfarin remains commonly prescribed [44, 47,48,49,50,51,52,53,54,55], and recent surveys show that warfarin is favoured amongst a significant proportion of clinicians , particularly for patients in whom use is established. In Australia, reports indicate that around one-quarter to one-third of initial anticoagulant prescribing is for warfarin [48, 49, 54]. There is also growing evidence that DOAC prescribing behaviour is influenced by the same perceived barriers that limited the wider use of warfarin indicating that DOACs do not appear to have bridged the evidence-practice gap as initially hoped and are subject to the same perceived limitations of warfarin. Recent studies report a high proportion of uncoagulated patients, for example, 32%  and 57% , and evidence also demonstrates that doctors remain cautious about bleeding risks, leading to non-evidence-based DOAC use. One large multicentre study of general practice patients showed that 40% of patients with AF and 40% of those at a high risk of stroke were not receiving anticoagulation despite trends towards increasing DOAC use . Moreover, DOACs have been preferentially prescribed to younger patients, those with lower bleeding risks, fewer comorbidities and/or with lower stroke risk [50,51,52,53]. This has also been shown to occur in Australia, where patients with lower comorbidity burden are preferentially prescribed higher dose DOACs . The underuse of anticoagulation in patients over the age of 85 remains significant . In one analysis of over 100,000 anticoagulated AF patients, DOACs were associated with a higher risk of ischaemic stroke and shorter prothrombin time compared with warfarin use , suggesting under-coagulation in patients using these newer agents. Under-dosing is common [47, 59,60,61,62] and has been associated with an increased risk of thromboembolic events and mortality, although findings have not been consistent [47, 59,60,61,62]. Contraindicated use or use inconsistent with recommendations is another concern [60, 61]. In an Australian study of AF patients admitted to hospital, 40% were judged to have been prescribed a DOAC inappropriately. Specifically, DOACs were contraindicated in around one-in-eight patients, with both under-dosing and overdosing reported in patients (~ 15 and 10%, respectively) .
Moreover, discontinuation of all oral anticoagulants is prevalent. In a study analysing data from over 45,000 newly-anticoagulated patients in the United States, 50% receiving Apixaban, 65% receiving Dabigatran, 57% receiving Rivaroxaban and 71% receiving warfarin discontinued treatment within 1 year of initiating it and no more than 10% of patients switched to another agent . In another large-scale study of over 64,000 patients, only 47% of those taking DOACs adhered to treatment for more than 80% of the time at a median of 1.1 years of follow-up, compared with 40% of those taking warfarin  while in another report based on over 62,000 patients newly diagnosed with AF, discontinuation of DOACs at 3 years was 46% . These results underscore that stroke risk reduction in AF remans ripe for knowledge translational activity and, contrary to expectations, barriers to wider and appropriate anticoagulation persist in contemporary practice.
DOACs were heralded to bridge the evidence-practice gap in AF management. Wider use of DOACs will increase the absolute numbers of patients who may experience a bleeding event if such patients would not have previously been anticoagulated given their risk of an adverse event would have been lower without any anticoagulation. If GPs encounter more patients experiencing adverse events, this may negatively reinforce their prescribing, resulting in more cautious future use and less evidence-based prescribing. These altering perceived risks may potentially impact on prescribing intentions, motivating GPs to reduce bleeding risk (or an act of commission) at the expense of reducing stroke risk (an act of omission) . For these reasons, our results may inform future studies evaluating strategies to increase anticoagulation uptake.
Strengths and limitations
The large number of GPs involved in this study is a strength. Limitations include the unblinded self-reported outcomes and the 15% attrition rate of participants. Losses to follow-up were non-differential between groups and any bias would have likely favoured the intervention. Further, the three-month scheduled time between delivery of the intervention and post-test assessment may have limited the opportunities for GPs to change prescribing practices as recalling patients for a consultation was not specifically encouraged. Specialists to whom GPs referred patients may have been predominately responsible for determining treatment and GPs may have been reluctant to deviate from and contradict those decisions. The “control group” of the study was an active arm. Consequently, we do not know whether either expert decisional support, academic detailing or self-audit minimised discontinuation of anticoagulation compared to usual care. We do not have insights into GPs’ views about the intervention as we did not conduct qualitative interviews at the end of the study. Specifically, it would be of interest to understand the reasons GPs did not change prescribing practice despite expert recommendations. We did not ask GPs whether they or another doctor had initiated treatment and do not know to what extent this contributed to the null findings. These issues are an important focus for future research.
GPs were advised to select patients for whom management presented challenges. This made case selection less costly in comparison to an independent audit. Only half of selected patient cases were receiving anticoagulation, even though 90% had a risk of stroke justifying anticoagulant use, characteristics which are consistent with those reported by others [4, 7, 8]. We saw evidence of decisional conflict with one-in-five GPs indicating difficulties in weighing up the benefits and risks of anticoagulation with one-third disagreeing that warfarin was the best choice for patients. These findings indicate that GPs selected appropriate cases for this knowledge translation exercise and self-audit may be relied upon in future research.
The selection of patients for whom management was considered challenging may have skewed the cohort towards a group of patients with more complex needs. More than two-thirds of patients had stroke risk warranting anticoagulation and 70% had a stroke risk assessed as “high” according to their CHADS2 score (that is, a score of two or more); fewer than 10% were deemed at low risk of stroke. The distribution of stroke risk is skewed towards higher risk compared with registrants in multicentre registries , although it is comparable to another knowledge translation Australian trial of AF patients recruited from general practice . An audit of 82 Australian general practices carried out at the time this study was conducted reports just over one-third of patients having high risk of stroke compared with 70% here, although, like here, fewer than 10% were considered low risk .
The study cohort is best described as one of motivated volunteers as only 3% of GPs initially contacted eventually participated in the study. In the context of a randomised evaluation, however, comparisons should be internally valid. The national scope of the research and our limited resources precluded more resource intensive recruitment strategies, such as those involving follow-up phone calls, site visits and repeated mailings which would have likely increased participation. The characteristics of the selected cases are in keeping with previous studies [4, 7, 8] and the proportion of GPs in full-time practice was consistent with national GP workforce statistics. Low participation rates amongst GPs in research studies is not an isolated experience. In an implementation trial carried out in the United Kingdom aimed at improving anticoagulant prescribing in AF, Holt et al.  recruited 8% of 570 general practices invited to participate via a mailed letter. Two Australian cluster randomised trials carried out in general practice have reported a 6% and an 8% participation rate amongst GPs located within the researchers’ local areas which were predominately urban [67, 68]. We utilised a study base of a random sample of GPs which could be expected to result in a lower participation rate compared with enrolling participants from a registry of self-nominated GPs interested in research.
To address the challenges in recruiting GPs, we deliberately approached a large number of potential participants and carried out two “Studies Within a Trial” (SWATs) evaluating various response-aiding strategies  at the time of initial contact. Interest in carrying out SWATs is gaining momentum to generate high quality evidence to improve trial administrative processes  and our findings will generate evidence to inform improved recruitment in primary health care implementation research.