A systematic review of factors affecting intended and actual adherence with antiviral medication as treatment or prophylaxis in seasonal and pandemic flu

The aim of this review was to identify factors predicting actual or intended adherence to antivirals as treatment or prophylaxis for influenza. Literature from inception to March 2015 was systematically reviewed to find studies reporting predictors of adherence to antivirals and self‐reported reasons for non‐adherence to antivirals. Twenty‐six studies were included in the review; twenty identified through the literature search and six through other means. Of these studies, 18 assessed predictors of actual adherence to antivirals, whereas eight assessed predictors of intended adherence. The most commonly found predictor of, and self‐reported reason for, non‐adherence was the occurrence of side effects. Other predictors include perceptions surrounding self‐efficacy, response efficacy and perceived personal consequences as well as social influences of others' experiences of taking antivirals. Predictors identified in this review can be used to help inform communications to increase adherence to antivirals in both seasonal and pandemic influenza.


| INTRODUCTION
In April 2009, a new strain of influenza virus was detected in humans (influenza A(H1N1)pdm09). The illness, commonly referred to as "swine flu," was first identified in Mexico and spread rapidly around There are many reasons why people may choose not to adhere to their antiviral medication as prescribed. Research into medication adherence has suggested that reasons for non-adherence include social and economic factors (such as social support, family/ caregiver factors and socio-economic status), therapy-related factors (such as the presence of adverse effects, duration of treatment and drug effectiveness), patient-related factors (such as age, gender and education) and condition-related factors (such as the presence of symptoms and disease severity), amongst others. 6 To the best of our knowledge, there is no systematic review describing factors affecting adherence to antiviral medication for influenza specifically.
Given that adherence is a complex behavioural process, 7 the successful implementation of interventions that promote adherence requires a thorough understanding of the factors associated with that behaviour. The Capability, Opportunity and Motivation (COM-B) framework was developed from existing theories of behaviour change. 8 In essence, the framework hypothesises that the interaction between capability (C), opportunity (O) and motivation (M) causes the performance of behaviour (B) and can provide a framework for understanding why a desired or recommended behaviour is not performed.
Capability is defined as the individual's psychological and physical capacity to engage in the activity and includes having the necessary knowledge and skills; Opportunity is defined as all the factors that lie outside the individual that make the behaviour possible or prompt it; and Motivation is defined as the mental processes that energise and direct behaviour, not just goals and conscious decision-making but also habitual processes and emotional responses. The COM-B approach has recently been applied to medication adherence. 9 It is able to account for a wide range of factors affecting adherence and to inform behaviour change interventions that can be used to guide healthcare practitioners involved in the care of non-adherent patients.
To inform policy regarding the distribution of and communication about antivirals as treatment or prophylaxis in seasonal and future pandemic influenza, we conducted a systematic review to investigate factors associated with adherence and non-adherence to antiviral medication for influenza. The outcome measures of this review included predictors of adherence and non-adherence, and self-reported reasons for non-adherent behaviours.

| METHOD
The review was conducted in accordance with the PRISMA guidelines, 10 using systematic methods to identify and select studies, and assess their risk of bias. No formal protocol exists for this review.

| Search strategy
We searched EMBASE, MEDLINE, PsycINFO and the Web of Science Core Collection from inception to 26 March 2015, using combinations of terms relating to antivirals (e.g. antiviral, Tamiflu, oseltamivir, Relenza, zanamivir), influenza (e.g. pandemic, influenza, H1N1, H5N1) and adherence (e.g. adherence, uptake, compliance). There was no limit on publication date imposed. Where possible, databases were also searched using MeSH headings. We also undertook reference tracking to identify further papers for inclusion. Other articles were located through previous non-systematic searches carried out by members of the research team.

| Inclusion criteria
Eligibility criteria for inclusion in the systematic review were as follows: Participants: Studies were included if they asked people whether they took antivirals as either treatment or prophylaxis for influenza or whether they intended to take antivirals as treatment or prophylaxis for influenza. Participants could be drawn from the general public, patient groups or specific occupational groups.
Predictors/Exposures: We only included studies if they assessed demographic or psychosocial predictors of adherence to antivirals in the context of influenza or if they assessed self-reported reasons for adherence or non-adherence to a course of antivirals.
Outcomes: Studies were included if they reported actual or intended adherence to antivirals, which were intended as either treatment or prophylaxis for pandemic, avian or seasonal influenza.
Study reporting: All study designs, aside from those published only as conference papers, editorials or abstracts, were included. For pragmatic reasons, we only accepted papers published in English.

| Data extraction
For every included paper, we tabulated details relating to the author, date of publication, influenza virus, country, sample size, methodology, adherence with antivirals, length of course of antivirals, and reasons for, and predictors of, non-adherence.

| Data synthesis
Where possible, we grouped study results together depending on whether they related to actual vs intended adherence, antivirals prescribed as a treatment vs antivirals prescribed as a prophylaxis and pandemic (influenza A(H1N1)pdm09) vs non-pandemic situations. In practice, it was not always possible to make these subdivisions due to a limited amount of data being available.
A meta-analysis of the data was not planned, because we expected that the literature would be too heterogeneous. Instead, we carried out a narrative synthesis of the data. No general consensus exists on the best methodological approach to the narrative synthesis of data. 11 In this review, we created a comprehensive list of predictors of actual and intended adherence to antivirals that have been studied to date.
Effect sizes were not synthesised.

| Risk of bias
Risk of bias was determined according to an adaptation of the Scottish Intercollegiate Guidelines Network (SIGN) critical appraisal methodology checklist for cohort studies 12 and supplemented by relevant items from the Cochrane Collaboration's Risk of Bias tool. 13 Table 1 shows the criteria used in the assessment.

| Procedure
LS and VJ developed and conducted our literature search. Studies were screened by LS, and data extraction, assessment of risk of bias and data synthesis were carried out independently by LS and DD with advice from JP, GJR and JW. Any discrepancies were resolved through discussion.

| Study characteristics
Eighteen articles assessed actual adherence to antiviral medication, while eight assessed intended adherence. The studies spanned thirteen countries (Australia, Brunei, Canada, Cyprus, Hong Kong, Israel, Japan, Madagascar, Netherlands, Norway, UK, USA and Singapore) and included samples of healthcare workers, military personnel, pregnant or recently pregnant women, poultry farmers, end-stage renal patients, school children and staff as well as the general public. All but three studies investigated pandemic influenza (actual influenza A(H1N1)pdm09 or hypothetical scenarios), with other studies investigating influenza H5N1, 15 H7N3 16 and seasonal influenza. 17 The majority of studies were cross-sectional, with one RCT, 18 and two qualitative studies. 19,20 All but one study 18 investigated adherence with antivirals via self-report. All studies investigated predictors of actual or intended adherence to oseltamivir; no studies included in the review investigated predictors of adherence to zanamivir (trade name "Relenza") or other antivirals such as amantadine. Tables 2 and   3 provide information about the samples and methods of the included studies, which addressed actual and intended adherence, respectively. Tables 4 and 5 provide adherence rates in each study, together with self-reported reasons for, and predictors of, actual and intended nonadherence. Where reported by individual studies, numerical data for significant predictors of adherence are stated in Tables 4 and 5.

| Actual adherence
In studies investigating actual adherence with oseltamivir, adherence ranged from 30% to 88.8%. 38 Most studies reported a high overall adherence in the range of 70%-89%. Eight studies investigated adherence to antivirals only as a preventative measure (six in the context of influenza A(H1N1)pdm09), three investigated adherence to antivirals only as a therapeutic measure (two in relation to influenza A(H1N1) pdm09), five investigated predictors of adherence to antivirals as both treatment and prophylaxis (all in reference to influenza A(H1N1)pdm09), and two studies did not report the reason for antiviral prescription. The study addresses an appropriate and clearly focused question

| Antivirals as prophylaxis
When antivirals were prescribed as prophylaxis for influenza A(H1N1) pdm09, the only demographic predictor of adherence was sex, with male students being more likely to take oseltamivir than female students. 39 Other predictors included an earlier time in the pandemic, 18 country of study (higher adherence in students studying in Singapore than those studying in the United States), 39 previous compliance with other precautionary advice about pandemic flu, 39 beliefs that the recommended preventative measures were necessary 39 and having discussed taking oseltamivir with someone who had not experienced side effects. 30 Where antivirals were prescribed as prophylaxis for avian influenza (H7N3), having had direct contact with infected poultry was a significant predictor for adherence to antivirals. 16 Where self-reported reasons for non-adherence to antivirals as prophylaxis for influenza A(H1N1)pdm09 were stated, occurrence of adverse events was the most commonly reported reason for the discontinuation of oseltamivir. 18,30,31,38 Fear (cause unspecified) was also given as a reason for discontinuation of oseltamivir in end-stage renal failure patients, 38 as was not wanting to take it. 38 For those taking oseltamivir in conjunction with their work, there being no direct contact with the virus was given as a reason for non-adherence. 30 Forgetting to take oseltamivir 31,38 and losing tablets 38 were also given as reasons for discontinuation.
Where self-reported reasons for non-adherence to antivirals as prophylaxis in non-pandemic situations were presented, the presence of adverse events was given as a reason for discontinuation of antivirals for influenza H5N1 15 as was not receiving enough tablets. 15 For those taking oseltamivir as a result of avian influenza outbreak (H7N3) at work, not being in direct contact with the virus was a reason given for discontinuation of oseltamivir. 16

| Antivirals as treatment
No predictors of adherence were found when antivirals were prescribed only as treatment for influenza A(H1N1)pdm09, nor were any statistically significant predictors of adherence found in nonpandemic situations.
Where self-reported reasons for non-adherence to antivirals as treatment for influenza A(H1N1)pdm09 were stated, occurrence of adverse events was the most commonly reported reason for the discontinuation of oseltamivir. 24,25 The fear of developing side effects was also sufficient to stop people from beginning oseltamivir. Perception of clinical improvement was given as a self-reported reason for stopping treatment with oseltamivir. Self-reported reasons for non-adherence to antivirals as treatment in non-pandemic situations (seasonal influenza) included the presence of adverse events. 15

| Antivirals as treatment and prophylaxis
One study identified taking a lower, prophylactic, daily dose rather than a higher, treatment daily dose as being a predictor of adherence to antivirals for influenza A(H1N1)pdm09. 36 Two studies investigating antivirals prescribed both as treatment and as prophylaxis for influenza A(H1N1)pdm09 identified significant predictors of adherence, but unfortunately the studies did not differentiate between predictors of adherence to antivirals as treatment vs prophylaxis. In these cases, the presence of adverse effects, 29 age, with higher adherence to antivirals for influenza A(H1N1)pdm09 in those aged <40, and the presence of adverse effects were identified as predictors of adherence.
Self-reported reasons for discontinuation of antivirals for influenza A(H1N1)pdm09 in studies that did not differentiate between treatment and prophylactic use of antivirals included the presence of adverse events 29,36 as well as the perception that antivirals were not effective. 23,29 People also stopped taking oseltamivir because they had medical advice to stop taking the drug, 36 forgot to take it 28,29 and because they had difficulty swallowing the tablets. 36 The perception that there was a greater risk of developing side effects after having taken oseltamivir than there was a risk of catching influenza A(H1N1) pdm09 23 also stopped people from beginning to take antivirals for influenza A(H1N1)pdm09.

T A B L E 4 (Continued)
Studies which did not specify at all the reason why antivirals were prescribed as treatment or prophylaxis for influenza A(H1N1)pdm09 also identified fear of developing side effects as a reason not to begin oseltamivir and perception of clinical improvement as a reason for stopping oseltamivir.

| Intended adherence
Studies investigating predicted adherence with antivirals found an intended adherence rate ranging from 43.5% 20 to 85.5% (Phern-Chern, personal communication). All studies reported intended adherence to antivirals in pandemic situations, but one also investigated hypothetical seasonal flu situations. 20 All studies found significant predictors of adherence. Female sex was associated with higher intended adherence to antivirals for oneself 22 and intention to divert antivirals to family members. 35 Older age was also predictive of

| DISCUSSION
Identifying predictors of adherence can help to inform areas which can be targeted to improve adherence further still. Attempts to identify demographic predictors of adherence have given mixed results, with male sex being found as a predictor for actual adherence in summer school students 39  There were no major qualitative differences between predictors of adherence to antivirals as treatment or prophylaxis for influenza. Any minor differences are likely because prophylactic courses of antivirals tended to be longer than treatment courses, with one study in particular indicating a 20-week prophylactic prescription period. 18 Due to the lack of major qualitative differences between predictors, results for predictors of adherence to antivirals as treatment and prophylaxis are discussed together in the context of the wider literature. Furthermore, only three studies investigated predictors of actual adherence to antivirals and self-reported reasons for discontinuation of antivirals in non-pandemic influenza situations. [15][16][17] The results of these three studies did not differ qualitatively for pandemic and non-pandemic influenza and results from pandemic and non-pandemic studies are therefore also discussed together.
The most commonly found predictor of actual and intended adherence, and reason given for non-adherence, was the incidence of adverse side effects, with studies indicating that the fear of side effects was enough to stop people from beginning a course of oseltamivir. Members of the public may perceive the incidence of side effects from antivirals to be considerably higher than is actually the case. According to clinical trial data, the most common side effect of oseltamivir as a treatment for influenza is nausea (without vomiting) affecting 10% people, with the most common side effect of oseltamivir when used as prophylaxis being headache, affecting 18% people.

Risk of bias
Bults et al. 21 Perceived risk, feelings of anxiety, behavioural responses Probably/certainly intend to take antiviral  1.93, P<.05).

B, C
Divert antivirals to family (all/some).  Risk perception, including perceived severity of and vulnerability to the outbreak, was also found as a predictor of intended adherence to antiviral medication. When communicating with the public, the perception of the overall severity of the outbreak should not be overlooked, as it influences the adoption of behavioural changes. 44 It has been suggested that during the pandemic, the general public were complacent and passive, making them vulnerable to influenza A(H1N1)pdm09 through their lack of uptake of safety behaviours. 45 In terms of motivation, perception of the risk of catching influenza, perception of the severity of influenza, perceived personal consequences of influenza, attitude towards the prevention of influenza beliefs associated with the necessity of precautionary behaviours and concern about the safety, side effects and effectiveness of the pharmaceutical recommendation could be classified as reflective motivation.
Automatic motivation comprises emotions of fear (towards antivirals themselves and side effects elicited by antivirals), and higher anxiety levels, as well as previous engagement in precautionary behaviours including previous acceptance of the influenza vaccine. Social opportunity includes predictors such as trust in the government and speaking to someone who had previously experienced side effects when taking antivirals and physical opportunity incorporates losing the pills or not having enough pills. The psychological capability aspect of the COM-B model is reflected in forgetting to take the pills, whereas the physical capability aspect could be seen in having difficulty swallowing them.
Predictors that are less easy to categorise within the COM-B model were also identified. One study identified time in the study period as a predictor of adherence to oseltamivir with participants being more adherent in the first 10 weeks of the study than in the second 10 weeks. This is possibly due to changing perceptions of the risk of contracting flu as the pandemic season continued and concern about the effects of long-term prophylaxis (both linked to reflective motivation in the COM-B model) as well as the increasing burden in everyday life of continuing to take oseltamivir prophylaxis daily (linked to physical capability in the COM-B model). Higher adherence with lower daily doses of oseltamivir was also found as a predictor of adherence and could be due to the perception that taking less of a drug is safer and better tolerated, but this speculative interpretation should be treated with caution.

| Limitations of the reviewed literature
Overall, the studies reviewed left room for improvement in methodological rigour. Of the 26 included studies, nine showed high risk of selection bias, fifteen showed high risk of detection bias, fourteen showed high risk of reporting bias and seven showed high risk of other sources of bias (see Tables 4 and 5). Studies of particular concern had small sample sizes (<60 participants), 18 Although we did not systematically search for rates of adherence to antivirals and therefore any interpretation of adherence rates should be taken with caution, it is interesting to note that in some cases intended adherence rates to antivirals were lower than actual adherence. This is surprising as although people may intend to carry out a particular health behaviour, they do not necessarily always do so; this is termed the "intention-behaviour gap". 48 However, an explanation for this pattern may lie in the populations of participants assessed in the studies reviewed. Many of the studies reporting actual adherence to oseltamivir were conducted in specified populations who were prescribed oseltamivir as part of their job, either to protect themselves from catching influenza (e.g. poultry farmers 15,16 ) or to protect already at-risk patient populations (e.g. healthcare workers 18,30 ). Other studies included populations of people who were already ill or at-risk from influenza (e.g. renal patients 38 ) or who had been in close contact to a confirmed case of pandemic influenza (school studies 23,29,31,33,36,39 ), with all participants enrolled in these studies having already been prescribed antivirals.
Conversely, three studies investigating intended adherence to oseltamivir were conducted in samples of the general population in different countries (representative samples of the Netherlands 21 and United States 22,27 ). Amongst these participants, there will be a large proportion of people for whom antivirals would not be prescribed in the hypothetical scenarios set out by the studies, or who may fit into "at-risk" populations but who may be unaware of the recommendations relating to antivirals. 49,50 There may also be a number of people who display symptoms of flu, but who do not present to health care providers. Such people would not be identified or included in research investigating actual adherence rates to antivirals. In this way, studies investigating actual adherence in the subset of patients who present to primary care may overestimate likely adherence in the general population.
The difference between adherence defined as collecting and then finishing a course of antivirals, and adherence defined as completing a course once the first tablet has been taken, can also be demonstrated by one of the papers in this review. 39 In this study, all students attending a summer school were offered oseltamivir without cost. Of these students, only 47.6% students picked up oseltamivir, of whom 70.2% took the medication. This means that 33.4% of students took oseltamivir, a much lower number than an adherence rate calculated from only those who picked up the drug.

| Limitations of the review
To the best of our knowledge, this is the first systematic review investigating predictors of antiviral adherence. One particular strength of this review is that studies were not limited only to influenza A(H1N1) pdm09, but also included predictors of adherence of antivirals during seasonal influenza, and so can help inform routine communications with the public.
The limitations of this systematic review are inherently bound to the methodologies of the original research studies reviewed. Studies used very different methodologies to assess adherence and predictors of adherence to antivirals in a wide range of participant populations.
In addition to this, differences in definitions of adherence and length of recommended use of antivirals (spanning 5 days to 20 weeks) between studies further limited interpretation of results. Although the presence of these factors has broadened the results, with predictors related to time within the study being identified, the influence of these factors made comparisons between studies difficult. Due to this wide variation in studies reviewed, meta-analysis of results was not appropriate. The poor reporting of effect sizes for individual studies was also problematic.
Another limitation of the review is that the neither the grey literature, nor unpublished papers were searched for pragmatic reasons, nor did we undertake forward citation tracking; articles not written in English and conference abstracts were also excluded. This increases the likelihood that some research into predictors of adherence to antivirals has been overlooked. The nature of publication bias, in which studies finding a statistically significant result are more likely to be published, suggests that any literature that we missed is unlikely to have found significant predictors of adherence.
One further thing to consider when interpreting the results of this review is that there was a wide variety of populations included, from poultry workers, to renal patients and that results were not subcategorised according to population type. Had these further distinctions been made, there would have been few articles in each group.

| CONCLUSIONS
The most consistently found predictor of actual or intended adherence and reason given for non-adherence to antivirals was the incidence of side effects. Other predictors can help to inform future strategies to increase adherence to antivirals. In particular, increasing knowledge about the risks of influenza, correcting misperceptions about side effects and putting into perspective the risks of antivirals may help to increase future adherence to antivirals.