Barriers to physician adherence to nonsteroidal anti-inflammatory drug guidelines: a qualitative study

SUMMARY

Background

Despite wide availability of physician guidelines for safer use of nonsteroidal anti-inflammatory drugs (NSAIDs) and widespread use of these drugs in the US, NSAID prescribing guidelines have been only modestly effective.

Aim

To identify and describe comprehensively barriers to provider adherence to NSAID prescribing guidelines.

Methods

We conducted interviews with 25 physicians, seeking to identify the major influences explaining physician non-adherence to guidelines. Interviews were standardized and structured probes were used for clarification and detail. All interviews were audio-taped and transcribed. Three independent investigators analysed the transcripts, using the constant-comparative method of qualitative analysis.

Results

Our analysis identified six dominant physician barriers explaining non-adherence to established NSAID prescribing guidelines. These included (i) lack of familiarity with guidelines, (ii) perceived limited validity of guidelines, (iii) limited applicability of guidelines among specific patients, (iv) clinical inertia, (v) influences of prior anecdotal experiences and (vi) medical heuristics.

Conclusions

A heterogeneous set of influences are barriers to physician adherence to NSAID prescribing guidelines. Suggested measures for improving guideline-concordant prescribing should focus on measures to improve physician education and confidence in guidelines, implementation of physician/pharmacist co-management strategies and expansion of guideline scope.

INTRODUCTION

The routine use of nonsteroidal anti-inflammatory agents (NSAIDs) is commonplace in the US, with 20 million regular users and 111 million prescriptions filled annually.¹ Because of their widespread use, the potential for related complications has been well documented. Side effects include renal toxicity, fluid overload and upper gastrointestinal events (UGIEs), with the latter constituting a significant mortality risk.² The high volume of NSAID prescriptions and the number of complications related to their use have led to development of clinical guidelines to promote safer prescribing of NSAIDs by physicians.^3–5 These guidelines focus on identification of high-risk patient populations and promotion of gastroprotective strategies in vulnerable populations. Guideline recommendations for high-risk patients include concomitant prescription of gastroprotective agents [proton pump inhibitors (PPIs) or misoprostol], or the use of cyclooxygenase-2 selective medications.

Guideline-concordant prescribing patterns have been associated with decreased incidence of UGIE.⁶ Unfortunately, studies have also demonstrated poor provider adherence to these guidelines,⁷ suggesting only modest effectiveness at changing physician prescribing behaviours.^8–10 Cabana et al.¹¹ proposed a conceptual model to explain the process and factors involved in changing physicians’ practices in response to guideline recommendations based on the premise that knowledge would shape attitudes and change in attitudes would then influence provider behaviour. However, Cabana noted that his framework may not be generalizable to all situations as barriers may be unique to each clinical practice setting. The specific barriers to physician adherence to NSAID prescribing guidelines have not yet been described. Thus, we endeavoured to explore physician barriers to the adoption of currently published NSAID prescription guidelines to better understand why guidelines have had limited impact on NSAID prescribing behaviour, and to describe comprehensively the taxonomy of barriers that exist.

METHODS

Study design and population

This study was approved by the Institutional Review Board of Baylor College of Medicine, Houston, TX. Qualitative methods were used to explore the factors that influence NSAID prescribing behaviours of physicians and the specific role of practice guidelines in shaping prescribing behaviours. A sample of physicians recruited from diverse practice settings participated in structured, in-depth interviews. In-depth interviews were the chosen qualitative methodology, given our aim of describing the complex relationship between factors that influence opinions, behaviours, or motivation regarding NSAID prescribing among physicians from diverse specialities.^{12, 13}

One investigator identified individuals who met the study eligibility criteria of being a licensed (nontrainee) physician who routinely cared for patients taking chronic NSAID therapy. Following standard qualitative methods, we purposively sampled participants¹⁴ from a variety of practice settings, including a veterans’ hospital, publicly funded health system, private hospital and academic ambulatory clinic and a diversity of medical and surgical specialities. Purposive sampling identifies cases that will provide the greatest depth, breadth or novel interrelations among variables with the intent of generating new hypotheses, identifying novel modifying or moderating variables or describing the complexity of multiple interrelations simultaneously among variables. We ceased recruiting participants when we reached thematic saturation or when no new concepts emerged from the ongoing analysis of at least five interviews.^{13, 14} Thematic saturation is defined as the ongoing collection and analysis of data until no new information is obtained (no variations or exceptions) and there is a ‘redundancy’ in the theme categories and relationships in subsequent interviews. Our a priori definition of thematic saturation required observation of five consecutive, redundant interviews by all investigators prior to cessation of recruitment.

Data collection and procedures

Interviews were scheduled at the convenience of the physician, based on their availability, to ensure minimal interference with each participant’s schedule. Each interview lasted between 45 and 60 min. Prior to all interviews, written consent was obtained from all study participants. Participants completed a demographic form, and then participated in a one-on-one interview conducted by the same investigator (AW). Interviews consisted of a series of open-ended questions to elicit data regarding knowledge, attitudes and behaviours associated with their NSAID prescribing practices (Table 1). All interviews began with the following opening question:

Table 1.

Interview structure (sample questions)

Gauging general NSAID knowledge

Are there NSAIDs you consider inherently safer than others from a GI standpoint?

When do you decide to recommend gastroprotection for your patients who are taking NSAIDs?

How does anticipated duration of therapy impact your thinking in terms of GI complications?

Impressions of NSAID toxicity

When prescribing NSAIDs, what side effects or complications do you worry about?

What patients would you worry more about than others with respect to complications?

How do you decide which NSAID to prescribe?

Knowledge of NSAID prescribing guidelines

What resources have you used to obtain recommendations?

Are you aware of published recommendations for safer NSAID prescribing strategies?

With which aspects of published recommendations do you agree or disagree?

Do you believe that published recommendations are applicable to your patients?

Impact of NSAID-related adverse experiences

What adverse experiences have you had with patients taking NSAIDs?

Is there a type of patient to whom you would not prescribe NSAIDs?

How does the presence or absence of risk factors for GI complications affect your prescribing practices for individual patients?

We know that NSAIDs are used very commonly for pain relief and for control of acute and chronic inflammation. I’m sure you have prescribed them many times for your patients, as well. When prescribing NSAIDs to your patients, how do you decide which NSAID to prescribe?

Responses to this question were further explored with probes focused on (i) assessment of over-the-counter NSAID use, (ii) gastrointestinal (GI) complications and other adverse effects, (iii) concerns about specific comorbidities, (iv) types of patients for whom one would not prescribe NSAIDs and (v) specific examples of adverse experiences with NSAID prescriptions (Table 1).

A number of strategies were implemented to minimize interview fatigue. First, interviews were structured such that the questions of greatest potential data richness were placed after the introductory or opening questions. Additionally, the interview guide was adapted and re-ordered during data analysis to maximize the richness of novel responses. Secondly, the interviewer was trained to observe for signs of participant fatigue by monitoring subject willingness to provide case examples or in-depth responses. If interview fatigue was detected, the interviewer was trained to focus on the structured probes designed to clarify areas where thematic saturation had not yet been achieved.

Data analysis

Interviews were audio-taped and transcribed for analysis. Data were analysed using the constant-comparative method of qualitative analysis.^12,13 Three investigators (NA, AW, AN) independently reviewed each transcript line by line to identify and sort segments of data with similar concepts into distinct categories, with particular attention to the following content areas: (i) knowledge of complications arising from NSAID use, (ii) attitudes regarding formal and informal guidelines for NSAID use and (iii) predictors and modifiers of NSAID prescribing behaviour. The sorted categories evolved into a coding system, which investigators applied independently to each transcript. Coding occurred soon after each interview and prior to completion of all interviews. After each round of independent coding, investigators convened as a group for careful review, negotiation and consensus building to resolve discrepancies in coding.

As new themes arose, the interview guide was revised to investigate the emerging themes further. With the coding of successive transcripts, the coding system was expanded, refined, and applied to previously coded data.¹⁴ During the last five interviews, no new themes emerged, and thematic saturation was believed to have been achieved.¹³ Recruitment was considered complete at the point of thematic saturation and the final code key was reapplied to each transcript. The resulting categories that emerged from the final coding scheme reflected the dominant themes used for elaboration of study findings.^{12, 13} Illustrative quotations were chosen to provide justification for the definition/basis of themes and potential variables, as is standard practice for qualitative studies.¹⁴ In general, selected quotations were chosen as representative of utterances garnered from interviews, from which no meaningful variation occurred.

RESULTS

A total of 25 physicians were interviewed from January to March 2006. The descriptive characteristics of study participants are shown in Table 2. Interviewees represented a multitude of medical specialties, along with varied practice settings and number of years in practice.

Table 2.

Demographics and practice characteristics (n = 25)

Baseline characteristic	Frequency (%)
Age* (years)
<35	6 (26.1)
36–44	11 (47.8)
45–55	4 (17.4)
56–64	2 (8.7)
Gender†
Male	11 (45.8)
Female	13 (54.2)
Race†
Caucasian	16 (66.7)
African-American	4 (16.7)
Other	4 (16.7)
Speciality
Internal Medicine	16 (64.0)
Cardiology	1 (4.0)
Rheumatology	1 (4.0)
Geriatric	1 (4.0)
General Surgery	1 (4.0)
Physical Medicine/Rehabilitation	2 (8.0)
Infectious Disease	2 (8.0)
Nephrology	1 (4.0)
Average years in practice [s.d.)	16.0 (10.2)
1–5	3 (12.0)
6–10	6 (24.0)
11–15	7 (28.0)
16–20	2 (8.0)
21–25	1 (4.0)
26–30	3 (12.0)
31–35	3 (12.0)
Practice setting
Academic – VAMC	10 (40.0)
Academic – County Hospital	8 (32.0)
Academic – Private Hospital	7 (28.0)
Patient-care responsibilities (%)
100	3 (12.0)
75–99	7 (28.0)
50–74	10 (40.0)
25–49	5 (20.0)
Interaction with pharmaceutical representatives
Daily	3 (12.0)
>3 times per week	1 (4.0)
1–2 times per week	6 (24.0)
1–2 times per month	12 (48.0)
No interaction	3 (12.0)
Role in setting formulary	15 (60.0)
Scope of practice
Out-patient only	6 (24.0)
In-patient only	3 (12.0)
Both	16 (64.0)

^*Two participants did not respond.

^†One participant did not respond.

Dominant themes

Interview analysis identified three dominant themes and six major barriers to physician use of published guidelines for safer NSAID prescription. These included (i) barriers related to knowledge: (a) lack of familiarity with guidelines, (ii) barriers related to perception: (b) perceived limited validity of guidelines, (c) limited applicability of guidelines among specific patient populations and (iii) barriers related to prior experiences: (d) clinical inertia, (e) anecdotal experiences and (f) clinical heuristics.

Barriers related to knowledge

Lack of familiarity with guidelines: Physicians frequently noted being unfamiliar with published guidelines (Table 3). In our study, 21 of 24 participants (84%) acknowledged being unaware or unfamiliar with NSAID prescribing guidelines. The lack of familiarity was often attributed to the overwhelming number of published medical guidelines and difficulties in keeping abreast with more recent recommendations. Other physicians acknowledged they were simply unaware that specific NSAID guidelines were available.

Table 3.

Barriers related to knowledge

Lack of familiarity

‘…[there are] tons of different places making guidelines and, knowing this, sometimes it can be difficult to be aware of every single one out there.’ [MD 16]

‘There are very well-defined guidelines on cholesterol, ypertension, yes, and diabetes, but not for GI use of NSAIDs.’ [MD 17]

‘…you can’t really predict which one (patient) is going to have an ulcer or a complication much all on your own, so there are actually no particular guidelines…to my knowledge it hasn’t been developed.’ [MD 1]

‘…There are guidelines for all the statins and for the drugs for hypertension and for diabetes…so when you come down to guidelines for NSAIDs…that’s lower in my priority list than all the other guidelines.’ [MD 18]

Barriers related to physician perception of guidelines

Perceived limited validity of guidelines: A substantial number (60%) of participants voiced concerns about accepting the recommendations (Table 4). When probed, physicians admitted their sentiments were rooted in the belief that many guidelines originate from unreliable sources. Author credibility was cited as an important influence in providers’ willingness to accept recommendations. In some instances, the guidelines were felt to lack author credibility because of concerns that guideline authors lacked clinical experience. For others, the feeling of being pressured to follow guidelines endorsed by major medical societies was uncomfortable.

Table 4.

Barriers related to physician perception

Limited validity

‘There were guidelines that were…the opinion of some expert and then there were guidelines that were truly based on the evidence…the guidelines become like reading the medical literature: the buyer beware.’ [MD 10]

‘…many of them, including some of the really basic ones like the JNC (Joint National Committee) and hypertension …are not evidence-based documents…99% of what we’ve ever been told is semi-processed drug-company propaganda and physicians are just tired of hearing about it.’ [MD 22]

‘…there are probably a myriad of reasons that people don’t follow guidelines…sometimes guidelines are viewed as being developed by a group of folks in an ivory tower who absolutely have never taken care of a patient. ‘ [MD 14]

‘I think the other thing that’s a barrier to influencing practice in a positive way is the organizations that adopt a guideline and then turn it into a dogma: Everybody must follow this guideline.’ [MD 10]

‘…so many organizations label something or another as a guideline, and…I think that’swhy physicians have a certain detachment from guidelines…the evidence is overwhelming.’ [MD 22]

Limited applicability

‘…guidelines need to fit within the context of all the various disease parameters, all the other diseases that are going on, which often the guidelines don’t take into account.’ [MD 10]

‘It’s kind of cookbook medicine…like somebody comes in and hands you the map; you do this and that, and you give this and that…but…people are not cookbooks and they always have other factors that you need to consider.’ [MD 11]

‘I always like to use our formulary, so the patients get [medications] for free.’ [MD 3]

‘…because we were dealing with a restrictive formulary…I can usually get around not prescribing it (Celebrex) if I don’t think it’s warranted.’ [MD 2]

Limited applicability for specific patients: Almost all participants expressed concern regarding the applicability of guidelines in their own practice (Table 4). Physicians frequently expressed difficulty in applying generic guidelines to specific patients, in particular, the elderly and patients with multiple comorbidities. The logistics of applying guideline recommendations to specific populations was also noted. The recommendation to provide gastroprotection to high-risk patients was observed to be a barrier, particularly among patients who are treated in environments with restrictive formularies, or for patients who have financial constraints. Often, these external constraints influenced a provider’s decision to follow guideline recommendations among specific high-risk patient populations.

Barriers related to prior experiences

Clinical inertia: Clinical inertia refers to physician failure to deviate from established practice patterns, and is typically rooted in physician ignorance, personal disagreement with guideline recommendations or overestimation of the quality of current medical care they provide (Table 5). Established practice habits, including prescribing patterns, were well described by most participants. Initiating change with regard to long-standing habits was acknowledged as a difficult task. If a routine for PPI gastroprotection was established, clinical inertia with this routine would dictate the degree of provider adherence to recommended NSAID prescribing guidelines. Physicians also recognized the role of clinical inertia in staying current with practice guidelines.

Table 5.

Barriers related to prior experiences

Clinical inertia

‘A new drug comes out. Well, there’s a certain amount of work that I’m going to have to do, mental work…before I’m comfortable using the drug, and if I’ve got an alternative that’s almost as good, well then … the incremental benefit I may gain does not outweigh the work I’m going to have to do to get over that.’ [MD 10]

‘…there’s an inherent reluctance to change on the part of just about every physician…the older physicians, you’re always going to have some degree of resistance because they’ve been practicing a certain way for so long.’ [MD 24]

‘… It’s on the formulary. I know the dosing. It’s routine. It’s habit. I know that; I know that they’re protected when I put them on a PPI.’ [MD 12]

‘I know they’re there. I want to read them (guidelines), but more than likely what I’ll end up doing is going to a new conference and wait(ing) for somebody to tell me…’ [MD 5]

Anecdotal experiences

‘Well, I learned to ask about the gastritis and peptic ulcer disease because I gave it to a little old lady who came back with black-coffee stools, and her daughter said, ‘Yeah, it’s just like last time when she had gastritis before.’ That’s when I learned, you always ask…’ [MD 3]

‘I have always been a cautious doctor…so that I can’t, right off the top of my head, tell you of anybody that I know that I tipped into renal failure…or that I caused to have a GI bleed. But no one has ever gotten back to me with a major crisis that I created.’ [MD 13]

‘I was never a big fan of Vioxx, just because I tried it once myself, and it gave me a GI (bleed). And Celebrex, same thing. Bextra never bothered me. So I actually have a lot of arthritis now and use those things.’ [MD 3]

Medical heuristics

‘You certainly see people who can have a GI bleed with short-term use (NSAIDs), but I think the longer you use them, and the more you use them, the more it’s going to occur. I don’t know whether that’s actually true or not. I think it is, but it just seems so.’ [MD 8]

‘If I’m just going to give it (NSAID) to somebody for, you know, a brief 7-day course or something like that for an acute issue, I still give them the caveat of taking it with food to protect their stomach, but I’m not going to generally prescribe anything to protect the stomach.’ [MD 5]

‘I give gastroprotective agents to anybody who’s taking NSAIDs regularly. If they tell me they just take it every now and then, then I just tell them to take it with food.’ [MD 12]

‘I don’t have any problems giving them ibuprofen, high dose, 800 mg for that short period of time, so I think in the short-term setting that that’s very safe, taken with food.’ [MD 25]

Anecdotal experiences: A very powerful influence on a physician’s uptake of guideline recommendations proved to be the individual experiences of that physician (Table 5). The presence or lack of clinical feedback, whether adverse or not, was highly influential and had a major impact on prescribing practices. Participants often stated that their inability to recollect negative outcomes related to current prescribing patterns reinforces their prescribing strategies. If they could not recall a negative outcome, then a negative outcome did not occur. The personal experiences of physicians, whether positive or negative, also created lasting impressions.

Medical heuristics: Also prevalent among the participants was the use of medical heuristics to guide decision-making (Table 5). Heuristics are clinical ‘rules of thumb’ used to guide medical decision-making where no established guidelines exist. These heuristics are often derived from physicians’ assessments,personal opinions and clinical experiences. For many, heuristics are simply rooted in lore.¹⁵ Frequently, adoption of a heuristic led to deviation from NSAID prescribing guidelines.

DISCUSSION

The purpose of this study was to understand better why guidelines have not had a greater impact on NSAID prescribing behaviour and to describe comprehensively the barriers to physician adherence to guidelines. Six major barriers stemming from three major themes were identified and the results demonstrate that a heterogenous set of influences is the driving force behind guideline-discordant prescribing behaviour.

Lack of familiarity with guidelines was the most frequently cited barrier by participants. Physicians’ lack of familiarity with guidelines has been shown to influence a number of clinical outcomes including control of hypertension, promotion of exercise stress testing, colon-cancer screening, smoking cessation and asthma compliance,^{11, 16–18} resulting both in under- and overtreatment of the medical condition.^{16, 19} In our study, many physicians voiced concerns that their unfamiliarity with published guidelines often resulted in non-evidence-based prescribing patterns.

Among physicians who were aware of the existence of NSAID prescription guidelines, reservations regarding the recommendations were motivated by the perception that guidelines were too strongly influenced by pharmaceutical companies and major medical societies with ‘secondary agendas’. The potent influence of pharmaceutical companies in academia, particularly in clinical trials, has been previously described^{20, 21} and can sometimes extend to the design and development of practice guidelines. Choudhry et al. described the interactions between guideline authors and pharmaceutical companies, with numerous authors receiving financial support or employment, but few disclosing these relationships.^{11, 22}

Among our respondents, the perception also existed that current NSAID prescribing guidelines were insufficiently ‘evidence-based’. This perception among physician providers may not be unfounded. A number of published guidelines have been shown to vary in terms of quality of evidence and interpretation of trial data. For example, guideline recommendations for treatment of heart failure and hypertension were shown to differ significantly, despite referencing the same randomized controlled trials,^{23, 24} Nuckols et al.²⁵ demonstrated that, despite rigorous development standards for major musculoskeletal guidelines, most omitted common clinical scenarios and had diverging interpretations of literature when an expert panel reviewed them. This documented discordance among guideline developers validates these physician concerns.

Many interviewed physicians cited concerns about applying generic guidelines to their own patients. Participants often expressed concern that guidelines could not accommodate the complexity of physicians’ unique patients. In addition to impeding guideline implementation, comorbidities affect guideline quality. McAlister et al. used a guideline-appraisal tool to evaluate management guidelines for three major cardiovascular risk factors and found only 45% of guidelines to be of high quality. The most frequent reason for a low quality rating was lack of applicability of guideline recommendations to patient populations with significant comorbid conditions.²⁶

Among all physicians, the impact of prior clinical experiences proved a powerful influence. Both negative and positive outcomes related to NSAIDs were influential in shaping current prescribing strategies. A single negative outcome was enough to have a lasting impact (i.e. availability bias) and positive outcomes reinforced current practices without regard for base rates of adverse events (i.e. base-rate fallacy). Schwartz et al.²⁷ found that physicians cited experience with ‘real-world medicine’ as having more utility than the results of clinical trials, further highlighting the impact of personal experience.

Reluctance to deviate from established practice also raises the issue of clinical inertia. Most often, inertia is related to one of several factors including physician ignorance (being unaware of guidelines that recommend different practices), knowingly deviating from guidelines because of disagreement or personal reasons and overestimation regarding the effectiveness of current care.²⁸ This study identified unique challenges including physician reluctance to learn and apply practice guidelines. Physicians were content with their current treatment strategies and were often discouraged by the amount of effort needed to adopt new prescribing habits.

Published medical literature, although wide in scope, is unable to guide every decision in medical practice.¹⁵ The use of clinical heuristics or cognitive rules of thumb is thought to guide day-to-day medical decisions.²⁹ Heuristics are employed by clinicians to simplify medical decision-making in scenarios when the course of action is not clearly described by the available data.³⁰ Their utility has been demonstrated in complex and time-constrained scenarios, when they serve to condense relevant information and streamline the decision-making process.³¹

Most heuristics provide guidance for situations in which evidence-based data are lacking. Studies assessing the quality of heuristics have advocated critical evaluation of established heuristic models to assess their validity.¹⁵ It is reasonable that, if certain heuristics are deemed valid, their use may not necessarily be precluded. However, in our study, the most frequent heuristics pertained to clinical decisions regarding NSAID dose, duration and gastroprotection – all areas well-addressed by current published recommendations for safer NSAID prescribing. The clinical heuristics used by physician participants often conflicted with the recommendations of these guidelines and often resembled the cognitive biases that distort clinical judgement like availability bias and base-rate fallacy, as described above.³²

Improving physician adherence to NSAID prescribing guidelines may require dissemination and implementation strategies that are better tailored to individual practice settings. Active barriers should be identified; and specific measures designed to address these barriers should be implemented. The respondents in our study expressed the desire for streamlined educational workshops to improve familiarity with guidelines.³³ Implementation of a structured guideline development process that is strongly rooted in evidence-based medicine significantly improves the quality of the generated recommendations. The RAND/University of California at Los Angeles Appropriateness Method, and the Delphi model have been shown to improve the validity of the guideline development process and assuage physician concerns regarding the validity of guideline recommendations.^{34, 35} In addition, adequate disclosure of thirdparty influences is essential to ensure credibility. The commonly cited concerns of our respondents regarding potential conflict of interest highlight the importance of full disclosure of the role pharmaceutical companies may play in guideline development. In addition, there is a need for transparency regarding existing financial relationships between guideline panel members and industry, to alleviate concerns regarding the credibility of guideline recommendations. Finally, the importance of recognizing and combating clinical inertia was highlighted in our study. Although traditionally difficult to overcome, some suggested methods for combating clinical inertia and biased or invalid medical heuristics include physician–pharmacist co-management strategies, as well as audit-feedback and academic detailing techniques.^36–38 The effectiveness of these strategies in promoting physician adherence to NSAID prescription guidelines is yet to be fully explored.

Our study sheds light on the common perceived barriers to physician adherence with published NSAID prescription guidelines; however, limitations of this study must be acknowledged. First, these findings are based on responses from practicing physicians recruited from academic hospitals in one geographical area. Consistent with qualitative methodology, we used purposive sampling to identify a broad representation of clinical specialties, training backgrounds and practice settings within this geographical area. For this study, we used a hybrid of ‘maximum variation sampling’ (to document important common patterns that cut across variations) and ‘confirming and disconfirming sampling’ (which identifies cases to elaborate and deepen initial analyses, then seek exceptions and sources of variation).³⁹ We chose this sampling methodology to identify all physicians who may be regular prescribers of NSAIDs. We began with the most common case (an internist) to elaborate our initial findings and seek saturation of themes among that group. We then sampled potential variations (i.e. other specialties that prescribed NSAIDs routinely) to identify novel themes or potential exceptions or variations from the already expressed themes. At times, we enrolled only a single case from a particular specialty because (i) he/she only confirmed existing themes and data without providing meaningful variation in the data (i.e. responses were exactly the same as the other specialists), or (ii) he/she was the only willing participant in the local area from that specialty.

Second, the small sample size and qualitative methods are not designed to generate quantitative estimates of the observed patterns and themes. Our objective was to generate hypotheses about why guidelines for safer NSAID prescribing are not used in clinical practice and to identify relationships between the barriers to guideline implementation and the reasons for persistence of inappropriate NSAID prescribing behaviours. The number of physicians recruited was dictated by the methodological desire to achieve thematic saturation. As per our a priori definition of thematic saturation (i.e. attained after observation of five consecutive redundant interviews), subject enrolment became even more deliberate to ensure that the widest ranges of responses and types of participants (within the parameters of the study aims) had been recruited and interviewed. Thematic saturation was achieved after recruitment of a relatively small number of providers, as is often observed in qualitative studies using in-depth interviews. It is unlikely that the recruitment of additional participants would have yielded new information, given that no new themes were elicited among the final five participants. Finally, interviews were scheduled in advance, so it is possible that some physician responses may not have been entirely candid. However, respondents were not told the purpose of the interview to limit the potential for information bias.

Several strengths of our study are noteworthy. To our knowledge, this is the first study that identifies determinants of physician NSAID prescription, with a focus on barriers to guideline adherence. The qualitative design provides a robust illustration of how barriers to guideline adherence create a vacuum by which inappropriate NSAID prescribing behaviours persist. All physician interviews were conducted using an accepted and validated methodology for conducting qualitative in-depth interviews based on the principles of grounded theory.^{12, 13} The standardization of each interview minimized the possibility of interviewer bias and information bias. Finally, the dominant themes were derived from participant-generated ideas following an iterative process of independent coding by three separate investigators using the constant comparative method of qualitative data analysis.^{12, 13}

CONCLUSION

In this qualitative study, we have clarified the major barriers to physician adherence with published safe NSAID prescribing guidelines. Dominant themes driving noncompliance with guidelines included difficulty with guideline familiarity, applicability and validity; as well as clinical inertia, influence of anecdotal experiences and use of heuristics. Suggested measures for improving guideline concordance include use of educational workshops and continued medical education to improve guideline familiarity, implementation of physician/pharmacist co-management strategies, expansion of guideline scope and overall efforts to improve physician confidence in the validity and the quality of guideline recommendations. Future studies are required to assess the effectiveness of these strategies in clinical practice.