Think-aloud method

A method that is often used to describe the sequence of thoughts involved in decision-making is the think-aloud technique.10 Subjects are instructed to say their thoughts aloud while performing a task, and the verbal reports are usually audiotaped, transcribed to written form and then analysed. The main objective in using the think-aloud technique is not to judge the outcomes of a participant’s cognitive process as either successful or unsuccessful decisions but, rather, to explore the process of performance.10 11 As our goal was to unravel which arguments, barriers and facilitators play a role in the diagnosis of real-world GP patients with CHF, the think-aloud method seemed well suited to achieve this. For the methodological orientation to underpin the study, we used the constructivist grounded theory methodology as described by Glaser and Strauss.12 13 Constructivist grounded theory is an approach in which researchers generate a theory of a process, action or interaction. This theory development is shaped by or ‘grounded’ in the viewpoints of various participants as well as in the viewpoints of the researchers, as the latter interpret the data.13 The Consolidated Criteria for Reporting Qualitative Research was used as guidance to report our study.14

Ethical considerations

The research ethics committee of the KU Leuven approved the study (mp19078). Before the think-aloud sessions, all participant GPs were asked informed consent based on written information about the aim and methods of the study. There was no remuneration provided for participation in the study.

Patient and public involvement

No patients and/or public were involved in the development of the research question, study design or interpretation of the data.

Participants

The setting of this think-aloud study was general practice in Belgium. All general pratices in Belgium form of the public healthcare system. According to the theoretical sampling method, participating GPs and practices were selected as the analysis progressed for their ability to provide data that would confirm, challenge or expand the emerging theory.15 We aimed to include a representative sample of gender, years of practice experience, practice type and location that were consistent with the reality standards. Initially, three family practices were selected: two urban and one more rural. One of the urban practices was a district health centre that was financed at practice level and receives a fee for each registered patient. The other practices work in a pay-for-performance system where GPs get paid for each patient who consults them. The latter is the most common system in Belgium. After 11 think-aloud sessions in these three practices, we decided to select one more rural practice with GPs not involved in academic teaching or research to guarantee a wide range of GP profiles (table 1). GPs were all approached personally or by email. All approached GPs consented to participate.

Data collection

Since patients with HF are often not registered as such in the GPs’ electronic health record (EHR), we first performed a clinical audit in each EHR to identify patients with possible HF. This clinical audit consisted of the search on a registered (coded or free-text) diagnosis of HF, combined with the search on coded or free-text diagnoses of risk factors for HF, HF symptoms, and signs and combinations of HF medication (online supplementary file 1). The list of all patients with possible HF was then presented to each treating physician and they were asked to judge which patients had HF or not (0/1) and grade how certain they were about the diagnosis (Likert scale with a range of 0%–10%–25%–50%–75%–90%–98%), based on their knowledge of the patient file. Afterwards, four patients of each GP were chosen at random for the think-aloud session: two of each binary code (HF 0/1) and, in each category, one with a high grade of certainty (ie, >75%) and one with a low grade of certainty (ie, <75%). The order of the cases was chosen at random for all participants. The GPs were asked to think aloud about why they did or did not appoint the HF diagnosis in their own real-world patients based on the patient file in the EHR. The think-aloud session took place in the GPs’ own offices. All sessions were audio recorded. The only intervention of the researcher during the think-aloud session was that a participant who was silent for more than approximately 15 s was reminded to say his or her thoughts aloud about the information presented.10 The interviewer made field notes during the think-aloud session. After each think-aloud session, the participant was asked to clarify some thoughts in a follow-up interview. This whole process and the verbatim transcriptions made afterwards were led by one of the authors (PDW), a GP trainee at the time of the study. The interviewer was familiar with two of the (urban) participating practices since he was trained there. He was unfamiliar with the other two practices. As a GP trainee he had a medical background and experience as a GP, hence understanding the medical terminology and barriers associated with HF in general practice. If requested, the GPs first got a test case (not recorded), also selected at random from the audit, in order to get acquainted with the think-aloud method. Then, they continued with their four personal study cases. Data collection techniques were piloted under the supervision of a qualitative research expert (SP). The data of the test performances were not included in the study. Data collection was continued until data saturation was reached. Data saturation was defined as the moment when the last two interviews no longer contributed any new elements and when a certain category had been exhaustively described in all its dimensions and variations. This means that conducting additional interviews would no longer provide new insights.11

Data analysis

The Qualitative Analysis Guide of Leuven (QUAGOL) was held as a guide for the data analysis.16 The QUAGOL is a theory and practice-based guide that offers a comprehensive method to guide the process of qualitative data analysis within the grounded theory approach. It promotes thorough (re)reading, thinking and discussion about the research data before starting the actual coding process.16 All transcripts were entered into NVivo V.11 software (QSR International, Melbourne, Australia), a data management platform, for qualitative data analysis. After two of the authors (PDW and MS—a GP specialised in HF) familiarised themselves with the data by making one-sheet summaries to aid categorisation and conceptualisation, they independently coded each line of text according to its meaning and content. Codes were created inductively. After reading and coding the findings of a sample of transcripts, the two researchers discussed and compared the codes for similarities and differences until a primary coding framework was constructed. Subsequently, the findings of the other transcripts were independently read and coded. The two authors discussed their respective coding frameworks frequently to reach consensus. Codes were added, modified or merged when necessary. This process resulted in a tree structure with several layers for organising the descriptive themes (online supplementary file 2). From these, a set of analytical themes emerged that were discussed by the research team. Our collaborative approach and the iterative constant comparison limited the extent to which individual perspectives or background could dominate our interpretation.13

Step 1: assessing the likelihood of HF

Implicitly, every GP assessed how likely HF was in their patient. To estimate the likelihood, many factors were considered, such as cardiovascular (CV) risk factors, CV diseases and medication, comorbidities, past HF hospitalisations, and HF signs and symptoms. In patients without any or few risk factors, GPs considered HF unlikely (figure 1).

He is 43 and is still a young man. As far as I know, something cardiac never occurred to him. I can also see this in his antecedents list. He never had any complaint that I can link with HF like dyspnoea, oedema, etc. So, actually, I am not 98%, but 100% sure he doesn’t have HF! (GP 3, Think-aloud session)

In patients with a CV history, the presence of HF signs and symptoms was cited as the most important discriminator to distinguish between being at risk for HF and having HF (figure 1). Consequently, for almost all GPs, the clinical assessment determined the further diagnostics process and their risk assessment.

If a patient has symptoms we look further but without symptoms we don’t. (…) In my opinion, the clinical aspect is the most important. For instance, the oedema, the gain of weight and the minimized exercise capacity. (GP 14, Follow-up interview)

However, many GPs recognised barriers in this clinical approach (table 2). It was apparent that almost every GP experienced difficulties caused by overlap in signs and symptoms with comorbidities in these real-world cases.

Yes, of course, there are many overlapping symptoms. She suffers from chronic hypoxia, and this is causing her fatigue. Her limited exercise tolerability, dyspnoea, cough and abnormal lung auscultation can also be caused by this. The core symptoms are overlapping, which makes it very difficult to appoint a clear (HF) diagnosis to someone with chronic obstructive lung disease on clinical grounds. (GP 4, Follow-up interview)

Even without concurrent comorbidities, it remained difficult to assess patients clinically due to the non-specificity of HF signs and symptoms.

She had a little peripheral oedema in summer. Is this linked with HF? I really don’t know. (GP 2, Follow-up interview)

Additionally, many patients already received CV medication, possibly masking HF symptoms and signs.

It could always be possible that he has a slight aspect of HF and because of the ACE-inhibitor he takes it is just masked. That it suffices to control his symptoms. (GP 7, Think-aloud session)

Other barriers mentioned were difficulties to assess the HF risk in immobile patients, which was often the case in this population of elderly; the relapsing remitting course of HF was also a challenge (table 2).

Step 2: considering further diagnostic steps

The decision about further diagnostic steps was influenced by patient and social factors and by the GPs’ attitudes towards HF diagnosis (table 3).

Step 3: choice and implications of further diagnostic steps

When GPs chose to refer, they almost immediately opted for a cardiologist and/or cardiac US referral (figure 1). GPs rarely mentioned ECG or NT-proBNP spontaneously. Influencing factors on how they decided and dealt with these investigations and their results were described in table 4.

Step 1: assessing the likelihood of HF

Every GP assessed the likelihood of HF as a first step. The arguments GPs used in their assessment coincided with the concept of the CV continuum or the HF stages of the American Heart Association guideline. These concepts describe HF as a syndrome that progresses from asymptomatic structural heart disease in patients with CV risk factors to symptomatic HF, making this likelihood assessment a valuable approach.17 18 Additionally, in line with former studies, the assessment of HF signs and symptoms by the GP was seen as the main discriminating factor to withhold a diagnosis of HF and/or to consider further diagnostic steps.7 9 However, at the same time, GPs reported many barriers in this clinical approach, especially in real-world older patients with comorbidities, as confirmed in other studies.5 6 Furthermore, it was shown that 77% of the patients with HF diagnosed in primary care in Belgium are already in New York Heart Association (NYHA) stages III–IV and thus unmistakably symptomatic at the time of diagnosis.19 Consequently, one might suspect that GPs tend to overestimate the value of their clinical assessment, especially in older patients with HF, leading to delayed or missed diagnoses.4–6 20 This provides important points of action to improve the diagnostic process in HF. First, a paradigm shift is needed towards early identification of patients with HF and prevention of disease progression. Early intervention in patients with stage B HF (asymptomatic structural heart disease) led to a reduction in morbidity and mortality.21 Therefore, we would like to emphasise the importance to consider further diagnostic steps when signs and symptoms are present and when substantial risk factors or comorbidities are known (figure 1, curled arrow).22 23 Second, the question arises whether the current definition of HF is applicable in the very old. The definition of HF restricts itself to stages at which clinical symptoms are apparent. As shown by our study, this is particularly difficult in elderly as signs and symptoms lose their value in this age group. Demonstrating an underlying cardiac cause is another essential part of the HF definition. This is especially challenging in patients with HFpEF. A high proportion of patients with HFpEF have concurrent atrial fibrillation while diastolic dysfunction is very difficult to assess in this patient group.24 25 Additionally, some patients only exhibit symptoms (and echocardiographic abnormalities) on exertion.24 25 Conversely, the prevalence of mild to moderate diastolic dysfunction is very high in elderly but it is difficult to decide on the clinical significance of these cardiac phenotypes.6 25 26 In response to these diagnostic problems, the Heart Failure Association introduced a new consensus on the HFpEF diagnosis on the latest European Society of Cardiology (ESC) congress.27 Time will tell whether this new HFpEF definition will resolve all diagnostic doubts.

Step 2: considering further diagnostic steps

Our study revealed patient-related, social and GP-related factors that were not reproducible by case-vignette studies9 or discussed in HF guidelines.1 18 Patient-related and social factors are generally not modifiable; however, GP-related factors are. The importance GPs attach to person-centred care was highlighted. As many GPs mentioned in our study, what matters to them and to the patients is the prognostic and therapeutic implications of cardiac abnormalities. This is an issue that HF guideline developers should consider.1 18 NT-proBNP could support GPs in this risk stratification, as it provides prognostic information in cardiac outpatients.1 3 20 28 29 Additionally, in our study, the concepts of HFrEF and HFpEF were cited by very few GPs. Interestingly, some GPs thought that they could distinguish HFrEF from HFpEF based on the patients’ clinical profile, while cardiac US remains the gold standard.1 18 Therefore, we recommend that GPs have a better understanding of cardiac US reports and propose targeted education as an area of improvement.7 30

Step 3: choice and implications of further diagnostic steps

NT-proBNP and ECG are recommended by all HF guidelines.1 18 Conversely, they were rarely mentioned by GPs in this think-aloud study. In the follow-up interview, the lack of reimbursement for NT-proBNP tests was cited as a significant barrier. In Belgium, the cost of natriuretic peptide testing is relayed on the patient (±€25 per test) due to an impasse in the negotiations with clinical biologists. This accounted for the GPs’ unfamiliarity with the test and the uncertainty regarding how to integrate it in their practice. However, one GP correctly quoted the value of NT-proBNP in the early stages of HF but voiced scepticism towards extended use in follow-up.31 Furthermore, the cost of the test could account, in part, for the fact that 77% of Belgian patients with HF in primary care are already in NYHA stages III–IV at the time of HF diagnosis.19 To withhold ECG as a valuable diagnostic tool, repeated training of GPs to fine-tune their interpretation skills remains important.32 However, as quoted pertinently in our study, a negative ECG excludes HF, but the elderly rarely have a completely normal ECG.1 6 Our study also showed that GPs sought more information on the correct interpretation of cardiac US reports when diagnosing CHF. In contrast with UK studies, there were few practical barriers for cardiologist referrals and cardiac US.7 8 33 Thus, better access alone to cardiac US is not sufficient because a better understanding of and education in interpreting cardiologists’ reports is needed.8 9 Additionally, GPs noted that cardiologists are responsible to describe HF diagnoses clearly in their reports, which, in the case of HFpEF, remains a difficult task.1 24

Implications for practice

The diagnostic flow chart of the ESC HF guideline already promotes using NT-proBNP or BNP as a diagnostic test when patients have a history of ischaemic heart disease, hypertension, cardiotoxic medication or chronic diuretic use.1 Belgian GPs rather use the national GP guideline about CHF published in 2011. Recommendations are generally in line with the ESC HF guideline but the use of natriuretic peptides is not actively promoted in this guideline since the test is not reimbursed in Belgium.34 The paradigm shift to an earlier risk assessment should be more widely disseminated in practice. However, access to natriuretic peptides is indispensable to achieve this.1 22 28 29 The main modifiable barrier for further investigations in this study is not, as formerly described, a lack of access or a pure lack of knowledge. It is a lack of belief in the added value of further investigations. A remaining uncertainty after cardiologist referral contributes to this. Uncertainty remains because cardiologists do not assign HFpEF diagnoses easily themselves, and GPs are not able to correctly interpret echocardiography reports.35 Additionally, GPs are confronted with a high percentage of patients with HFpEF, whose prognosis does not change much with available treatment. Therefore, assigning a diagnosis in this patient group seemed less important in their opinion. Education and guidelines for GPs should target these beliefs because a distinction between HFrEF and HFpEF cannot always be made clinically and a correct HFpEF diagnosis does have important prognostic implications.36

Methodological strengths and limitations of this study

The variability in the years of experience, the nature of the practice and the gender of the 14 participating GPs reflected the general GP population. Therefore, differences in working conditions, access to cardiac US and cooperation with specialists were well covered. However, there were no solo-operating GPs included in the study. This form of practice is declining but is currently still represented in the Belgian healthcare system. In this survey, there was also a higher fraction of participant GPs involved in academic teaching and/or research, explaining our recruitment success rate of 100%. To correct for this imbalance, one extra rural practice with three GPs who were not involved in academic teaching or research was added in the study. As far as we know, this is the first study where a think-aloud approach on real-world patients was performed. This enabled us to highlight the importance GPs attach to person-centred care and to analyse GPs’ diagnostic reasoning with respect to CHF and the differences with the existing guidelines.