Ethics statement

This study was approved by the ethic committee of the University de Bretagne Occidentale. The participants provided their written informed consent to participate in this study. The ethic committee approved this consent procedure.

Research team

The research team included two FPs (MB and PB) with academic backgrounds and two female trainees (NM and AC) in family practice doing their Master's degree. After being coached in interview techniques, the two trainees conducted all the interviews. The theme of pulmonary embolism was cited during the call to request an appointment.

Participant selection

Two groups of participants were selected. The first group consisted of 14 FPs who had referred a patient to the emergency unit of the local hospital in the area of Brest, Brittany and where a pulmonary embolism was ultimately diagnosed. Data of all the cases of pulmonary embolism from June to November 2011 were collected. The patient's FP was identified where he/she had been the physician who had sent the patient to the emergency unit. We had no information on the reason why the physician referred the patient to the emergency unit when we interviewed him/her. We undertook the FPs' interviews a few days after the positive diagnosis of PE. The first consecutive 14 FPs who agreed to participate made up the first group. A second group of FPs was chosen using a purposeful sampling method. The aim of selecting a second group was to include the widest possible range of perspectives, experiences, points of view, and, in particular, to enrich the sample with FPs from rural areas who had not referred their patients to the regional hospital. Sixty FPs were approached through a phone call during which the theme of the interview was presented. Fourteen agreed to participate and they made up the second group. Reasons to decline included prior engagements and lack of time. Information on participant profiles is detailed in the “results” section.

Data Collection

The research team developed the interview form with a topic guide, drawn up to answer the research question (see table 1). It was composed of open-ended questions for exploration and closed questions to refine the participants' answers. The first question focused on the case report of a hospitalized patient for the first group and a recollected consultation about a positive diagnosis of PE for the second group. In order to match what emerged from the interviews, we added new questions to the interview guide when participants raised aspects which had not previously been mentioned [15]. All the interviews were audio-recorded. The recordings were transcribed and checked. All the interviews took place at the FP's office. In order to improve validity and credibility, all transcripts were returned to participants for member checking. The duration of the interview was between 5 and 40 minutes. We considered data saturation achieved when no new code emerged from the analysis of the verbatim accounts. It occurred after the twelfth interview in the first group and after the ninth interview in the second group.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Topic guide for the interviews in both sampled groups of family physicians (FPs).

https://doi.org/10.1371/journal.pone.0098112.t001

Data analysis

A thematic analysis was conducted using the technique of constant comparison, originating from grounded theory [16]. The first stage: open coding was done by two researchers (AC and NM) working independently without any framework for the written data. After the first stage, they shared their results. Any discrepancies between the two researchers were discussed with a third member of the research team (MB) until a consensus was reached. Through an iterative process of constant comparison, an axial coding framework was developed at the second stage. The axial coding involved linking categories found within the open coding. The same procedure of working independently before sharing the results was applied. The codebook was revised, with the 3 researchers going back to the data until mutual consent was reached. QSR N vivo 10.0 Software was used to perform the analysis.

Sample

Participants.

Characteristics of the participants and practices are summarized in table 2. One of the FPs interviewed had been a mentor to one of the trainees during the training period.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Characteristics of the 28 FPs interviewed for data collection.

https://doi.org/10.1371/journal.pone.0098112.t002

Analysis

Analysis of the text fragments resulted in 65 open codes, which after an inductive interpretation and categorization process could be structured in 16 axial codes and 3 main categories (table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Themes and axial codes.

https://doi.org/10.1371/journal.pone.0098112.t003

Key points

Main results

The suspicion of pulmonary embolism arose out of four considerations: the absence of indicative clinical signs for diagnoses other than PE, a sudden change in the condition of the patient, a gut feeling that something was wrong and an FP's experience of previously failing to diagnose PE. The FPs interviewed did not use rules in their diagnostic process.

Strengths and limitations of the study

As far as we know, this is the first study describing FPs' diagnostic reasoning processes in cases of pulmonary embolism. The FPs interviewed revealed their diagnostic errors, and sometimes recounted dramatic stories of failure to diagnose PE. This openness during the interview confirmed that the ambiance created by the interviewers was appropriate. The data were based on real life experiences and not on general opinions.

The first group of FPs interviewed was composed of FPs who had had a positive diagnosis of PE. We did not interview FPs who had missed a PE. In accordance with the objective of the study, we focused on the diagnostic reasoning process in cases where PE was suspected. Some FPs did not think about a PE, but about another serious disease which needed emergency care. In the second group, one FP had never even suspected a PE. We decided to include these interviews to broaden the analysis.

Another limitation of this study is the recall bias related to the FPs recruited in the second stage. We did not present a specific case to start with and sometimes their stories had occurred months or years before. In the first group the FPs did remember the cases very well which made it much easier for them to describe their diagnostic reasoning.

Two young family practice trainees carried out the interviews, which could have influenced the content of what FPs talked about. The FPs interviewed knew the study was conducted by young FPs wanting to know what occurred in primary care, rather an intern or pulmonologist who might be judgmental. However, a more hierarchical, rather than collaborative, relationship might have developed between the young interviewers and the experienced physicians. This situation may have influenced the answers of one FP who had previously been the tutor of one of the interviewers. This factor might also explain some reactions to the question about rules.

The sample was selected using two recruitment procedures. The purposive sample allowed us to interview rural FPs, female FPs, and younger FPs (see table 2). It reinforced the maximal variation of the sample.

None of the FPs interviewed used either prediction rules or the point of care D-dimer test although it is the recommended strategy according to international guidelines. Our aim was to describe what was done in real French practices, and not in an ideal situation. Relying on gut feelings in the first stage of the diagnostic process does not exclude the usefulness of clinical decision rules such as the Well's rule. On the contrary, gut feelings should trigger the next clinical process and especially the use of rules such as the Well's rule. The fact that the FPs interviewed revealed their real practice which, although far from what is recommended, is nevertheless another illustration of the openness achieved during the interview.

Key points

The diversity of the clinical pictures of PE did not provide a foundation on which FPs could base their decisions. FPs interviewed in this study, attached importance to the presence of symptoms of thrombosis and patient anxiety. They mainly emphasized the absence of objective and indicative signs in parallel with patients' complaints. A sudden change in the condition of the patient was considered as the most important indication. This point had already been raised, in cases where coronary heart disease was suspected [17], as a reason for referral of patients with chest pain [18], and in cases of meningococcal infection in children [19]. Background knowledge about the patient and person-specific discrepancy were tools used where a serious condition was suspected. The results of our study contribute to the improvement of a specific strategy in primary care: knowing the patient, his risk factors and being sensitive to a discrepancy in the patient's behavior seemed to be decisive for FPs when clinical signs were vague but serious conditions were suspected.

The FPs also based their decision on what they felt during the consultation. That uneasy feeling experienced by the FPs interviewed matched the sense of alarm described in the “gut feeling” concept [20]. This sense of alarm implies that an FP worries “about a patient's health status, even though he/she has found no specific indications yet”; it is a sense that ‘there's something wrong here’” [20]. Their sense of alarm involves specific management whereby the patient has to be referred to an emergency unit, or to a specialist to prevent serious health problems. The gut feeling concept was originally formalized from statements raised in family medicine in the Netherlands and in Belgium [21]. Its transculturality was proven in a Romance language [22]. The FPs interviewed in our study affirmed the existence of the sense of alarm in the specific situation of a suspected PE. Van Den Bruel et al. found that a gut feeling had a higher specificity than a clinical impression in the context of serious infection in children in primary care. The authors recommended that this gut feeling should not be ignored in diagnostic reasoning but has to be used as a red warning flag [23]. One factor which contributed to gut feeling was the perception of parental concern. In our study FPs who had recently diagnosed PE, attached importance to the anxiety expressed by the patient. They felt that this anxiety was indicative of PE whereas other FPs considered the anxiety to be the result of the dyspnea. In our study the patient's anxiety and the sense of alarm perceived by the FPs steered the diagnostic reasoning process.

The FPs interviewed did not use the Wells rule during their diagnostic process because this prediction rule was not considered a helpful tool for detecting PE in primary care. The Wells score, combined with a qualitative D-dimer test, safely and efficiently excludes PE in primary care [11]. This procedure provides a concrete way to deal with the suspicion of PE depending on whether the probability is high or low: if the score is ≤4, the probability of PE is low and a D-dimer test is required and if the score is >4, the probability of PE is moderate to high and further investigations in hospital (compression ultrasonography, pulmonary vascular imaging) are required [24]. The Wells rule brings a synthesis between clinical symptoms, clinical signs and the physician's assessment with the allocation of three points to the physician's clinical judgement of whether PE is more likely than an alternative diagnosis. The physician may express his feeling of alarm with these three points. This first stage in the suspicion of PE, including the sense of alarm, should drive the FPs forward to the second stage of the diagnostic process using the Well's prediction rule. This strategy has a clinical impact in the FP's decision and fits the criteria of a relevant rule for decision-making in primary care [25]. How can we explain this non-use of a safe decision strategy by the FPs interviewed in this study? The use of the point of care D-dimer test is not widespread in France: all the FPs interviewed had to refer their patients to an independent laboratory to have the blood sample analysed. Most of the older FPs had never heard of the Wells rule; indeed, even if the younger ones had learnt how to use it during initial training at the university, they had not used the scores in the cases they related. All the FPs interviewed agreed about their willingness to use their global evaluation of the case, guided by their knowledge of the patient. For them, using a score did not fit the patient-centered approach [26]. This is consistent with one of the main aspects of the definition of family practice where the patient-centered approach belongs to the core competencies of the discipline [5]. Clinical impression, global empirical clinical assessment (also called “gestalt”) are the tools used to qualify a holistic rather than an atomistic approach based on clinical context [27]–[29]. Lucassen et al. compared the predictive value of the gestalt and clinical decision rule when used in combination with D-dimer testing for excluding pulmonary embolism [10]. The sensitivity of gestalt was similar to clinical rules (Wells, Geneva, Revised Geneva) but its specificity was lower [10]. Using a point of care D-dimer test combined with the Wells rule was useful in reducing false positives [10]. The FPs interviewed apparently gave priority to sensitivity rather than specificity. This may be explained by the fact that, in the context of low incidence serious diseases [30], the FPs appear to attach greater value to true-positive decisions (correct decisions to provide care to patients who need it) than to true-negative decisions (correct decisions to withhold care from patients who do not need it). Providing care for patients whom they suspect may be suffering from a life threatening disease, is possibly considered much more important than withholding care from a patient in good health [25].

Moreover the FPs interviewed told us about diagnostic errors they or their colleagues had made when diagnosing PE and how the sense of alarm popped up from their experience and knowledge. This description of the sense of alarm fits into the Reason's model of error prevention in decision making [31]. The experienced FPs function as expert decision makers: as a result of their experience, they have established a set of spontaneous reactions to patterns of diseases they identify at a glance [32]. These rapid, effortless and unconscious ways of thinking are named heuristics: they are considered as powerful tools [33] but with specific errors [34]. When the FPs are confronted by a new situation they have to call upon slow and demanding analytical reasoning. The “uneasy feeling” perceived by the FP corresponded to a perception of cognitive dissonance: the mismatch situation with pre-established patterns triggered the sense of alarm in the FP [32]. The perception of alarm compelled the physician to quit his routine-based reasoning to an analytical reasoning by generating and considering the PE hypothesis [35], [36]. The sense of alarm acted as a feedback mechanism, allowing the questioning of a possibly wrong decision at a very early stage of the diagnostic process. This feeling is based on medical, experiential and contextual knowledge. The traumatic experiences FPs narrated during the interviews are images which are tagged with negative affect in their memories [37]. This shortcut was named the affect heuristic [38]. In the situation where PE was suspected, we may hypothesize that this affect heuristic emphasized the FP's sense of alarm, forcing him to switch to analytical reasoning. The sense of alarm was used here as a tool to prevent the diagnostic error of missing a PE.

Implications for practice and future research

The accuracy of the sense of alarm needs to be evaluated in the context of PE. The data were collected in a qualitative retrospective manner. We cannot generalize from our results without testing them in a prospective way on a large sample of situations in primary care. A short questionnaire has been developed to determine gut feelings in primary care settings [39]. We are aiming to use this questionnaire to study the predictive value of the sense of alarm when confronted with dyspnea and/or thoracic pain at the office. The objective is to compare the results to the questionnaire, filled in at the end of the consultation, with the diagnostic outcome four weeks later. Before implementing the gut feelings concept in educational programs, we need to study its accuracy in real settings.