Informants were selected to represent clinicians who regularly encountered patients with dyspeptic symptoms in their daily practice and who thus were expected to have a treatment policy for these patients. We therefore invited Swedish clinicians who were specialists in medical gastroenterology, surgery, and general practice to participate in this study. All clinicians gave their consent to be interviewed. They also suggested other clinicians who might be interested in participating in the study. Altogether, 27 clinicians participated, nine from each of the three specialities. This number is the number commonly used in studies that use judgement analysis to ensure a variation in answers 12.

The specialists in medical gastroenterology (four women and five men) all worked at Lund University Hospital at the time of the study. The surgeons (one woman and eight men) worked at Lund University Hospital or Helsingborg Hospital. The general practitioners (two women and seven men) worked in the Public Health Service in the southern part of Sweden. The clinicians were 35–56 years (average 47 years) of age and their experience in their speciality ranged from 1 to 22 years (average 10 years).

A useful approach for studying clinicians' strategies is judgement analysis 14, which describes the cognitive process involved in making a decision. It focuses on the actual decision made by the clinician and on which information, for example symptoms and clinical findings, that the clinician uses to reach that decision.

In our study, each of the clinicians was interviewed separately. The interviews were based on four case histories. The histories described frequent symptoms and findings in 30-year-old men seen in general practice and in specialist out-patient clinics (Table 1). Besides age and sex, all histories contained information on the localisation of symptoms, but information on subjective symptoms and findings from the examination varied from history to history. The questions, listed in Table 1, focused on how the clinician viewed whether there was a need for treatment, what management the clinician would suggest, which factors the clinician thought were most important to consider, and what decisions the clinician thought colleagues in their own speciality or in other specialities would make. The same questions were asked for each patient. The case histories were presented in the sequence described in Table 1. Thus, the questions were answered four times by each informant.

One of the authors (KK) with experience in judgement and decision analysis interviewed each informant at her or his office. Each interview began with the interviewer reading the first history. The interviewer then asked the questions aloud and wrote down the informant's answers after each question. If the informant wished, the history and questions were reread. At the end of the question-and-answer session after each case history, the interviewer read aloud the participant's answers and in cases of ambiguity, adjusted the answers. The interviews thus took the form of an oral questionnaire, but with the possibility to explain the answers more fully.

The analysis was performed after all data had been collected from the informants. Two of the authors (KK, MT) with experience in qualitative studies read the interview notes. In a first step, each of the clinicians' answers was scrutinised to find content or evaluations in common, which were then coded as concepts. The following is an example of answers coded as a concept: To question A ("Do you think that it is a problem to decide whether there is a need for treatment in a case like this?" and if yes: "What problems?") informants gave answers like "there is no demonstrable disease, but the patient has symptoms" and "there is no diagnosis but a suffering patient" (patients 1 and 2). These answers were coded as the concept "The patient suffers, but there is no distinct diagnosis".

In the second step, related concepts were combined into categories, common to all the case histories. The authors compared their coding and resolved any differences through discussion to reach a consensus. The concepts and categories were compared with the original answers to ensure that the answers were covered by the classifications.

The four case histories in this study are representative of patients with conditions frequently seen in medical practice. Diseases of the gastrointestinal tract account for about 6% of all reported consultations in Sweden 13. Gastritis, dyspepsia, and unspecified diseases in the ventricle and duodenum (patients 1 and 2) are diagnosed in 1.6% of the population; oesophagitis and reflux (patient 3) in 0.9%; and different ulcers (patient 4) in 0.6%.

The clinicians who participated in this study were not representative of a randomised selection of clinicians in Sweden but rather of a select group with a special interest in patients with symptoms from the gastrointestinal tract. Since their experience with these patients was extensive, they should have been able to develop a treatment policy. One would expect a variation in treatment strategy in this group to be more limited than among a randomised selection of clinicians.

The results underpin existing evidence that variations in medical practice exist 141516. This study presents variations both in treatment strategy and in what information is considered important for making treatment decisions. The histories of patients with diffuse symptoms but no objective findings (patients 1 and 2) gave rise to more extensive variations than did the case histories of patients 3 and 4 where the symptoms were more obviously related to a diagnosis (patient 3) or where more obvious organic changes existed (patient 4). Uncertainty in diagnosis could lead to uncertainty regarding which outcome of alternative interventions is optimal 1718. A meta-analysis that evaluates the most effective treatment in patients with functional dyspepsia 10 recommended the eradication of H. pylori if the treatment is to be effective from the patient's perspective, whereas other randomised, double-blind, controlled studies find that eradication has no beneficial effect 11. These results were underpinned by the variations observed in this study.

Uncertainty due to lack of knowledge or professional competence may lead to qualitative differences. In this study, different management strategies were recommended for patients with reflux disease (patient 3), for example, eradication treatment as one strategy and changes in life-style as another. However, evidence in the literature presenting eradication as an optimal management of reflux is lacking 17. Instead, this management strategy raises the cost to society and to the patient and causes unnecessary antibiotic pressure. Randomised, double-blinded, prospective studies conclude that the only indication for treating dyspepsia with proton pump inhibitors (PPIs) is the presence of an ulcer or reflux 7. However, the majority of clinicians prescribed PPIs and H₂-receptor antagonists for the treatment of dyspepsia, although no evidence of an effect with this treatment compared to placebo has been documented 910. Thus, this regimen has led to high costs for society without any benefits. The drug costs for treatment of ventricular and duodenal ulcers and of reflux, including eradication treatment, were SEK 1.6 billion [USD 210 million] in 1998 in Sweden 13, while The Swedish Council on Technology Assessment in Health Care reports that society's direct and indirect costs for dyspepsia were between SEK 3.7 and 4.4 billion [USD 490 and 590 million] in 2000 3.

From the informants' answers to the question "Which factors are most important to consider in your decision?", the completeness of some of the answers given in the interviews could be questioned. Most likely, the clinicians failed to answer the question in full as they left out many factors worthy of consideration. Other factors in the patient's life such as stress, mental mood, working situation, and diet were not mentioned by the clinicians in the interviews but should often be considered in treatment strategies in an actual clinical situation.

Within each of the three specialities, clinicians believed that their colleagues would treat patients in the same way as they themselves did. This aspect of "professional certainty" implies that clinicians believe that their practice is correct, irrespective of how much it in fact differs from that of other clinicians 1819. This unawareness of variation in the management of frequently seen patients may indicate a lack of communication and discussion about everyday cases. Such discussions are perhaps reserved for more "complex" and rare cases.

Among the causes of variation in medical practice, the influence of factors like patient characteristics (e.g. age, sex, morbidity, and personal preferences) could be more or less regarded as legitimate to explain variations in practice 20. Other factors, like resource capacity, could be influenced by, for example, budget restrictions while management policy and practice style are more resistant to change 20. These latter factors should therefore be the targets of efforts to change. However, many attempts to implement evidence by information alone, for example in the form of clinical practice guidelines, have failed to change management strategies 20212223. Instead, a combination of methods is most likely needed if a permanent change is to occur 23. Furthermore, it is of utmost importance that potential barriers are identified and that the clinicians who will be affected support the clinical practice guidelines to be implemented.

In recent years, the individual autonomy of patients and letting patients' preferences influence the choice of intervention have been emphasised 24. In this study, such a strategy could explain why some clinicians mentioned that they felt pressured by patients to perform an intervention, even when the optimal strategy was non-intervention.