This validation study was nested in a large multicentre observational study exploring the management of thoracic pain in primary care [9
] in which a random sample of patients seeing a GP for chest pain was included. The early diagnostic guess, which occurred prior to the patients' history being taken, was then compared to the final diagnosis obtained after one year of follow-up.
In patients with chest pain, we measured the prevalence of cases for which physicians had an early diagnostic guess, which was confirmed by further investigations and a one year follow-up. Furthermore, we examined if this predictive ability was influenced by previous encounters with the patient, previous manifestation of similar chest pain, the severity of the illness and by the patients' anxiety, age, and gender.
Fifty-eight general practitioners (GP)working in private practice and six working as residents in an academic primary care outpatient department in Western Switzerland volunteered to participate to this study (Table ). For practical reasons, the six supervised residents were grouped under one common code.
General practitioners' characteristics
Patients sixteen years of age and over reporting any type of chest pain during the first minutes of their visit were consecutively enrolled. The presence of chest pain was ascertained according to the usual practice of each GP. Chest pain due to obvious causes such as trauma or known body metastases was also included and was not necessarily the chief complaint on presentation.
Early diagnostic guess
Physicians gave their early diagnostic guess after the first minutes of the encounter with the patient. They were asked to complete the initial part of the case report form (CRF) before investigating the patient's history or performing any medical examination. Physicians based their early diagnostic guess on their previous knowledge of the patient, initial contact, and spontaneous presentation of complaint. GPs were free to report the early diagnostic guess in any terms and made no guess if the most probable cause of pain was unclear. Reported diagnoses were then divided into six categories (Table ).
Categorisation of reported early diagnostic guess for chest pain.
GPs also recorded if a diagnosis for a similar complaint was already known and whether the patient was feeling anxious about the pain. The diagnostic guess was recorded at four time points during the initial patient encounter: prior to history taking and physical exam, post history taking, post physical exam, and at the end of the encounter.
We used the diagnosis eventually retained after one year of follow-up as the definite diagnosis to be contrasted with the early diagnostic guess. An independent panel of physicians confirmed the follow-up diagnosis. Any new investigation, additional tests, reports from hospitals, or specialist referral that occurred after the initial diagnosis made at the end of the initial encounter was considered to reach the definite diagnosis. Adjudicators remained blinded to the early diagnostic guess. For the patients that were lost at follow-up, the information collected during the study and the patient's up to date medical records were used by the adjudicators. This method is not believed to be perfect but is the best acceptable solution for studies in family practice settings [11
]. Quality control of the reported diagnosis was done using patients up to date medical records at the GP's office for ten percent of the included patients. All reported diagnoses at the one year follow-up were then categorized by grouping of disorders in the same manner as the initial guess (Table ). The definition of a severe, potentially life-threatening illness included myocardial infarction, stable or unstable angina, pulmonary embolism, pneumonia and pleurisy, acute asthma, and neoplasm.
Prevalence of cases for which the early diagnostic guess was to be confirmed was calculated at a cluster level for each GP. Summary measure for all GPs was given using a frequency weighed mean value with a 95% CI. Influence of age (<50 yrs vs. >50 yrs), sex (male vs. female), known vs. unknown patient, new vs. known manifestation of chest pain, severe vs. non severe illness and patient anxiety over the predictive ability of early diagnostic guess were estimated. Predictive ability was calculated stratifying the results for each of these variables. Odds of correctly diagnosing the illness early were calculated using random effect logistic regression, adjusting for cluster effects verified by quadrature check. Homogeneity of these effects across GPs was verified by calculating the intraclass correlation coefficient (ρ). No correction for multiple testing was planned; significance level was set at p < 0.05. The study protocol was approved by the official Ethical Commission of Internal Medicine (Prot. 41/2000).