Scenario

On inpatient teaching rounds, we recently discussed the admission of a 45 year old woman with chest pain. This patient presented to the emergency department after an episode of chest pain lasting approximately 30 minutes. The pain was pleuritic in nature, located on the right side of her chest, and was associated with shortness of breath. The patient reported no palpitations, diaphoresis, nausea, or vomiting. She had no relevant medical history, was taking oral contraceptives, and had no family history of cardiac or thromboembolic disease. Her vital signs, including oxygen saturation, were normal, as were the physical examination results. The working diagnosis was pulmonary embolus (PE), and she was started on intravenous heparin and scheduled for a ventilation-perfusion (V/Q) scan to be done within the next half hour.

After interviewing the patient with a team (2 residents, 3 interns, and 2 medical students) and reviewing the data, I asked the team to individually record their best estimate of the probability that this patient had PE. At first, the team was reluctant, in particular the third year residents, but with encouragement all members of the team scribbled down their best estimate. The probabilities were reviewed and were found to range from 5% to 80%.

The importance of accurately determining the pre-test probability of disease is highlighted in figure 1. For our patient, resident 1 estimated the patient’s pre-test probability of PE as being high (ie, 80%). This pre-test probability combined with a low probability V/Q scan result (likelihood ratio = 0.4) gives a 61.5% probability of PE.¹ This post-test probability is not sufficient either to rule in or rule out the diagnosis of PE and would prompt the resident to order further diagnostic tests, such as spiral computed tomography or pulmonary angiography, to better clarify the diagnosis. Resident 2 estimated the patient’s pre-test probability as being low (ie, 5%). When combined with the same V/Q scan results, a final probability of PE of 2.1% is estimated. Resident 2 would probably stop any further diagnostic work up. Therefore, for the same patient, differing assessments of the pre-test probability of a PE by the 2 residents results in very different management strategies.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Resident 1 estimates the pre-test probability of PE at 80%. This estimate results in a post-test probability of 61.5%, given a low probability V/Q scan (likelihood ratio=0.4). Resident 2 estimates the pre-test probability of PE at 5% for the same patient. This results in a post-test probability of 2.1%, given a low probability V/Q scan.

As a group, we decided it would be useful to see whether we could quickly find valid information about the patient’s pre-test probability to help us with the diagnostic decision making process.

Search and appraisal

Information about the accuracy of diagnostic tests is difficult to find easily and quickly. Even more challenging is trying to find information about pre-test probability. Several potential sources could be used to determine pre-test probability, including our own clinical expertise, an audit of our practice, and the primary literature.

Clinical prediction rules (CPRs)* have been created to help clinicians arrive at more accurate estimates of pre-test probability. A CPR is a tool that quantifies the individual contributions that various components of the history, physical examination, and basic laboratory test results make towards the diagnosis, prognosis, or probable response to a treatment of an individual patient.² CPRs are most useful when directed at frequent problems for which the stakes are high or cost saving is possible.

CPRs are most useful at the patient’s bedside. However, they often require solving complex algorithms or performing tedious calculations to obtain a result. Therefore, many computer-based models have been developed on the internet and for personal digital assistants (PDAs). Electronic CPRs enable easy calculation of pre-test probability and give quick and perhaps more accurate estimates.

A small library of CPRs can be found at the Mount Sinai EBM web site (http://med.mssm.edu/ebm). These CPRs were chosen by a team of academic general internists for common medical problems and are organised by level of evidence,² a measure of quality that describes whether the CPR is merely derived (level 4) or validated (levels 2 to 3) or whether an impact analysis of the CPR has been done (level 1). Such key information as the study population and the point of decision making at which the CPR should be applied are a link away.^3,4 CPRs can also be downloaded for PDA devices from this web site.

Applying the evidence

In the scenario described above, the team referred to a CPR on the Mount Sinai EBM web site that was based on information from a study by Wells et al⁵ and from the PIOPED study.⁶ We first noted that the Wells CPR is considered level 3 evidence and that the study population included patients similar to ours. On the first page of the web site we found a list of respiratory symptoms and noted that our patient had 2 of these (dyspnea and pleuritic chest pain) (fig 2). 3 subsequent screens asked the team to record whether the patient had additional signs or symptoms and any of 8 listed risk factors for PE. We were also asked to consider whether an alternative diagnosis could be less likely than PE in this patient. Clicking the “calculate” button on this last screen (fig 3) produced an estimate of the patient’s pre-test probability (3.4%, 95% CI 2.5 to 5.0) and a post-test probability of PE of 1.4% (CI 1.0 to 2.0) following her low probability V/Q scan. With this low post-test probability, the team felt comfortable ruling out PE.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Respiratory symptoms.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Alternative diagnosis, V/Q scan result, pre-test probability, and post-test probability.

Using the CPR and our patient’s unique clinical circumstances, the calculation process took less than 1 minute. Our team, however, also spent several minutes reviewing the links for the level of evidence, population, and clinical decision model on the web site.

Evaluating the process

Medicine is still an art, and all clinicians have their own threshold of uncertainty that they are willing to accept before making a diagnostic or therapeutic intervention. CPRs are only tools in this process and may serve to provide an objective standard by which clinicians can better quantify the pre-test probability of disease.

Several internet sites have both browser-based and PDA-based CPR calculators (table). These sites help us to practise evidence-based medicine in real time. In this scenario we were able to find and use a CPR in less than 2 minutes. However, few of these sites identify the level of evidence of a particular CPR or the population from which the rule was derived. Both types of information are necessary to apply the rule safely and in the correct setting.