In a large cohort of primary care patients presenting with chest pain, we demonstrated important gaps in quality and safety, with high risk patients not receiving recommended care and low risk patients undergoing many low-yield tests. Our electronic alerts provided evidence-based recommendations based on real-time calculation of the Framingham Risk Score, however did not significantly impact clinical practice patterns.
Electronic health records are promoted as an important patient safety tool,12,18
as well as having great potential to increase the efficiency of health care and reduce unnecessary testing.19,20
As the nation considers a substantial investment to support the broad implementation of electronic health records,21
the failure of our intervention highlights the need for deeper insight into how to use them to change physician behavior. While there are data to support the use of decision support to improve quality and medication safety,22,23
our data add to prior evidence suggesting that these benefits are not universally realized.24–26
The design of electronic decision support systems play a large role in their effectiveness.27
Many chest pain algorithms rely on symptom description and are difficult to implement using real-time decision support.28–30
The Framingham Risk Score represents a promising decision support tool for patients with chest pain.5,31,32
In our study, a risk score of at least 10% was associated with a higher occurrence of acute myocardial infarction within 30 days of presentation, highlighting the relevance of this risk assessment in the primary care evaluation of chest pain. The electronic health record can automate the calculation of this risk score and provide real time recommendations directly integrated into the workflow without requiring additional information input on the part of clinicians.
Our surveys indicate that the electronic alerts were well received by clinicians. Rather than problems of workflow integration or usability, it may be that the clinical benefits of electronic decision support as a stand-alone intervention do not extend to more complex clinical scenarios. Clinicians bring significant clinical intuition and experience to these encounters, and this experience may have superseded the information provided by the risk scores. Electronic alerts may be only one piece of a multi-targeted effort required to improve the management of complex scenarios such as chest pain.
An important distinction is whether our program failed to change care patterns due to limitations of the technology-based intervention, or due to lack of clinician trust in the clinical recommendations being offered. Our data indicate the latter issue did not play a large role as the majority of clinicians endorsed the validity of the Framingham Risk Score as a tool when evaluating patients complaining of chest pain.
To our knowledge, this is the largest prospective analysis of chest pain management in primary care. Errors in diagnosis represent a leading ambulatory patient safety concern,18,33,34
and our data showed that while the occurrence of acute myocardial infarction was infrequent, misdiagnosis was common. Over one-third of acute myocardial infarctions were misdiagnosed, compared to less than 5% in the emergency department setting.35
This is not surprising, given that only one-half of high risk patients had an electrocardiogram performed, despite its key role in evaluating patients with chest pain.9–11
The challenges to improving patient safety in the outpatient setting are substantial,34
particularly as primary care physicians may not view errors in diagnosis as an important patient safety concern.36
The majority of the patients in our study were low risk, and approximately 10% of these patients underwent cardiac stress testing. Current guidelines recommend against the use of such testing for low risk patients based on the poor positive predictive value.14,37
The common use of this low yield test represents a key area for improving efficiency, though recommending against such testing for patients with chest pain may be particularly challenging given physician concerns regarding patient safety and malpractice. Prior studies of decision support have focused on reducing unnecessary testing represented by avoiding redundant testing, though greater value may be achieved by avoiding tests that are not needed at the outset.
Our study should be interpreted in the context of some limitations. We relied on medical assistants to identify patients with chest pain. This decision was based on the need to identify patients prior to the evaluation by the clinician to deliver real-time risk information. We conducted extensive training of all medical assistants including performance feedback, and validated their identification of patients using medical record review. We chose the Framingham Risk Score as a tool to risk stratify primary care patients with chest pain. Many other risk prediction instruments exist to risk stratify primary care patients with chest pain.28,30
The Framingham Risk Score provides a substantial advantage over other instruments by producing a valid risk estimate without the need for additional input such as detailed symptom description, and without requiring testing not typically available in primary care such as serial cardiac enzymes or immediate cardiac stress testing.
In conclusion, this study of primary care management of acute chest pain demonstrated important quality and safety concerns. A well-designed electronic decision support system was acceptable to clinicians, but did not impact clinical practice patterns, with errors in diagnosis and overuse of stress testing persisting. Future work is needed to understand how advanced electronic health records can be used to improve the quality and safety of health care delivery.