Author + information
- Phillip D. Levy, MD, MPH∗ ( )()
- Department of Emergency Medicine and Integrative Biosciences Center, Wayne State University, Detroit, Michigan
- ↵∗Address for correspondence:
Dr. Phillip D. Levy, Department of Emergency Medicine, Wayne State University, Integrative Biosciences Center, 6135 Woodward Avenue, Detroit, Michigan 48202.
“To wish was to hope, and to hope was to expect.”
—Jane Austen (1)
In this issue of the Journal, Nejatian et al. (2) report findings from a registry-based cohort study of 66,506 patients evaluated in 16 Swedish hospital emergency departments (EDs) for chest pain prior to and post-initiation of clinical routine high-sensitivity troponin T (hsTnT) testing. The primary aim was to investigate whether 30-day incidence of major adverse cardiac events (MACE) was reduced, with a secondary aim to determine if risk factor profiles of those tested were different following uniform implementation of the hs-TnT assay. The investigation was specifically focused on the subgroup of patients who were discharged, either directly from the ED (88%) or after a short observation/hospital stay (12%), with a diagnosis of “chest pain, unspecified” using International Classification of Diseases-10th Revision (ICD-10) codes as the identifier. The emphasized findings include a marginal difference in MACE incidence (absolute difference −0.3%) for those discharged directly from the ED, and a more substantive change in MACE risk (absolute difference +3.8%) for those who were placed in observation status or admitted to the hospital after introduction of hsTnT. There were also statistically significant differences in cardiovascular risk factor prevalence post–hsTnT implementation: those patients who were observed or admitted were slightly older and somewhat sicker, and patients in the directly discharged group were younger with fewer comorbidities. Based upon this, the authors conclude that the “hsTnT assay appears to be of great help in the ED when deciding whether a patient can be sent home or not” (2).
With the U.S. Food and Drug Administration’s recent approval of the hsTnT assay (Roche Diagnostics, Indianapolis, Indiana) for commercial applications in the United States, understanding the clinical effect of this test, and framing the use case for its implementation is of utmost importance. On the surface, the work by Nejatian et al. (2) appears to add to the growing body of published data supporting the clinical benefit of hsTnT (3,4); however, on closer scrutiny, the precise contribution appears to be less about the assay, and more about the pitfalls of causal inference based on population-level data.
To be clear, there is a role for population-level data analyses, particularly as they relate to cardiovascular disease risk. But, as with all studies that report associations and not true causal relationships, data from such work must be allowed to speak for itself, free from attachment of meaning or subjective interpretation. While the findings of Nejatian et al. (2) do support that absolute differences in MACE incidence exist before and after introduction of the hsTnT assay, they also show that these differences vanish when age, sex, diabetes mellitus, hyperlipidemia, hypertension, and ED visit date are accounted for. Although seemingly pivotal, this information receives only brief mention at the end of the Study Limitation section, and is explained away as an anomaly of risk factor differences, rather than embraced as an indicator of probable neutrality.
Did the authors misinterpret study findings? Doubtful. More likely, there was unintentional bias arising from connection of seemingly evident patterns in the data. There were indeed substantive differences in the prevalence of cardiovascular risk factors among patient subgroups pre-/post-assay change. The authors interpret this an indicator of better risk profiling; yet, we have no information on whether standardized methods of chest pain risk stratification such as the HEART (History, Electrocardiogram, Age, Risk Factors and Troponin) score (5,6) or protocols for disposition planning were implemented at any of the 16 hospitals that contributed data. Absent such insight, it is unclear what interaction may exist between risk profiling and testing, and what incremental value the troponin assay may have provided within that context. One could reasonably speculate that perhaps older, sicker patients would be more likely to have detectable or even low-level elevations in troponin concentration (and vice versa for younger, less-sick patients), and that this was the driver of observed differences. However, we have no actual troponin data (no values and no information on what percent of the study population was even tested), and know little other than the fact that hsTnT testing was available at participating hospital centers after a certain date. However, what is certain, is that all of these patients had a final diagnosis of unspecified chest pain implying that the treating clinician: 1) did not primarily suspect acute coronary syndrome; and 2) did not have a definitive, alternative diagnosis. With this in mind, it would be quite unusual in clinical practice for a patient with an elevated hsTnT and symptoms at least suspicious enough for acute myocardial injury to order the test to be diagnosed with unspecified chest pain, rather than a more definitive attributable etiology (7). Whether concurrent patterns for more specific potentially related diagnoses such as acute coronary syndrome, unstable angina, non–ST-segment elevation myocardial infarction, pulmonary embolism, and acute heart failure shifted during the study period would help to clarify.
So, what then does this study say about routine availability of hsTnT? Interestingly, the best answer to that question is hiding in the data itself, right there in plain sight. As shown in Table 1, the average number of patients diagnosed with unspecified chest pain rose by 125 a month (from 829 to 954 patients) post-assay implementation, whereas the number observed or admitted per month dropped by nearly 40% (from 138 to 85 patients). Stated simply, after assay introduction, more people were given an unspecified chest pain diagnosis, and more of those given this diagnosis were directly discharged from the ED. This could be interpreted as demonstrating that hsTnT implementation leads to fewer observed or admitted patients with unspecified chest pain, and better identification of those who are at higher risk. This would be consistent with other studies conducted using similar source data from the SWEDEHEART (Swedish Web-system for Enhancement and Development of Evidence-based care in Heart disease Evaluated According to Recommended Therapies) registry (8), but it could also indicate that the net result is testing of many more low-risk people in the ED. Either way, this produces a dilution effect on MACE, with lower-risk patients being added to the pool driving down the proportion of outcome events for the ED discharge group. A movement of low-risk patients from observed or admitted to directly discharged may have added to this, perhaps triggering a collateral increase in acuity for those who were hospitalized post-assay change.
It is natural to want to understand why such patterns exist, and to posit that, somehow, implementation of the assay altered clinician practice is logical, as we know disruptive technology tends to do this. But, it may be that knowledge of assay availability prompted more low-risk patients to seek evaluation for chest pain, or led more general practitioners to send their patients to the ED for a work-up. Perhaps it is all explained by changes in coding patterns over time, with unspecified chest pain being used with increased frequency for ED patients and decreased frequency on the hospital side. It may even mirror a general trend toward increasing ED visits over time, as we have experienced in the United States (9). However, unlike the United States, Sweden is a country with universal health care coverage and an extensive, integrated network of general practitioners and nurses who work with patients to handle many health concerns outside of the ED setting (10), making this explanation less plausible.
Whatever attribution is assigned, it will still fall prey to the challenge of causal inference and the issue of ecological fallacy. As can be seen by close analysis of study data, explanation of meaning requires intent, and intent is ultimately a human quality displayed at the individual level. Thus, inherent to any of the potential explanations described in the preceding paragraph, or those put forth in the Discussion and Conclusion sections of Nejatian et al. (2), is an individual decision-making process that, be it by a physician or nurse, patient or caregiver, coder or biller, was altered entirely by presence of the hsTnT assay. Does it make sense that a high-sensitivity troponin assay would be able to have such an effect all on its own? It depends on what you wish for, hope for, and expect.
↵∗ Editorials published in the Journal of the American College of Cardiology reflect the views of the authors and do not necessarily represent the views of JACC or the American College of Cardiology.
Dr. Levy has received consultant fees from Siemens and Roche Diagnostics.
- 2017 American College of Cardiology Foundation
- Austen J.
- Nejatian A.,
- Omstedt Å.,
- Höijer J.,
- et al.
- Body R.,
- Burrows G.,
- Carley S.,
- et al.
- McCord J.,
- Cabrera R.,
- Lindahl B.,
- et al.
- Roffi M.,
- Patrono C.,
- Collet J.P.,
- et al.