The aims of this study were to assess the effectiveness of 2 automated electrocardiogram interpretation programs in patients with suspected acute coronary syndrome transported to hospital by ambulance in 1 rural region of Denmark with hospital discharge diagnosis used as the gold standard and to assess the effectiveness of cardiologists’ triage decisions for these patients based on initial electrocardiogram. Twelve-lead electrocardiograms were recorded in ambulances using a LIFEPAK 12 monitor/defibrillator (Physio-Control, Inc., Redmond, Washington) and transmitted digitally to an attending cardiologist. If a diagnosis of ST elevation myocardial infarction was made, a patient was taken to a regional interventional center for primary percutaneous coronary intervention or to a local hospital. One thousand consecutive digital electrocardiograms and corresponding interpretations from LIFEPAK 12 were available, and these were subsequently interpreted by the University of Glasgow program. Electrocardiogram interpretations and cardiologists’ decisions were compared to hospital discharge diagnoses. The sensitivity, specificity, and positive predictive values for a report of ST elevation myocardial infarction with respect to discharge diagnosis were 78%, 91%, and 81% for LIFEPAK 12 and 78%, 94%, and 87% for the Glasgow program. Corresponding data for attending cardiologists were 85%, 90%, and 81%. In conclusion, the Glasgow program had significantly higher specificity than the LIFEPAK 12 program (p = 0.02) and the cardiologists (p = 0.004). Triage decisions were effective, with good agreement between cardiologists’ decisions and discharge diagnoses.
The aims of the present study were (1) to evaluate the accuracy of automated interpretation of electrocardiograms (ECGs) acquired in an ambulance using a LIFEPAK 12 defibrillator/monitor (Physio-Control, Inc., Redmond, Washington) with respect to ST elevation myocardial infarction (STEMI) and to compare it to the accuracy of ECG interpretation made by the University of Glasgow ECG analysis program, (2) to compare automated interpretations to an on-call cardiologist’s decision of whether to send a patient directly to an interventional center for primary percutaneous coronary intervention (pPCI) or to a local hospital, (3) to assess the validity of the cardiologist’s triage decision against the final diagnosis, and (4) to investigate if there was any evidence that the decision to refer a patient was unduly influenced by gender or age considerations.
Methods
Consecutive patients from Zealand, Denmark were included in the study with electrocardiograms recorded from January 1, 2007 to March 1, 2008. All patients had symptoms suggesting an acute coronary syndrome (ACS). A digital prehospital electrocardiogram was recorded in an ambulance and transmitted to an on-call cardiologist at Rigshospitalet, Copenhagen University Hospital (Copenhagen, Denmark). If a diagnosis of acute myocardial infarction (MI) were confirmed, the patient was taken directly to an interventional center; otherwise the patient was taken to a local hospital. If, in the local hospital, pPCI was subsequently thought necessary, the patient was transferred to an interventional center. Data were collected in all patients and consisted of digital prehospital 12-lead ECG tracings with automated interpretations and 3 databases of related patient information. The first database held the demographic details for all patients and discharge diagnoses based on clinical course, biomarkers, and angiographic findings. A patient’s hospital discharge diagnosis could consist of several diagnoses and, for a patient being transferred between hospitals, a discharge diagnosis was available from each hospital. This database contained the time and outcome of the triage decision, i.e., whether the patient was taken directly to an interventional center or admitted to a local hospital. The second database contained data on the subset of patients who subsequently had an angiogram, and the third contained data on patients who had a pPCI.
Some patients had multiple entries in the databases because they had been transported by ambulance with suspected ACS on different dates. Each admission was treated separately. In some cases, a discrepancy was found between the discharge diagnosis recorded in the demographic database and findings as recorded on angiogram and PCI databases. These cases were reviewed by the local authors (M.S., P.C.) without any knowledge of the Glasgow ECG report to determine the diagnosis based on all data available for individual patients including electrocardiograms, biomarker data and angiographic/PCI results.
Each automated report from LIFEPAK 12 was examined for presence of an acute MI statement, which was considered equivalent to STEMI. LIFEPAK 12 uses the 12SL 14 ECG algorithm (GE Marquette Medical Systems, Milwaukee, Wisconsin), which outputs a statement, “ACUTE MI SUSPECTED” when an injury pattern suggestive of an MI has been identified. Criteria for an injury include abnormal ST elevation, increased ST:T ratio, and reciprocal changes. Thresholds used for determining ST elevation are 0.1 mV in the inferior and lateral leads and 0.2 mV in chest leads (leads V 1 to V 4 ).
Raw data for electrocardiograms were extracted retrospectively by Physio-Control from its ECG management system and sent to Glasgow. Electrocardiograms were reanalyzed using the Glasgow program. This program uses age- and gender-dependent criteria based on ST amplitude in 2 contiguous leads and occasionally in 1 lead only. Criteria use continuous equations for determining upper limits of normal ST amplitudes and the program outputs the statement, “CONSIDER ACUTE ST ELEVATION MI” if appropriate.
Four sets of comparisons were made. The first comparison was between the occurrence of STEMI in automated reports and discharge diagnosis. Second, the sensitivity and specificity of the occurrence of acute MI in the automated ECG output were calculated for the 2 analysis programs and compared against each other. Third, the triage decision of where the patient should be sent, i.e., to an interventional center or a local hospital, was equated to a cardiologist’s decision of STEMI or non-STEMI, respectively, and compared to the discharge diagnosis. Fourth, a cardiologist’s triage decision was compared to the automated computer output for the LIFEPAK 12 and Glasgow programs.
Analysis was repeated for subgroups of men and women separately to see if the automated interpretation programs were equally efficient for men and women and to see if there was any gender bias when patients were referred to an interventional center for pPCI. Subgroups were then subdivided into age categories and reanalyzed.
When analyzing differences in sensitivities or specificities, nonparametric McNemar test was run to compare results using a binomial distribution due to small numbers in some categories. The statistical package used was SPSS 15 (SPSS, Inc., Chicago, Illinois). Level of statistical significance was set at a p value <0.05.
To establish the procedures to be used in the collaboration between the universities of Copenhagen and Glasgow, a pilot study was carried out on 200 randomly selected electrocardiograms. Results from the pilot study were presented at the Computers in Cardiology conference in 2009. The 200 electrocardiograms were included in this study.
Results
One thousand consecutively recorded electrocardiograms were initially included but 88 were subsequently excluded from analysis. Table 1 presents the reasons for exclusion. The dataset therefore consisted of 912 cases. Nineteen patients had been entered in the study on 2 different occasions and 2 patients on 3 different occasions. Three hundred ninety-three of the 912 patients had an angiogram. Three hundred ten patients had a pPCI. The discharge summary included biomarker values in 123 cases. In most cases, biomarker results were not copied to the discharge note but were available to the physician making the diagnosis. Conflicting diagnoses among databases were recorded in 45 patients (5%).
| Reason for Exclusion | Number of Electrocardiograms |
|---|---|
| Implanted cardiac pacemaker | 33 |
| Missing discharge diagnosis | 25 |
| Sudden death | 11 |
| Ventricular or wide complex tachycardia | 8 |
| Missing triage data | 4 |
| No data | 3 |
| Technically unsatisfactory electrocardiogram | 3 |
| Not admitted to hospital | 1 |
| Total | 88 |
Results from the comparison of LIFEPAK 12 and Glasgow programs to discharge diagnosis are presented in Table 2 . The 2 programs had a sensitivity of 78%. The κ value for the measurement of agreement between the 2 programs was 0.753, implying substantial agreement. However, the Glasgow program had a significantly higher specificity than the LIFEPAK 12 program (94% vs 91%, p = 0.02).
| Discharge Diagnosis | Glasgow | LIFEPAK 12 | ||
|---|---|---|---|---|
| Not STEMI | STEMI | Total | ||
| Not STEMI | Not STEMI | 529 | 37 | 566 |
| STEMI | 19 | 18 | 37 | |
| Total | 548 | 55 | 603 | |
| STEMI | Not STEMI | 48 | 21 | 69 |
| STEMI | 20 | 220 | 240 | |
| Total | 68 | 241 | 309 | |
To assess the validity of the triage decision, the admission route followed by the 912 patients was analyzed ( Figure 1 ). Of the 325 patients sent directly to an interventional center for pPCI by the attending cardiologist, 81% were discharged with a diagnosis of STEMI. Of the 587 patients sent to a local hospital, 541 (92%) did not have STEMI documented in their discharge diagnosis. Of the remaining 46 who were given a discharge diagnosis of STEMI, 21 were transferred to an interventional center after ECG changes or increased biomarker measurements. In 9 of the 21 patients, the ambulance electrocardiogram had been reported as STEMI by LIFEPAK 12. There were 25 patients who were sent to the local hospital and discharged with a diagnosis of STEMI.
Sensitivity and specificity indexes for the cardiologist’s triage decision on referral with respect to final discharge diagnosis were 85% and 90%. Although the Glasgow program had significantly higher specificity than the cardiologist (94% vs 90%, p = 0.004), the LIFEPAK 12 program showed no significant difference in specificity at 91% (p = 0.5). The 2 programs had significantly lower sensitivity at 78% versus 85% (p = 0.002 for comparison of the Glasgow program with the cardiologists and p = 0.004 for corresponding LIFEPAK 12 comparison). Positive predictive value for a report of STEMI was 87% for the Glasgow program compared to 81% for LIFEPAK 12 and the cardiologists.
Of the 912 patients transported by ambulance, 66% were men with a mean age of 63.4 ± 13.4 years and 34% were women with a mean age of 69.3 ± 14.1 years. Of the 309 patients for whom the discharge diagnosis was STEMI, 72% were men. Seventy-five percent of patients referred to an interventional center were men ( Figure 2 ).
For subgroups of men and women, there were no significant differences in the indexes reported by the 2 programs with respect to final diagnoses ( Table 3 ). There were also no significant differences for the subgroup of women when a cardiologist’s decision to refer to an interventional center was compared to the automated results. However, for the subgroup of men ( Table 4 ), specificity of the Glasgow program was significantly higher than a cardiologist’s decision (93% vs 88%, p = 0.006), whereas the corresponding sensitivity was significantly lower (79% vs 88%, p = 0.001). Levels of significance for comparing the LIFEPAK 12 program report to the triage decision for men were a p value equal to 0.4 for specificity (90% vs 88%) and a p value equal to 0.002 for sensitivity (79% vs 88%).
| Sensitivity | Specificity | |||||
|---|---|---|---|---|---|---|
| LIFEPAK 12 | Glasgow | Triage | LIFEPAK 12 | Glasgow | Triage | |
| Men (n = 601) | 79% | 79% | 88% | 90% | 93% | 88% |
| Women (n = 311) | 74% | 75% | 78% | 92% | 95% | 93% |
| Discharge Diagnosis | Glasgow | Triage | ||
|---|---|---|---|---|
| Not STEMI | STEMI | Total | ||
| Not STEMI | Not STEMI | 317 | 34 | 351 |
| STEMI | 14 | 12 | 26 | |
| Total | 331 | 46 | 377 | |
| STEMI | Not STEMI | 19 | 29 | 48 |
| STEMI | 8 | 168 | 176 | |
| Total | 27 | 197 | 224 | |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
