- Email: [email protected]
Validation of a short rhythm strip compared to ambulatory ECG monitoring for ventricular ectopy
Journal of Clinical Epidemiology 53 (2000) 491–497
Validation of a short rhythm strip compared to ambulatory ECG monitoring for ventricular ectopy Kelly R. Evensona,*, Verna L. Lamar Welcha, Wayne E. Casciob, Ross J. Simpson Jr.b a
Department of Epidemiology, School of Public Health, University of North Carolina–Chapel Hill, 137 East Franklin Street, Bank of America Center, Suite 306, Chapel Hill, NC 27514, USA; b Preventive Cardiology Program, Division of Cardiology, Department of Medicine, University of North Carolina–Chapel Hill, Chapel Hill, NC 27599-7075, USA Received 4 August 1998; received in revised form 11 June 1999; accepted 17 September 1999
Abstract Premature ventricular contractions (PVCs) are associated with an increased risk of cardiovascular disease and mortality. Many epidemiologic studies measure a continuous short rhythm strip to ascertain PVCs as a screening tool to identify persons at highest risk. Despite its widespread use in epidemiologic studies, the rhythm strip has not been completely validated. Therefore, a continuous 2-min rhythm strip was measured on 242 consecutive individuals referred for ambulatory ECG monitoring. Prevalence of at least one PVC on the 2-min rhythm strip was compared to a gold standard, the average number of PVCs per hr on ambulatory recording. The prevalence of any PVCs on the 2-min rhythm strip was 19%. As average PVCs per hr increased on the ambulatory ECG recording, sensitivity increased while specificity slowly decreased. Sensitivity ranged from 26–100% and specificity ranged from 81–100% across the distribution of average PVCs per hr on ambulatory monitoring. Area under the receiver operator characteristic (ROC) curve of the 2-min rhythm strip compared to 24-hr results was 0.943. Area under ROC curves were not statistically different (P ⬎ 0.05) by age, gender, hypertension status, or history of myocardial infarction. In this clinical population, utilizing the 2-min rhythm strip as an indicator of average PVCs per hr had excellent specificity and moderate to low sensitivity across most of the distribution of average PVCs per hr. The use of a short rhythm strip to detect PVCs may be considered useful in epidemiologic investigations of cardiovascular disease and mortality for detecting high frequency PVCs in populations. The use of a short rhythm strip as a screening tool to detect PVCs in clinical practice is not warranted, based on our findings and the existing literature. However, an awareness that PVCs on a 2-min rhythm strip consistently identify high frequency PVCs on 24-hr recordings should be helpful to clinicians. © 2000 Elsevier Science Inc. All rights reserved. Keywords: Ventricular premature complexes; Electrocardiography; Arrhythmia; Ambulatory electrocardiography; Epidemiology; Validation
1. Introduction Occurrences of premature ventricular contractions (PVCs) on ambulatory electrocardiograph (ECG) are associated with an increased risk of subclinical [1] and clinical cardiovascular disease [2,3], as well as mortality in post-myocardial infarction (MI) patients [2,4–19]. In apparently healthy men, PVCs on rhythm strip recordings as short as 2 min were correlated with an increased risk of sudden cardiac death [20] and complex or frequent ventricular arrhythmias from a 1-hr recording were associated with all-cause and coronary heart disease mortality [21]. Additionally, for men at least 40 years of age without apparent heart disease, PVCs on routine ECG were related to an increased risk of ischemic heart disease, especially sudden cardiac death [22]. * Corresponding author. Tel.: 919-966-1967; fax: 919-966-9800. E-mail address: [email protected] (K.R. Evenson)
Many cardiovascular epidemiologic studies measure a continuous short rhythm strip to ascertain PVCs as a screening tool to identify high-risk individuals. Despite its widespread use in epidemiologic studies, the rhythm strip has not been completely validated. Validity of a short rhythm strip was reported only for men participating in a single cohort study [23,24]. Therefore, the main objectives of this study were to validate the ability to the standard 2-min rhythm strip to detect PVCs relative to ambulatory ECG monitoring and to describe the relationship between patient characteristics and the occurrence to PVCs on the 2-min rhythm strip. 2. Methods Inpatients and outpatients referred to the University of North Carolina Memorial Hospital in Chapel Hill, NC, for a clinically indicated continuous ambulatory ECG monitor from July 17, 1995 to January 26, 1996 were consecutively
0895-4356/00/$ – see front matter © 2000 Elsevier Science Inc. All rights reserved. PII: S0895-4356(99)00 1 9 0 - 0
492
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497
enrolled in this cross-sectional study. A 2-min continuous rhythm strip was obtained at the same time the individual began ambulatory ECG monitoring. Participants were excluded if they were less than 18 years of age, institutionalized, had a pacemaker, or had any of the following findings on the rhythm strip: sick sinus syndrome (heart rate less than 50 beats per min), atrial fibrillation, or atrial flutter. Ambulatory ECG monitoring was indicated for the following conditions: 22% syncope or dizziness, 18% angina or ischemia, 19% palpitations, and 41% arrhythmias. Only the participant’s first-time visit during the study period contributed to these analyses. 2.1. Rhythm strip With the patient lying supine, electrode position for the 2-min rhythm strip was determined using standard criteria to best identify P waves. A modified chest lead two (MCL2) was most often used. Two cardiologists independently evaluated the 2-min rhythm strip to determine the number of PVCs. In case of disagreement on the number of PVCs, a cardiologist adjucated the discrepancy. Before adjucation, agreement between the two cardiologists on the presence of any PVCs on the 2-min rhythm strip was evaluated, yielding a kappa coefficient of 98.7%. 2.2. Ambulatory ECG monitoring The simultaneous two-lead system (MCL2 and modified V5) was used for ambulatory ECG monitoring. The ambulatory ECG tape was read according to a standard, partially automated protocol by one of two trained ECG technicians masked to the rhythm strip results. For the 252 participants, the average length of ambulatory ECG recording was 22.7 hr. To avoid possible bias, only complete hr from the ambulatory ECG were used in these analyses, assuming for this purpose that PVCs occurrences were uniform over each hr. The 1-, 2-, and 3-hr results were taken from the first 1, 2, and 3 hr, respectively, of the monitoring tape. 2.3. A priori definitions For the 1- and 2-min rhythm strip, a positive test was indicated if at least one PVC occurred during the recording. For the 1-, 2-, and 3-hr results, a positive test was indicated if there was an average of at least 10 PVCs per hr during the 1, 2, or 3 hr of recording, respectively. The gold standard for validation was the average number of PVCs per hr from all complete hr of ambulatory ECG monitoring. To examine complex ectopy, we defined the gold standard in two ways: (1) presence of any ventricular triplets or tachycardia on ambulatory ECG; and (2) presence of any ventricular couplets, triplets, or tachycardia on ambulatory ECG. Hypertension and history of MI, as well as smoking and ingestion of any food or caffeine in the 2 hr prior to the rhythm strip recording were self-reported and logged by the ECG technicians for this study. Age and ethnicity were obtained from the medical record.
2.4. Statistical analyses Categorical analyses were performed using Mantel-Haenszel chi-square tests. Median tests were calculated to determine whether median average PVCs per hr from ambulatory ECG recordings differed significantly by various factors. Because sensitivity is inversely related to specificity, the likelihood ratio has been proposed to be more stable [25]. The pretest odds multiplied by the likelihood ratio provides the post-test odds of disease. Sensitivity, specificity, positive likelihood ratios [sensitivity/(1-specificity)], and negative likelihood ratios [(1sensitivity)/specificity] were calculated, along with their 95% confidence intervals (CI) [26]. Graphs of these were truncated at 600 average PVCs per hr from ambulatory recording due to a reduced sample size beyond this cutpoint. Accuracy was calculated as the correct number of classified patients (both positive and negative) divided by the total number of patients [27]. Unconditional multivariable logistic regression was used to identify independent predictors of PVCs on the 2-min rhythm strip. The full model included age, gender, ethnicity, hypertension, MI, and both minimum and maximum heart rate from ambulatory ECG monitoring. Both age and heart rate were left continuous in the model. Receiver operating characteristic (ROC) curves can efficiently quantify the relationship between true positive (sensitivity) and false positive (1-specificity) rates for tests with continuous outcomes [28]. Area under the ROC curve was derived from unconditional logistic regression, modeling the average number of PVCs per hr from the ambulatory ECG monitor as the independent variable and whether or not PVCs occurred on the rhythm strip as the dependent variable. Area under ROC curves were used to compare the 2-min rhythm strip to the 1-min rhythm strip as well as to the first 1-, 2-, and 3-hr recording from the ambulatory ECG, taking into account correlation between samples [29]. Additionally, area under ROC curves was utilized to compare 2-min rhythm strip results for groups separately stratified by gender, age, hypertension, and MI status. Because these data were uncorrelated, the method described by Hanley and McNeil [30] was used for comparison. For all ROC curve comparisons, the gold standard was the average number of PVCs per hr from all complete hr of ambulatory ECG monitoring.
3. Results Among 252 participants, the prevalence of any PVCs on the 2-min rhythm strip was 19%, ranging from 0 to 53 PVCs. Less than 2% of participants had one or more ventricular couplets, and there were no occurrences of ventricular triplets or ventricular tachycardia on the 2-min rhythm strip. The average number of PVCs per hr over 24-hr on the ambulatory ECG monitor was 49.2 (median 0.38, range 0–894). Average PVCs per hr on the ambulatory ECG monitor was 198.2 (median 101.6, range 2.2–894) for those with PVCs on the rhythm and 14.1 (median 0.17, range 0–605) for those without PVCs on the rhythm strip. Table 1 displays
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497 Table 1 Average PVCs per hr from ambulatory ECG monitoring for the study population stratified by presence or absence of PVCs on the 2-min rhythm strip Average PVCs/hr on ambulatory ECG
PVCs on 2-min rhythm strip Present (n ⫽ 48)
Absent (n ⫽ 204)
0 PVCs/hr ⬎0–⬍0.5 PVCs/hr ⭓0.5–⬍5 PVCs/hr ⭓5–⬍10 PVCs/hr ⭓10–⬍60 PVCs/hr ⭓60 PVCs/hr
0 (0%) 0 (0%) 1 (2%) 5 (10%) 11 (23%) 31 (65%)
66 (32%) 65 (32%) 35 (17%) 6 (3%) 22 (11%) 10 (5%)
the distribution of average PVCs per hr from ambulatory ECG monitoring according to presence or absence of any PVCs on the 2-min rhythm strip. Ten participants had an average PVCs per hr of at least 60, but did not have any PVCs on the 2-min rhythm strip. Table 2 displays the bivariate relationship between patient characteristics and cardiovascular risk factors to the prevalence of PVCs on the 2-min rhythm strip. Prevalence of PVCs increased with age (P ⬍ 0.01). Patients with a history of MI (P ⬍ 0.01) were also more likely to manifest PVCs on the rhythm strip. PVC occurrences were not significantly different by gender, ethnicity, or hypertension (P ⬎ 0.05). These results were similar when evaluating average PVCs per hr from the 24-hr monitor to patient characteristics. When simultaneously controlling for gender, ethnicity, hypertension, MI, minimum and maximum 24-hr heart rate, older age was a statistically significant predictor of PVCs
Table 2 Prevalence of PVCs on the 2-min rhythm strip and median average PVCs per hr from ambulatory ECG monitoring according to patient characteristics Patient characteristics Age in years 18–29 30–39 40–49 50–59 60–69 70–79 80–89 Gender Male Female Ethnicity White Nonwhite Hypertension Yes No Previous MI Yes No
Prevalence of PVCs on rhythm strip (%)
Median average PVCs/hour over 24 hr
4 12 11 17 15 43 53
0.02 0.17 0.24 0.17 1.25 27.70 14.91
21 17
1.00 0.25
20 17
0.37 0.38
19 18
0.87 0.26
32 16
29.77 0.17
493
on the 2-min rhythm strip (prevalence odds ratio 1.06, 95% CI 1.03–1.09). The remaining factors in the logistic model were not statistically significant predictors of PVCs on the 2-min rhythm strip. Table 3 displays the sensitivity and specificity of the 2-min rhythm strip for several average PVCs per hr cutpoints and Fig. 1 displays the entire range graphically. As average PVCs per hr increased, sensitivity increased while specificity decreased gradually. To examine whether our results were affected by those with MI, we re-examined sensitivity and specificity excluding those individuals with MI. Specificity did not meaningfully change from 0 to 60 average PVCs per hr while sensitivity increased above 0.5 average PVCs per hr. Repeat analyses were also performed, after excluding 76 individuals who ate, ingested caffeine, or smoked two hr prior to the ECG recording. Sensitivity and specificity of the 2-min rhythm strip to detect 0 to 60 average PVCs per hr did not meaningfully change. This was also found when excluding 28 individuals who had a premature atrial contraction during the rhythm strip recording. The proportion of individuals classified incorrectly as negative (false negative) on the 2-min rhythm strip ranged from 74% at the gold standard cutoff of ⬎0 average PVCs per hr to 24% at ⭓60 average PVCs per hr on ambulatory ECG monitoring. The proportion of individuals classified incorrectly as positive (false positive) on the 2-min rhythm strip ranged from 0% at a gold standard cutoff of ⬎0 average PVCs per hr to 8% at ⭓60 average PVCs per hr on ambulatory ECG monitoring. During 24-hr ECG monitoring, 30% (75) had at least one ventricular couplet and 13% (32) had at least one ventricular triplet or one run of ventricular tachycardia. Those with any evidence of complex ventricular ectopy (e.g., ventricular tachycardia, ventricular triplets, ventricular couplets) during ambulatory ECG monitoring were more likely to incur PVCs on the 2-min recording. Among those with any ventricular couplets on the 24-hr recording, 52% had at least one PVC on the 2-min rhythm strip. Similarly, among those with any triplets or ventricular tachycardia on the 24-hr recording, 71% and 74%, respectively, had at least one PVC on the 2-min rhythm strip. The ROC curve comparing the 2-min rhythm strip to 24hr results is displayed in Fig. 2. The area under this ROC curve was 0.943, indicating that the test did far better than chance alone in identifying average PVCs per hr. To compare other recording times to the 2-min rhythm strip, ROC curves were compared statistically. When comparing a ROC curve for a 1-min (area 0.919) versus a 2-min rhythm strip to 24-hr monitoring, differences between the two were not statistically significant (P ⬎ 0.05). When comparing a 2-min rhythm strip to a 1-hr recording to detect PVCs over 24 hr, the 1-hr recording (area 0.978) showed improvement, although the differences did not reach significance (P ⬎ 0.05). Similarly, the 2-hr (area 0.982) and 3-hr recording (area 0.984) improved upon the 2-min recording, but neither
494
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497
Table 3 Sensitivity and specificity with 95% confidence intervals (CI) comparing the 2-min rhythm strip to 24-hr monitoring for all participants and for those without a myocardial infarction (MI), and median PVCs per hr over 24 hr for the study population Gold standard from ambulatory ECG Average PVCs/hour ⬎0
⭓0.5 ⭓5 ⭓10 ⭓15 ⭓30 ⭓60 Any ventricular triplets or tachycardia Any ventricular couplets, triplets, or tachycardia
Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity
All (95% CI) (n ⫽ 252)
Without MI (95% CI) (n ⫽ 215)
26% (20–32) 100% 40% (31–48) 100% 55% (45–66) 99% (98–100) 57% (45–68) 97% (94–99) 57% (46–69) 95% (92–98) 71% (58–83) 94% (91–97) 76% (62–89) 92% (88–96) 66% (49–82) 88% (83–92) 49% (38–60) 95% (91–98)
24% (17–30) 100% 40% (30–50) 100% 57% (45–70) 99% (98–100) 59% (45–72) 96% (93–99) 61% (47–75) 95% (92–98) 76% (62–91) 94% (91–98) 85% (71–98) 93% (89–96) 58% (39–78) 88% (84–93) 47% (35–60) 95% (91–98)
reached statistical significance (P ⬎ 0.05). With increasing recording time on the screening test, sensitivity increased while specificity decreased. Because sensitivity and specificity can vary by subgroups within a population [25], we re-examined the comparison of the 2-min rhythm strip to continuous ambulatory ECG results stratifying separately on gender, age, hypertension, and history of MI. Area under ROC curves were not statistically different by gender (men 0.934 vs. women 0.053, P ⬎ 0.05), age (⭓50 years 0.931 vs. ⬍50 years 0.964, P ⬎ 0.05), or hypertension status (hypertension 0.937 vs. normotension 0.946, P ⬎ 0.05). Because the choice of age categories may have affected our results, we included additional cutpoints in our calculations. Area under ROC curves were still not statistically different by age, al-
Median average PVCs/hr over 24 hr (n ⫽ 252) 2.69 17.00 56.69 70.10 81.11 148.11 206.17 126.92 36.75
though at the highest age cutpoint there was an indication of reduced area under the ROC curve (⭓70 years 0.876 vs. ⬍70 years 0.952, P ⬎ 0.05). ROC curve comparisons between those with and without a history of MI were not statistically significant (MI 0.827 vs. no MI 0.960, P ⫽ 0.13). We may have lacked adequate statistical power to detect differences within these groups, because our sample included only 47 patients at least 70 years of age and 37 patients with a history of MI. We also examined the data separately for sensitivity and specificity across the average PVCs per hr continuum, because sensitivity and specificity can vary independently. Women, older individuals, and those with hypertension tended to have higher sensitivity across most of the spectrum of average PVCs per hr on the ambulatory ECG record-
Fig. 1. Plot of sensitivity and specificity for the 2-min rhythm strip across average PVCs per hr on the ambulatory ECG.
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497
Fig. 2. ROC curve comparing the 2-min rhythm strip to the average PVCs per hr on the ambulatory ECG.
ing. Specificity was greater for younger individuals across the average PVCs per hr spectrum, but was similar by gender and hypertension status. For MI patients, sensitivity and specificity were lower across the continuum of average PVCs per hr when compared to those without a history of MI, resulting in lower positive likelihood ratios and larger negative likelihood ratios. 4. Discussion The use of a short rhythm strip to detect PVCs would offer considerable advantages if found to be valid because it is easy, inexpensive, and feasible on virtually all people. In previous epidemiologic studies, 2–7% of apparently healthy men had at least one PVC on a resting 2-min rhythm strip [31–33]. Our sample of men and women had a higher prevalence of ventricular ectopy (19%), which was expected because this was a clinical population with a higher prevalence of MI. When evaluating the ability to detect PVCs on a short recording compared to the 24-hr ambulatory ECG, in general shorter recordings had relatively high specificity at the expense of sensitivity, while longer recordings had increased sensitivity with a smaller compromise in specificity. The specificity of the 2-min rhythm strip was moderate to high across the entire distribution of average PVCs per hour on the 24-hr recording. Those with average PVCs on the ambulatory ECG recording less than the gold standard cutoff had a high probability of a negative test across the distribution of average PVCs per hr. When comparing the 2-min rhythm strip to the 24-hr ambulatory ECG monitor, sensitivity was moderate to high across high average PVCs per hr and reduced when detecting low average PVCs per hr. When categorizing a gold standard test as having an average of at least 60 PVCs per hr on the ambulatory ECG monitor, the sensitivity of the 2-min rhythm strip was 76%. Therefore, having a positive 2-min rhythm strip (at least one PVC) was better at confirming, rather than excluding a specific PVCs per hr average over 24 hr.
495
The optimal balance between sensitivity and specificity depends on how the test is to be used. For conditions of low prevalence or in cases where a false positive result will expose a patient to potentially risky testing or treatment, the test cutoff should minimize false positive results. On the ROC curve, this occurs in the lower left portion of the curve [28]. For high prevalence conditions, the test cutoff should minimize false negative results. In this case, the ideal value occurs on the ROC curve in the upper right-hand portion of the curve [28]. To maximize sensitivity and specificity simultaneously, the cutpoint should be chosen at the upper left portion of the ROC curve. In our data, this occurred at approximately 72 PVCs per hr average, yielding 83% sensitivity, 91% specificity, and 90% accuracy. The repeatability of a 2-min rhythm strip as a marker for PVC occurrences over 24 hr was not assessed in our study, but was reported in another cohort. Men aged 35 to 57 years screened for the Multiple Risk Factor Intervention Trial from the Twin Cities, MN, area who were free of coronary artery disease and hypertension participated in this study [23,24]. The 2-min rhythm strip was repeated on 49 men who had PVCs and 49 controls who did not have PVCs on their first rhythm strip. On the second 2-min recording, 22 (45%) men who had PVCs on the first recording also had at least one PVC on the second recording, while only two (4%) men in the control group incurred PVCs. The correlation between the first and second recording was 0.69. Additionally, the 2-min rhythm strip identified most men with frequent or complex PVCs on the ambulatory ECG monitor, while missing participants with infrequent occurrences of PVCs. Similar results were found in our study. The clinical use of the 2-min rhythm strip to identify those with PVCs is constrained by the lack of repeatability and sensitivity, especially when trying to detect infrequent occurrences of PVCs over 24 hr. Repeatability is reduced, because PVCs have a large circadian variation in frequency, which constrain the use of shortened ECG monitoring periods to assess occurrences among individuals [34–44]. The sensitivity of the 2-min rhythm strip was also low, especially when detecting infrequent PVCs per hr. In this study, the sensitivity of the 2-min rhythm strip to detect PVCs ⬎0 per hour over 24 hr was 26% with a corresponding false negative rate of 74%. The sensitivity of the 2-min rhythm strip to detect PVCs ⭓60 per hr over 24 hr was 76% with a corresponding false negative rate of 24%. If the purpose of the screening test is to identify individuals with PVC occurrences (especially of low frequency) over 24 hr, then according to our data, a 2-min recording period is not long enough to adequately characterize individual patients. When the purpose of monitoring PVCs is to identify a group of people at relatively high risk of mortality, then according to Ruberman and colleagues the presence of PVCs on short recordings “serves this epidemiologic objective very well” [12,18]. Numerous epidemiologic studies have utilized a short ECG recording period to examine the relationship between PVCs and mortality outcomes among
496
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497
those with and without coronary heart disease, several of which are referenced here. Among MI patients, recordings of less than 1 min (taken from a 12-lead ECG) [16] and of 1 hr [12,17,18] have been predictive of mortality and sudden death. Data from three studies on MI survivors indicate that using longer monitoring periods, in contrast to 1-hr recordings, to assess PVCs did not offer significant gains in predictive value when used for prediction of 1-year mortality after MI [45–47]. In fact, much of the predictive value of the 24-hr ECG was contained in the first hr of recording [46]. Among men without apparent heart disease over 40 years of age in the Manitoba Study, PVCs occurring during 12-lead ECG recording identified a group of men at high risk for sudden death [22]. Among apparently healthy men screened for the Multiple Risk Factor Intervention Trial, PVCs on a 2-min recording were associated with sudden death [20]. For Framingham Heart Study participants, a 1-hr recording to detect PVCs was predictive of mortality outcomes for those with no apparent coronary heart disease [21], as well as for those with echocardiographic evidence of left ventricular hypertrophy [48]. Short-term recordings yield relatively high specificity at the expense of sensitivity. However, this method may adequately identify people with high frequency PVCs over 24 hr. This study had several limitations. Our data were sparse at the higher end of the average PVCs per hr continuum on the ambulatory ECG monitor. Furthermore, the study population was drawn as a sample of convenience and it is unknown whether it represents other clinical populations, because referral patterns may affect our results [49]. However, ROC curves have been shown by computer simulation to be insensitive to selection bias across a wide range of clinically realistic rates and referral patterns [50]. Re-examination of results according to pre-existing disease has been recommended [51]. We reanalyzed our data, after excluding MI patients, which tended to increase sensitivity and specificity across average PVCs per hr over 24 hr. In this clinical population, utilizing the 2-min rhythm strip as an indicator for average PVCs per hr had excellent specificity and moderate sensitivity across most of the distribution of average PVCs per hr. Epidemiologic research suggests that frequent PVCs may be an important predictor of cardiovascular disease and subsequent mortality [1–22]. It has also been suggested that PVCs frequent enough to occur on a short rhythm strip may carry more prognostic significance than PVCs that require a longer recording to identify [20]. Our data suggest that although a 2-min rhythm strip meets many of the criteria for a good screening test (including high specificity, low cost, safety, objectivity, and ease of use), the lower sensitivity and repeatability of the 2-min rhythm strip to identify individuals with less frequent PVCs is of concern. The 2-min rhythm strip leaves many with recurrent PVCs unidentified due to the length of the monitoring period and the repeatability is compromised due to the large circadian variation of PVCs. However, PVCs on a 2-min rhythm strip are better at identifying those with high frequency PVCs over 24 hr.
The use of a short rhythm strip to detect PVCs, in light of our validation results, may be considered useful in epidemiologic investigations of cardiovascular disease and mortality for detecting high frequency PVCs in populations. The use of a short rhythm strip as a screening tool to detect PVCs in clinical practice is not warranted due to its sensitivity and repeatability, based on our findings and the existing literature. However, an awareness that PVCs on a 2-min rhythm strip consistently identifying high frequency PVCs on 24-hr recordings should be helpful to clinicians. Acknowledgments The authors thank Gerardo Heiss, Gail Hester, Mike Smith, and Sylvia Fletcher for their assistance. This study was conducted while Kelly R. Evenson and Verna L. Lamar Welch were supported by NSRA fellowships through NIH, #2 T32 HL07055 and #5 F31 H109284 respectively.
References [1] Manolio T, Furberg C, Rautaharju P, Siscovick D, Newman A, Borhani N, Gardin J, Tabatznik B. Cardiac arrhythmias on 24-hour ambulatory electrocardiography in older women and men: The Cardiovascular Health Study. J Am Coll Cardiol 1994;23:916–25. [2] Horan M, Kennedy H. Ventricular ectopy. JAMA 1984;251:380–6. [3] Califf R, McKinnis R, Burks J, Lee K, Harrell F Jr, Behar V, Pryor D, Nagner G, Rosati R. Prognostic implications of ventricular arrhythmias during 24 hour ambulatory monitoring in patients undergoing cardiac catheterization for coronary artery disease. Am J Cardiol 1982;50:23–31. [4] Bigger J Jr, Fleiss J, Kleiger R, Miller J, Rolnitzky L, The Multicenter Post-Infarction Research Group. The relationships among ventricular arrhythmias, left ventricular dysfunction, and mortality in the 2 years after myocardial infarction. Circulation 1984;69:250–8. [5] Connolly S, Cairns J, Group tCPS. Prevalence and predictors of ventricular premature complexes in survivors of acute myocardial infarction. Am J Cardiol 1992;69:408–11. [6] Kostis J, Byington R, Friedman L, Goldstein S, Furberg C, BHAT Study Group. Prognostic significance of ventricular ectopic activity in survivors of acute myocardial infarction. J Am Coll Cardiol 1987; 10:231–42. [7] Maggioni A, Zuanetti G, Franzosi M, Rovelli F, Santoro E, Staszewsky L, Tavazzi L, Tognoni G. Prevalence and prognostic significance of ventricular arrhythmias after acute myocardial infarction in the fibrinolytic era. Circulation 1993;87:312–22. [8] Moss A. Clinical significance of ventricular arrhythmias in patients with and without coronary artery disease. Prog Cardiovasc Dis 1980; 23:33–52. [9] Mukharji J, Rude R, Poole W, Gustafson N, Thomas L Jr, Strauss H, Jaffe A, Muller J, Roberts R, Raube D Jr, Croft C, Passomani E, Braunwald E, Willerson J, Group tMS. Risk factors for sudden death after myocardial infarction: two-year follow-up. Am J Cardiol 1984; 54:31–36. [10] Wilson A, Kostis J, BHAT Study Group. The prognostic significance of very low frequency ventricular ectopic activity in survivors of acute myocardial infarction. Chest 1992;102:732–6. [11] Hallstrom A, Bigger J Jr, Roden D, Friedman L, Akiyama T, Richardson D, Rogers W, Waldo A, Pratt C, Capone R, Griffith L, Theroux P, Barker A, Woosley R. Prognostic significance of ventricular premature depolarizations measured 1 year after myocardial infarction in patients with early postinfarction asymptomatic ventricular arrhythmia. J Am Coll Cardiol 1992;20:259–64.
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497 [12] Ruberman W, Weinblatt E, Goldberg J, Frank C, Chaudhary B, Shapiro S. Ventricular premature complexes and sudden death after myocardial infarction. Circulation 1981;64:297–305. [13] Nicod P, Gilpin E, Dittrich H, Henning H, Ross J Jr. Prognostic significance of complex ventricular arrhythmia for cardiac death during the first year after myocardial infarction. J Electrophysiology 1987;1:93–102. [14] Statters D, Malik M, Redwood S, Hnatkova K, Staunton A, Camm A. Use of ventricular premature complexes for risk stratification after acute myocardial infarction in the thrombolytic era. Am J Cardiol 1996;77:133–8. [15] The Multicenter Post-Infarction Research Group. Risk stratification and survival after myocardial infarction. N Engl J Med 1983;309: 331–6. [16] The Coronary Drug Project Research Group. Prognostic importance of premature beats following myocardial infarction. JAMA 1973;223: 1116–24. [17] Ruberman W, Weinblatt E, Goldberg J, Frank C, Shapiro S. Ventricular premature beats and mortality after myocardial infarction. N Engl J Med 1977;297:750–7. [18] Ruberman W, Weinblatt E, Frank C, Goldberg J, Shapiro S. Repeated 1 hour electrocardiographic monitoring of survivors of myocardial infarction at 6 month intervals: arrhythmia detection and relation to prognosis. Am J Cardiol 1981;47:1197–204. [19] Cats V, Lie K, Van Capelle F, Durrer O. Limitations of 24 hour ambulatory electrocardiographic recording in predicting coronary events after acute myocardial infarction. Am J Cardiol 1979;44:1257–62. [20] Abdalla I, Prineas R, Neaton J, Jacobs D Jr, Crow R. Relation between ventricular premature complexes and sudden cardiac death in apparently healthy men. Am J Cardiol 1987;60:1036–42. [21] Bikkina M, Larson M, Levy D. Prognostic implications of asymptomatic ventricular arrhythmias: The Framingham Heart Study. Ann Int Med 1992;117:990–6. [22] Rabkin S, Mathewson A, Tate R. Relationship of ventricular ectopy in men without apparent heart disease to occurrence of ischemic heart disease and sudden death. Am Heart J 1981;101:135–42. [23] De Backer G, Jacobs D Jr, Prineas R, Crow R, Kennedy H, Vilandre J, Blackburn H. Ventricular premature beats: screening and induction tests in normal men. Cardiology 1980;65:23–41. [24] De Backer G, Jacobs D, Prineas R, Crow R, Vilandre J, Blackburn H. Ventricular premature beats; reliability in various measurement methods at rest and during exercise. Cardiology 1978;63:53–63. [25] Moons K, van Es G, Deckers J, Habbema J, Grobbee D. Limitations of sensitivity, specificity, likelihood ratio, and Bayes’ theorem in assessing diagnostic probabilities: a clinical example. Epidemiology 1997;8:12–7. [26] Simel D, Samsa G, Matchar D. Likelihood ratios with confidence: sample size estimation for diagnostic test studies. J Clin Epidemiol 1991;44:763–70. [27] Fletcher R, Fletcher S, Wagner E. Clinical Epidemiology. Baltimore, MD: Williams and Wilkins, 1996. [28] Hagen M. Test characteristics: how good is that test? Med Decision Making 1995;22:213–33. [29] Hanley J, McNeil B. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983;148:839–43. [30] Hanley J, McNeil B. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:29–36. [31] Crow R, Prineas R, Dias V, Taylor H, Jacobs D, Blackburn H. Ventricular premature beats in a population sample. Circulation 1975;5152:211–5. [32] Elkon K, Swerdlow T, Myburgh D. Persistent ventricular ectopic beats. South African Med J 1977;52:564–6. [33] Fisher F, Tyroler H. Relationship between ventricular premature con-
[34]
[35] [36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49] [50]
[51]
497
tractions on routine electrocardiography and subsequent sudden death from coronary heart disease. Circulation 1973;47:712–9. Pratt C, Deklos G, Wierman A, Mahler S, Seals A, Leon C, Young J, Quinones M, Roberts R. The changing base line of complex ventricular arrhythmias. N Engl J Med 1985;313:1444–9. Winkle R. Antiarrhythmic drug effect mimicked by spontaneous variability of ventricular ectopy. Circulation 1978;57:1116–21. Sami M, Kraemer H, Harrison D, Houston N, Shimasaki C, DeBusk R. A new method for evaluating antiarrhythmic drug efficacy. Circulation 1980;62:1172–9. Raeder E, Hohnloser S, Graboys T, Podrid P, Lampert S, Lown B. Spontaneous variability and circadian distribution of ectopic activity in patients with malignant ventricular arrhythmia. J Am Coll Cardiol 1988;12:656–61. Morganroth J, Michelson E, Horowitz L, Josephson M, Pearlman A, Dunkman W. Limitations of routine long-term electrocardiographic monitoring to assess ventricular ectopic frequency. Circulation 1978; 58:408–14. Morganroth J. Evaluation of arrhythmias in clinical trials: relationship to the spontaneous variability of ventricular arrhythmias. In: Wenger N, Mock M, Ringqvist I, editors. Ambulatory Electrocardiographic Recording. Chicago, IL: Year Book Medical Publishers, Inc., 1980 pp. 285–98. Michelson E, Morganroth J. Spontaneous variability of complex ventricular arrhythmias detected by long-term electrocardiographic recording. Circulation 1980;61:690–5. Moller M. Reliability of serial 24 hour ambulatory electrocardiography in predicting cardiac death after myocardial infarction. Eur Heart J 1982;3:67–74. Bigger J Jr, Rolnitzky L, Leahey E Jr, LaPook J. Duration of recording: activity protocol. In: Wenger N, Mock M, Ringqvist I, editors. Ambulatory Electrocardiographic Recording. Chicago, IL: Year Book Medical Publishers Inc., 1980 pp. 87–102. Kennedy H, Chandra V, Sayther K, Caralis D. Effectiveness of increasing hours of continuous ambulatory electrocardiography in detecting maximal ventricular ectopy. Am J Cardiol 1978;42:925–30. Thanavaro S, Kleiger R, Hieb B, Krone R, DeMello V, Oliver G. Effect of electrocardiographic recording duration on ventricular dysrhythmia detection after myocardial infarction. Circulation 1980;62:262–5. Davis B, Friedman L, Lichstein E, for the Beta-Blocker Heart Attack Trial Study Group. Are 24 hours of ambulatory ECG monitoring necessary for a patient after infarction? Am Heart J 1988;115:83–91. Connolly S, Claims J, on behalf of the CAMIAT Pilot Study Group. Comparison of one-, six-, and 24-hour ambulatory electrocardiographic monitoring for ventricular arrhythmia as a predictor of mortality in survivors of acute myocardial infarction. Am J Cardiol 1992; 69:308–13. Petretta M, Bianchi V, Pulcino A, Carpinelli A, Valva G, Themistodakis S, Attisaro T, Salemme L, Bonaduce D. Continuous electrocardiographic monitoring for more than one hour does not improve the prognostic value of ventricular arrhythmias in survivors of first acute myocardial infarction. Am J Cardiol 1994;73:139–42. Bikkina M, Larson M, Levy D. Asymptomatic ventricular arrhythmias and mortality risk in subjects with left ventricular hypertrophy. J Am Coll Cardiol 1993;22:1111–6. Knottnerus J, Leffers P. The influence of referral patterns on the characteristics of diagnostic tests. J Clin Epidemiol 1992;45:1143–54. Diamond G. Affirmative actions: can the discriminant accuracy of a test be determined in the face of selection bias? Med Decision Making 1991;11:48–56. Knottnerus J. The effect of disease verification and referral on the relationship between symptoms and diseases. Med Decision Making 1987;7:139–48.
Validation of a short rhythm strip compared to ambulatory ECG monitoring for ventricular ectopy Kelly R. Evensona,*, Verna L. Lamar Welcha, Wayne E. Casciob, Ross J. Simpson Jr.b a
Department of Epidemiology, School of Public Health, University of North Carolina–Chapel Hill, 137 East Franklin Street, Bank of America Center, Suite 306, Chapel Hill, NC 27514, USA; b Preventive Cardiology Program, Division of Cardiology, Department of Medicine, University of North Carolina–Chapel Hill, Chapel Hill, NC 27599-7075, USA Received 4 August 1998; received in revised form 11 June 1999; accepted 17 September 1999
Abstract Premature ventricular contractions (PVCs) are associated with an increased risk of cardiovascular disease and mortality. Many epidemiologic studies measure a continuous short rhythm strip to ascertain PVCs as a screening tool to identify persons at highest risk. Despite its widespread use in epidemiologic studies, the rhythm strip has not been completely validated. Therefore, a continuous 2-min rhythm strip was measured on 242 consecutive individuals referred for ambulatory ECG monitoring. Prevalence of at least one PVC on the 2-min rhythm strip was compared to a gold standard, the average number of PVCs per hr on ambulatory recording. The prevalence of any PVCs on the 2-min rhythm strip was 19%. As average PVCs per hr increased on the ambulatory ECG recording, sensitivity increased while specificity slowly decreased. Sensitivity ranged from 26–100% and specificity ranged from 81–100% across the distribution of average PVCs per hr on ambulatory monitoring. Area under the receiver operator characteristic (ROC) curve of the 2-min rhythm strip compared to 24-hr results was 0.943. Area under ROC curves were not statistically different (P ⬎ 0.05) by age, gender, hypertension status, or history of myocardial infarction. In this clinical population, utilizing the 2-min rhythm strip as an indicator of average PVCs per hr had excellent specificity and moderate to low sensitivity across most of the distribution of average PVCs per hr. The use of a short rhythm strip to detect PVCs may be considered useful in epidemiologic investigations of cardiovascular disease and mortality for detecting high frequency PVCs in populations. The use of a short rhythm strip as a screening tool to detect PVCs in clinical practice is not warranted, based on our findings and the existing literature. However, an awareness that PVCs on a 2-min rhythm strip consistently identify high frequency PVCs on 24-hr recordings should be helpful to clinicians. © 2000 Elsevier Science Inc. All rights reserved. Keywords: Ventricular premature complexes; Electrocardiography; Arrhythmia; Ambulatory electrocardiography; Epidemiology; Validation
1. Introduction Occurrences of premature ventricular contractions (PVCs) on ambulatory electrocardiograph (ECG) are associated with an increased risk of subclinical [1] and clinical cardiovascular disease [2,3], as well as mortality in post-myocardial infarction (MI) patients [2,4–19]. In apparently healthy men, PVCs on rhythm strip recordings as short as 2 min were correlated with an increased risk of sudden cardiac death [20] and complex or frequent ventricular arrhythmias from a 1-hr recording were associated with all-cause and coronary heart disease mortality [21]. Additionally, for men at least 40 years of age without apparent heart disease, PVCs on routine ECG were related to an increased risk of ischemic heart disease, especially sudden cardiac death [22]. * Corresponding author. Tel.: 919-966-1967; fax: 919-966-9800. E-mail address: [email protected] (K.R. Evenson)
Many cardiovascular epidemiologic studies measure a continuous short rhythm strip to ascertain PVCs as a screening tool to identify high-risk individuals. Despite its widespread use in epidemiologic studies, the rhythm strip has not been completely validated. Validity of a short rhythm strip was reported only for men participating in a single cohort study [23,24]. Therefore, the main objectives of this study were to validate the ability to the standard 2-min rhythm strip to detect PVCs relative to ambulatory ECG monitoring and to describe the relationship between patient characteristics and the occurrence to PVCs on the 2-min rhythm strip. 2. Methods Inpatients and outpatients referred to the University of North Carolina Memorial Hospital in Chapel Hill, NC, for a clinically indicated continuous ambulatory ECG monitor from July 17, 1995 to January 26, 1996 were consecutively
0895-4356/00/$ – see front matter © 2000 Elsevier Science Inc. All rights reserved. PII: S0895-4356(99)00 1 9 0 - 0
492
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497
enrolled in this cross-sectional study. A 2-min continuous rhythm strip was obtained at the same time the individual began ambulatory ECG monitoring. Participants were excluded if they were less than 18 years of age, institutionalized, had a pacemaker, or had any of the following findings on the rhythm strip: sick sinus syndrome (heart rate less than 50 beats per min), atrial fibrillation, or atrial flutter. Ambulatory ECG monitoring was indicated for the following conditions: 22% syncope or dizziness, 18% angina or ischemia, 19% palpitations, and 41% arrhythmias. Only the participant’s first-time visit during the study period contributed to these analyses. 2.1. Rhythm strip With the patient lying supine, electrode position for the 2-min rhythm strip was determined using standard criteria to best identify P waves. A modified chest lead two (MCL2) was most often used. Two cardiologists independently evaluated the 2-min rhythm strip to determine the number of PVCs. In case of disagreement on the number of PVCs, a cardiologist adjucated the discrepancy. Before adjucation, agreement between the two cardiologists on the presence of any PVCs on the 2-min rhythm strip was evaluated, yielding a kappa coefficient of 98.7%. 2.2. Ambulatory ECG monitoring The simultaneous two-lead system (MCL2 and modified V5) was used for ambulatory ECG monitoring. The ambulatory ECG tape was read according to a standard, partially automated protocol by one of two trained ECG technicians masked to the rhythm strip results. For the 252 participants, the average length of ambulatory ECG recording was 22.7 hr. To avoid possible bias, only complete hr from the ambulatory ECG were used in these analyses, assuming for this purpose that PVCs occurrences were uniform over each hr. The 1-, 2-, and 3-hr results were taken from the first 1, 2, and 3 hr, respectively, of the monitoring tape. 2.3. A priori definitions For the 1- and 2-min rhythm strip, a positive test was indicated if at least one PVC occurred during the recording. For the 1-, 2-, and 3-hr results, a positive test was indicated if there was an average of at least 10 PVCs per hr during the 1, 2, or 3 hr of recording, respectively. The gold standard for validation was the average number of PVCs per hr from all complete hr of ambulatory ECG monitoring. To examine complex ectopy, we defined the gold standard in two ways: (1) presence of any ventricular triplets or tachycardia on ambulatory ECG; and (2) presence of any ventricular couplets, triplets, or tachycardia on ambulatory ECG. Hypertension and history of MI, as well as smoking and ingestion of any food or caffeine in the 2 hr prior to the rhythm strip recording were self-reported and logged by the ECG technicians for this study. Age and ethnicity were obtained from the medical record.
2.4. Statistical analyses Categorical analyses were performed using Mantel-Haenszel chi-square tests. Median tests were calculated to determine whether median average PVCs per hr from ambulatory ECG recordings differed significantly by various factors. Because sensitivity is inversely related to specificity, the likelihood ratio has been proposed to be more stable [25]. The pretest odds multiplied by the likelihood ratio provides the post-test odds of disease. Sensitivity, specificity, positive likelihood ratios [sensitivity/(1-specificity)], and negative likelihood ratios [(1sensitivity)/specificity] were calculated, along with their 95% confidence intervals (CI) [26]. Graphs of these were truncated at 600 average PVCs per hr from ambulatory recording due to a reduced sample size beyond this cutpoint. Accuracy was calculated as the correct number of classified patients (both positive and negative) divided by the total number of patients [27]. Unconditional multivariable logistic regression was used to identify independent predictors of PVCs on the 2-min rhythm strip. The full model included age, gender, ethnicity, hypertension, MI, and both minimum and maximum heart rate from ambulatory ECG monitoring. Both age and heart rate were left continuous in the model. Receiver operating characteristic (ROC) curves can efficiently quantify the relationship between true positive (sensitivity) and false positive (1-specificity) rates for tests with continuous outcomes [28]. Area under the ROC curve was derived from unconditional logistic regression, modeling the average number of PVCs per hr from the ambulatory ECG monitor as the independent variable and whether or not PVCs occurred on the rhythm strip as the dependent variable. Area under ROC curves were used to compare the 2-min rhythm strip to the 1-min rhythm strip as well as to the first 1-, 2-, and 3-hr recording from the ambulatory ECG, taking into account correlation between samples [29]. Additionally, area under ROC curves was utilized to compare 2-min rhythm strip results for groups separately stratified by gender, age, hypertension, and MI status. Because these data were uncorrelated, the method described by Hanley and McNeil [30] was used for comparison. For all ROC curve comparisons, the gold standard was the average number of PVCs per hr from all complete hr of ambulatory ECG monitoring.
3. Results Among 252 participants, the prevalence of any PVCs on the 2-min rhythm strip was 19%, ranging from 0 to 53 PVCs. Less than 2% of participants had one or more ventricular couplets, and there were no occurrences of ventricular triplets or ventricular tachycardia on the 2-min rhythm strip. The average number of PVCs per hr over 24-hr on the ambulatory ECG monitor was 49.2 (median 0.38, range 0–894). Average PVCs per hr on the ambulatory ECG monitor was 198.2 (median 101.6, range 2.2–894) for those with PVCs on the rhythm and 14.1 (median 0.17, range 0–605) for those without PVCs on the rhythm strip. Table 1 displays
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497 Table 1 Average PVCs per hr from ambulatory ECG monitoring for the study population stratified by presence or absence of PVCs on the 2-min rhythm strip Average PVCs/hr on ambulatory ECG
PVCs on 2-min rhythm strip Present (n ⫽ 48)
Absent (n ⫽ 204)
0 PVCs/hr ⬎0–⬍0.5 PVCs/hr ⭓0.5–⬍5 PVCs/hr ⭓5–⬍10 PVCs/hr ⭓10–⬍60 PVCs/hr ⭓60 PVCs/hr
0 (0%) 0 (0%) 1 (2%) 5 (10%) 11 (23%) 31 (65%)
66 (32%) 65 (32%) 35 (17%) 6 (3%) 22 (11%) 10 (5%)
the distribution of average PVCs per hr from ambulatory ECG monitoring according to presence or absence of any PVCs on the 2-min rhythm strip. Ten participants had an average PVCs per hr of at least 60, but did not have any PVCs on the 2-min rhythm strip. Table 2 displays the bivariate relationship between patient characteristics and cardiovascular risk factors to the prevalence of PVCs on the 2-min rhythm strip. Prevalence of PVCs increased with age (P ⬍ 0.01). Patients with a history of MI (P ⬍ 0.01) were also more likely to manifest PVCs on the rhythm strip. PVC occurrences were not significantly different by gender, ethnicity, or hypertension (P ⬎ 0.05). These results were similar when evaluating average PVCs per hr from the 24-hr monitor to patient characteristics. When simultaneously controlling for gender, ethnicity, hypertension, MI, minimum and maximum 24-hr heart rate, older age was a statistically significant predictor of PVCs
Table 2 Prevalence of PVCs on the 2-min rhythm strip and median average PVCs per hr from ambulatory ECG monitoring according to patient characteristics Patient characteristics Age in years 18–29 30–39 40–49 50–59 60–69 70–79 80–89 Gender Male Female Ethnicity White Nonwhite Hypertension Yes No Previous MI Yes No
Prevalence of PVCs on rhythm strip (%)
Median average PVCs/hour over 24 hr
4 12 11 17 15 43 53
0.02 0.17 0.24 0.17 1.25 27.70 14.91
21 17
1.00 0.25
20 17
0.37 0.38
19 18
0.87 0.26
32 16
29.77 0.17
493
on the 2-min rhythm strip (prevalence odds ratio 1.06, 95% CI 1.03–1.09). The remaining factors in the logistic model were not statistically significant predictors of PVCs on the 2-min rhythm strip. Table 3 displays the sensitivity and specificity of the 2-min rhythm strip for several average PVCs per hr cutpoints and Fig. 1 displays the entire range graphically. As average PVCs per hr increased, sensitivity increased while specificity decreased gradually. To examine whether our results were affected by those with MI, we re-examined sensitivity and specificity excluding those individuals with MI. Specificity did not meaningfully change from 0 to 60 average PVCs per hr while sensitivity increased above 0.5 average PVCs per hr. Repeat analyses were also performed, after excluding 76 individuals who ate, ingested caffeine, or smoked two hr prior to the ECG recording. Sensitivity and specificity of the 2-min rhythm strip to detect 0 to 60 average PVCs per hr did not meaningfully change. This was also found when excluding 28 individuals who had a premature atrial contraction during the rhythm strip recording. The proportion of individuals classified incorrectly as negative (false negative) on the 2-min rhythm strip ranged from 74% at the gold standard cutoff of ⬎0 average PVCs per hr to 24% at ⭓60 average PVCs per hr on ambulatory ECG monitoring. The proportion of individuals classified incorrectly as positive (false positive) on the 2-min rhythm strip ranged from 0% at a gold standard cutoff of ⬎0 average PVCs per hr to 8% at ⭓60 average PVCs per hr on ambulatory ECG monitoring. During 24-hr ECG monitoring, 30% (75) had at least one ventricular couplet and 13% (32) had at least one ventricular triplet or one run of ventricular tachycardia. Those with any evidence of complex ventricular ectopy (e.g., ventricular tachycardia, ventricular triplets, ventricular couplets) during ambulatory ECG monitoring were more likely to incur PVCs on the 2-min recording. Among those with any ventricular couplets on the 24-hr recording, 52% had at least one PVC on the 2-min rhythm strip. Similarly, among those with any triplets or ventricular tachycardia on the 24-hr recording, 71% and 74%, respectively, had at least one PVC on the 2-min rhythm strip. The ROC curve comparing the 2-min rhythm strip to 24hr results is displayed in Fig. 2. The area under this ROC curve was 0.943, indicating that the test did far better than chance alone in identifying average PVCs per hr. To compare other recording times to the 2-min rhythm strip, ROC curves were compared statistically. When comparing a ROC curve for a 1-min (area 0.919) versus a 2-min rhythm strip to 24-hr monitoring, differences between the two were not statistically significant (P ⬎ 0.05). When comparing a 2-min rhythm strip to a 1-hr recording to detect PVCs over 24 hr, the 1-hr recording (area 0.978) showed improvement, although the differences did not reach significance (P ⬎ 0.05). Similarly, the 2-hr (area 0.982) and 3-hr recording (area 0.984) improved upon the 2-min recording, but neither
494
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497
Table 3 Sensitivity and specificity with 95% confidence intervals (CI) comparing the 2-min rhythm strip to 24-hr monitoring for all participants and for those without a myocardial infarction (MI), and median PVCs per hr over 24 hr for the study population Gold standard from ambulatory ECG Average PVCs/hour ⬎0
⭓0.5 ⭓5 ⭓10 ⭓15 ⭓30 ⭓60 Any ventricular triplets or tachycardia Any ventricular couplets, triplets, or tachycardia
Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity Sensitivity Specificity
All (95% CI) (n ⫽ 252)
Without MI (95% CI) (n ⫽ 215)
26% (20–32) 100% 40% (31–48) 100% 55% (45–66) 99% (98–100) 57% (45–68) 97% (94–99) 57% (46–69) 95% (92–98) 71% (58–83) 94% (91–97) 76% (62–89) 92% (88–96) 66% (49–82) 88% (83–92) 49% (38–60) 95% (91–98)
24% (17–30) 100% 40% (30–50) 100% 57% (45–70) 99% (98–100) 59% (45–72) 96% (93–99) 61% (47–75) 95% (92–98) 76% (62–91) 94% (91–98) 85% (71–98) 93% (89–96) 58% (39–78) 88% (84–93) 47% (35–60) 95% (91–98)
reached statistical significance (P ⬎ 0.05). With increasing recording time on the screening test, sensitivity increased while specificity decreased. Because sensitivity and specificity can vary by subgroups within a population [25], we re-examined the comparison of the 2-min rhythm strip to continuous ambulatory ECG results stratifying separately on gender, age, hypertension, and history of MI. Area under ROC curves were not statistically different by gender (men 0.934 vs. women 0.053, P ⬎ 0.05), age (⭓50 years 0.931 vs. ⬍50 years 0.964, P ⬎ 0.05), or hypertension status (hypertension 0.937 vs. normotension 0.946, P ⬎ 0.05). Because the choice of age categories may have affected our results, we included additional cutpoints in our calculations. Area under ROC curves were still not statistically different by age, al-
Median average PVCs/hr over 24 hr (n ⫽ 252) 2.69 17.00 56.69 70.10 81.11 148.11 206.17 126.92 36.75
though at the highest age cutpoint there was an indication of reduced area under the ROC curve (⭓70 years 0.876 vs. ⬍70 years 0.952, P ⬎ 0.05). ROC curve comparisons between those with and without a history of MI were not statistically significant (MI 0.827 vs. no MI 0.960, P ⫽ 0.13). We may have lacked adequate statistical power to detect differences within these groups, because our sample included only 47 patients at least 70 years of age and 37 patients with a history of MI. We also examined the data separately for sensitivity and specificity across the average PVCs per hr continuum, because sensitivity and specificity can vary independently. Women, older individuals, and those with hypertension tended to have higher sensitivity across most of the spectrum of average PVCs per hr on the ambulatory ECG record-
Fig. 1. Plot of sensitivity and specificity for the 2-min rhythm strip across average PVCs per hr on the ambulatory ECG.
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497
Fig. 2. ROC curve comparing the 2-min rhythm strip to the average PVCs per hr on the ambulatory ECG.
ing. Specificity was greater for younger individuals across the average PVCs per hr spectrum, but was similar by gender and hypertension status. For MI patients, sensitivity and specificity were lower across the continuum of average PVCs per hr when compared to those without a history of MI, resulting in lower positive likelihood ratios and larger negative likelihood ratios. 4. Discussion The use of a short rhythm strip to detect PVCs would offer considerable advantages if found to be valid because it is easy, inexpensive, and feasible on virtually all people. In previous epidemiologic studies, 2–7% of apparently healthy men had at least one PVC on a resting 2-min rhythm strip [31–33]. Our sample of men and women had a higher prevalence of ventricular ectopy (19%), which was expected because this was a clinical population with a higher prevalence of MI. When evaluating the ability to detect PVCs on a short recording compared to the 24-hr ambulatory ECG, in general shorter recordings had relatively high specificity at the expense of sensitivity, while longer recordings had increased sensitivity with a smaller compromise in specificity. The specificity of the 2-min rhythm strip was moderate to high across the entire distribution of average PVCs per hour on the 24-hr recording. Those with average PVCs on the ambulatory ECG recording less than the gold standard cutoff had a high probability of a negative test across the distribution of average PVCs per hr. When comparing the 2-min rhythm strip to the 24-hr ambulatory ECG monitor, sensitivity was moderate to high across high average PVCs per hr and reduced when detecting low average PVCs per hr. When categorizing a gold standard test as having an average of at least 60 PVCs per hr on the ambulatory ECG monitor, the sensitivity of the 2-min rhythm strip was 76%. Therefore, having a positive 2-min rhythm strip (at least one PVC) was better at confirming, rather than excluding a specific PVCs per hr average over 24 hr.
495
The optimal balance between sensitivity and specificity depends on how the test is to be used. For conditions of low prevalence or in cases where a false positive result will expose a patient to potentially risky testing or treatment, the test cutoff should minimize false positive results. On the ROC curve, this occurs in the lower left portion of the curve [28]. For high prevalence conditions, the test cutoff should minimize false negative results. In this case, the ideal value occurs on the ROC curve in the upper right-hand portion of the curve [28]. To maximize sensitivity and specificity simultaneously, the cutpoint should be chosen at the upper left portion of the ROC curve. In our data, this occurred at approximately 72 PVCs per hr average, yielding 83% sensitivity, 91% specificity, and 90% accuracy. The repeatability of a 2-min rhythm strip as a marker for PVC occurrences over 24 hr was not assessed in our study, but was reported in another cohort. Men aged 35 to 57 years screened for the Multiple Risk Factor Intervention Trial from the Twin Cities, MN, area who were free of coronary artery disease and hypertension participated in this study [23,24]. The 2-min rhythm strip was repeated on 49 men who had PVCs and 49 controls who did not have PVCs on their first rhythm strip. On the second 2-min recording, 22 (45%) men who had PVCs on the first recording also had at least one PVC on the second recording, while only two (4%) men in the control group incurred PVCs. The correlation between the first and second recording was 0.69. Additionally, the 2-min rhythm strip identified most men with frequent or complex PVCs on the ambulatory ECG monitor, while missing participants with infrequent occurrences of PVCs. Similar results were found in our study. The clinical use of the 2-min rhythm strip to identify those with PVCs is constrained by the lack of repeatability and sensitivity, especially when trying to detect infrequent occurrences of PVCs over 24 hr. Repeatability is reduced, because PVCs have a large circadian variation in frequency, which constrain the use of shortened ECG monitoring periods to assess occurrences among individuals [34–44]. The sensitivity of the 2-min rhythm strip was also low, especially when detecting infrequent PVCs per hr. In this study, the sensitivity of the 2-min rhythm strip to detect PVCs ⬎0 per hour over 24 hr was 26% with a corresponding false negative rate of 74%. The sensitivity of the 2-min rhythm strip to detect PVCs ⭓60 per hr over 24 hr was 76% with a corresponding false negative rate of 24%. If the purpose of the screening test is to identify individuals with PVC occurrences (especially of low frequency) over 24 hr, then according to our data, a 2-min recording period is not long enough to adequately characterize individual patients. When the purpose of monitoring PVCs is to identify a group of people at relatively high risk of mortality, then according to Ruberman and colleagues the presence of PVCs on short recordings “serves this epidemiologic objective very well” [12,18]. Numerous epidemiologic studies have utilized a short ECG recording period to examine the relationship between PVCs and mortality outcomes among
496
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497
those with and without coronary heart disease, several of which are referenced here. Among MI patients, recordings of less than 1 min (taken from a 12-lead ECG) [16] and of 1 hr [12,17,18] have been predictive of mortality and sudden death. Data from three studies on MI survivors indicate that using longer monitoring periods, in contrast to 1-hr recordings, to assess PVCs did not offer significant gains in predictive value when used for prediction of 1-year mortality after MI [45–47]. In fact, much of the predictive value of the 24-hr ECG was contained in the first hr of recording [46]. Among men without apparent heart disease over 40 years of age in the Manitoba Study, PVCs occurring during 12-lead ECG recording identified a group of men at high risk for sudden death [22]. Among apparently healthy men screened for the Multiple Risk Factor Intervention Trial, PVCs on a 2-min recording were associated with sudden death [20]. For Framingham Heart Study participants, a 1-hr recording to detect PVCs was predictive of mortality outcomes for those with no apparent coronary heart disease [21], as well as for those with echocardiographic evidence of left ventricular hypertrophy [48]. Short-term recordings yield relatively high specificity at the expense of sensitivity. However, this method may adequately identify people with high frequency PVCs over 24 hr. This study had several limitations. Our data were sparse at the higher end of the average PVCs per hr continuum on the ambulatory ECG monitor. Furthermore, the study population was drawn as a sample of convenience and it is unknown whether it represents other clinical populations, because referral patterns may affect our results [49]. However, ROC curves have been shown by computer simulation to be insensitive to selection bias across a wide range of clinically realistic rates and referral patterns [50]. Re-examination of results according to pre-existing disease has been recommended [51]. We reanalyzed our data, after excluding MI patients, which tended to increase sensitivity and specificity across average PVCs per hr over 24 hr. In this clinical population, utilizing the 2-min rhythm strip as an indicator for average PVCs per hr had excellent specificity and moderate sensitivity across most of the distribution of average PVCs per hr. Epidemiologic research suggests that frequent PVCs may be an important predictor of cardiovascular disease and subsequent mortality [1–22]. It has also been suggested that PVCs frequent enough to occur on a short rhythm strip may carry more prognostic significance than PVCs that require a longer recording to identify [20]. Our data suggest that although a 2-min rhythm strip meets many of the criteria for a good screening test (including high specificity, low cost, safety, objectivity, and ease of use), the lower sensitivity and repeatability of the 2-min rhythm strip to identify individuals with less frequent PVCs is of concern. The 2-min rhythm strip leaves many with recurrent PVCs unidentified due to the length of the monitoring period and the repeatability is compromised due to the large circadian variation of PVCs. However, PVCs on a 2-min rhythm strip are better at identifying those with high frequency PVCs over 24 hr.
The use of a short rhythm strip to detect PVCs, in light of our validation results, may be considered useful in epidemiologic investigations of cardiovascular disease and mortality for detecting high frequency PVCs in populations. The use of a short rhythm strip as a screening tool to detect PVCs in clinical practice is not warranted due to its sensitivity and repeatability, based on our findings and the existing literature. However, an awareness that PVCs on a 2-min rhythm strip consistently identifying high frequency PVCs on 24-hr recordings should be helpful to clinicians. Acknowledgments The authors thank Gerardo Heiss, Gail Hester, Mike Smith, and Sylvia Fletcher for their assistance. This study was conducted while Kelly R. Evenson and Verna L. Lamar Welch were supported by NSRA fellowships through NIH, #2 T32 HL07055 and #5 F31 H109284 respectively.
References [1] Manolio T, Furberg C, Rautaharju P, Siscovick D, Newman A, Borhani N, Gardin J, Tabatznik B. Cardiac arrhythmias on 24-hour ambulatory electrocardiography in older women and men: The Cardiovascular Health Study. J Am Coll Cardiol 1994;23:916–25. [2] Horan M, Kennedy H. Ventricular ectopy. JAMA 1984;251:380–6. [3] Califf R, McKinnis R, Burks J, Lee K, Harrell F Jr, Behar V, Pryor D, Nagner G, Rosati R. Prognostic implications of ventricular arrhythmias during 24 hour ambulatory monitoring in patients undergoing cardiac catheterization for coronary artery disease. Am J Cardiol 1982;50:23–31. [4] Bigger J Jr, Fleiss J, Kleiger R, Miller J, Rolnitzky L, The Multicenter Post-Infarction Research Group. The relationships among ventricular arrhythmias, left ventricular dysfunction, and mortality in the 2 years after myocardial infarction. Circulation 1984;69:250–8. [5] Connolly S, Cairns J, Group tCPS. Prevalence and predictors of ventricular premature complexes in survivors of acute myocardial infarction. Am J Cardiol 1992;69:408–11. [6] Kostis J, Byington R, Friedman L, Goldstein S, Furberg C, BHAT Study Group. Prognostic significance of ventricular ectopic activity in survivors of acute myocardial infarction. J Am Coll Cardiol 1987; 10:231–42. [7] Maggioni A, Zuanetti G, Franzosi M, Rovelli F, Santoro E, Staszewsky L, Tavazzi L, Tognoni G. Prevalence and prognostic significance of ventricular arrhythmias after acute myocardial infarction in the fibrinolytic era. Circulation 1993;87:312–22. [8] Moss A. Clinical significance of ventricular arrhythmias in patients with and without coronary artery disease. Prog Cardiovasc Dis 1980; 23:33–52. [9] Mukharji J, Rude R, Poole W, Gustafson N, Thomas L Jr, Strauss H, Jaffe A, Muller J, Roberts R, Raube D Jr, Croft C, Passomani E, Braunwald E, Willerson J, Group tMS. Risk factors for sudden death after myocardial infarction: two-year follow-up. Am J Cardiol 1984; 54:31–36. [10] Wilson A, Kostis J, BHAT Study Group. The prognostic significance of very low frequency ventricular ectopic activity in survivors of acute myocardial infarction. Chest 1992;102:732–6. [11] Hallstrom A, Bigger J Jr, Roden D, Friedman L, Akiyama T, Richardson D, Rogers W, Waldo A, Pratt C, Capone R, Griffith L, Theroux P, Barker A, Woosley R. Prognostic significance of ventricular premature depolarizations measured 1 year after myocardial infarction in patients with early postinfarction asymptomatic ventricular arrhythmia. J Am Coll Cardiol 1992;20:259–64.
K.R. Evenson et al. / Journal of Clinical Epidemiology 53 (2000) 491–497 [12] Ruberman W, Weinblatt E, Goldberg J, Frank C, Chaudhary B, Shapiro S. Ventricular premature complexes and sudden death after myocardial infarction. Circulation 1981;64:297–305. [13] Nicod P, Gilpin E, Dittrich H, Henning H, Ross J Jr. Prognostic significance of complex ventricular arrhythmia for cardiac death during the first year after myocardial infarction. J Electrophysiology 1987;1:93–102. [14] Statters D, Malik M, Redwood S, Hnatkova K, Staunton A, Camm A. Use of ventricular premature complexes for risk stratification after acute myocardial infarction in the thrombolytic era. Am J Cardiol 1996;77:133–8. [15] The Multicenter Post-Infarction Research Group. Risk stratification and survival after myocardial infarction. N Engl J Med 1983;309: 331–6. [16] The Coronary Drug Project Research Group. Prognostic importance of premature beats following myocardial infarction. JAMA 1973;223: 1116–24. [17] Ruberman W, Weinblatt E, Goldberg J, Frank C, Shapiro S. Ventricular premature beats and mortality after myocardial infarction. N Engl J Med 1977;297:750–7. [18] Ruberman W, Weinblatt E, Frank C, Goldberg J, Shapiro S. Repeated 1 hour electrocardiographic monitoring of survivors of myocardial infarction at 6 month intervals: arrhythmia detection and relation to prognosis. Am J Cardiol 1981;47:1197–204. [19] Cats V, Lie K, Van Capelle F, Durrer O. Limitations of 24 hour ambulatory electrocardiographic recording in predicting coronary events after acute myocardial infarction. Am J Cardiol 1979;44:1257–62. [20] Abdalla I, Prineas R, Neaton J, Jacobs D Jr, Crow R. Relation between ventricular premature complexes and sudden cardiac death in apparently healthy men. Am J Cardiol 1987;60:1036–42. [21] Bikkina M, Larson M, Levy D. Prognostic implications of asymptomatic ventricular arrhythmias: The Framingham Heart Study. Ann Int Med 1992;117:990–6. [22] Rabkin S, Mathewson A, Tate R. Relationship of ventricular ectopy in men without apparent heart disease to occurrence of ischemic heart disease and sudden death. Am Heart J 1981;101:135–42. [23] De Backer G, Jacobs D Jr, Prineas R, Crow R, Kennedy H, Vilandre J, Blackburn H. Ventricular premature beats: screening and induction tests in normal men. Cardiology 1980;65:23–41. [24] De Backer G, Jacobs D, Prineas R, Crow R, Vilandre J, Blackburn H. Ventricular premature beats; reliability in various measurement methods at rest and during exercise. Cardiology 1978;63:53–63. [25] Moons K, van Es G, Deckers J, Habbema J, Grobbee D. Limitations of sensitivity, specificity, likelihood ratio, and Bayes’ theorem in assessing diagnostic probabilities: a clinical example. Epidemiology 1997;8:12–7. [26] Simel D, Samsa G, Matchar D. Likelihood ratios with confidence: sample size estimation for diagnostic test studies. J Clin Epidemiol 1991;44:763–70. [27] Fletcher R, Fletcher S, Wagner E. Clinical Epidemiology. Baltimore, MD: Williams and Wilkins, 1996. [28] Hagen M. Test characteristics: how good is that test? Med Decision Making 1995;22:213–33. [29] Hanley J, McNeil B. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983;148:839–43. [30] Hanley J, McNeil B. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:29–36. [31] Crow R, Prineas R, Dias V, Taylor H, Jacobs D, Blackburn H. Ventricular premature beats in a population sample. Circulation 1975;5152:211–5. [32] Elkon K, Swerdlow T, Myburgh D. Persistent ventricular ectopic beats. South African Med J 1977;52:564–6. [33] Fisher F, Tyroler H. Relationship between ventricular premature con-
[34]
[35] [36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49] [50]
[51]
497
tractions on routine electrocardiography and subsequent sudden death from coronary heart disease. Circulation 1973;47:712–9. Pratt C, Deklos G, Wierman A, Mahler S, Seals A, Leon C, Young J, Quinones M, Roberts R. The changing base line of complex ventricular arrhythmias. N Engl J Med 1985;313:1444–9. Winkle R. Antiarrhythmic drug effect mimicked by spontaneous variability of ventricular ectopy. Circulation 1978;57:1116–21. Sami M, Kraemer H, Harrison D, Houston N, Shimasaki C, DeBusk R. A new method for evaluating antiarrhythmic drug efficacy. Circulation 1980;62:1172–9. Raeder E, Hohnloser S, Graboys T, Podrid P, Lampert S, Lown B. Spontaneous variability and circadian distribution of ectopic activity in patients with malignant ventricular arrhythmia. J Am Coll Cardiol 1988;12:656–61. Morganroth J, Michelson E, Horowitz L, Josephson M, Pearlman A, Dunkman W. Limitations of routine long-term electrocardiographic monitoring to assess ventricular ectopic frequency. Circulation 1978; 58:408–14. Morganroth J. Evaluation of arrhythmias in clinical trials: relationship to the spontaneous variability of ventricular arrhythmias. In: Wenger N, Mock M, Ringqvist I, editors. Ambulatory Electrocardiographic Recording. Chicago, IL: Year Book Medical Publishers, Inc., 1980 pp. 285–98. Michelson E, Morganroth J. Spontaneous variability of complex ventricular arrhythmias detected by long-term electrocardiographic recording. Circulation 1980;61:690–5. Moller M. Reliability of serial 24 hour ambulatory electrocardiography in predicting cardiac death after myocardial infarction. Eur Heart J 1982;3:67–74. Bigger J Jr, Rolnitzky L, Leahey E Jr, LaPook J. Duration of recording: activity protocol. In: Wenger N, Mock M, Ringqvist I, editors. Ambulatory Electrocardiographic Recording. Chicago, IL: Year Book Medical Publishers Inc., 1980 pp. 87–102. Kennedy H, Chandra V, Sayther K, Caralis D. Effectiveness of increasing hours of continuous ambulatory electrocardiography in detecting maximal ventricular ectopy. Am J Cardiol 1978;42:925–30. Thanavaro S, Kleiger R, Hieb B, Krone R, DeMello V, Oliver G. Effect of electrocardiographic recording duration on ventricular dysrhythmia detection after myocardial infarction. Circulation 1980;62:262–5. Davis B, Friedman L, Lichstein E, for the Beta-Blocker Heart Attack Trial Study Group. Are 24 hours of ambulatory ECG monitoring necessary for a patient after infarction? Am Heart J 1988;115:83–91. Connolly S, Claims J, on behalf of the CAMIAT Pilot Study Group. Comparison of one-, six-, and 24-hour ambulatory electrocardiographic monitoring for ventricular arrhythmia as a predictor of mortality in survivors of acute myocardial infarction. Am J Cardiol 1992; 69:308–13. Petretta M, Bianchi V, Pulcino A, Carpinelli A, Valva G, Themistodakis S, Attisaro T, Salemme L, Bonaduce D. Continuous electrocardiographic monitoring for more than one hour does not improve the prognostic value of ventricular arrhythmias in survivors of first acute myocardial infarction. Am J Cardiol 1994;73:139–42. Bikkina M, Larson M, Levy D. Asymptomatic ventricular arrhythmias and mortality risk in subjects with left ventricular hypertrophy. J Am Coll Cardiol 1993;22:1111–6. Knottnerus J, Leffers P. The influence of referral patterns on the characteristics of diagnostic tests. J Clin Epidemiol 1992;45:1143–54. Diamond G. Affirmative actions: can the discriminant accuracy of a test be determined in the face of selection bias? Med Decision Making 1991;11:48–56. Knottnerus J. The effect of disease verification and referral on the relationship between symptoms and diseases. Med Decision Making 1987;7:139–48.