Abstract
Background: Despite using the same tool (ECG), the proportion of myocardial infarctions that goes unrecognized varies from 20% to 60% in population‐based studies. The reasons for such wide variations have not been studied. We sought to evaluate the effect of ECG‐MI criteria and study methodology on the prevalence of unrecognized myocardial infarction (UMI) and to identify the optimal ECG‐MI criteria for UMI detection in epidemiologic studies.
Methods: A random population‐based sample of 2042 adults, age ≥45 years, underwent history, medical record abstraction and ECG. Six different ECG‐MI criteria and two subjective recognized myocardial infarction (RMI) identification criteria, from different published studies, were applied to the same survey ECG. The operating test characteristics of different criteria were compared with the objective criterion standard of a RMI by Gillum criteria.
Results: The UMI proportion estimates varied from 32% to 61% due to variation in ECG‐MI criteria, while keeping the study population, MI recognition criteria, and ECG constant. Subjective criteria for MI recognition had limited value (positive predictive value of 44–93%) in picking up RMI. Depending on the ECG abnormality used to define MI, ECG reading had widely varying sensitivity (21–37%; P < 0.0001) with consistently high specificity (92–97%) for detection of RMI.
Conclusions: The prevalence estimates of UMI vary widely and are strongly dependent on the ECG‐MI and MI recognition criteria. Future studies of UMI should explicitly recognize this variation and select the ECG‐MI criteria that match their study aims.
Keywords: unrecognized myocardial infarction, ECG‐MI criteria, electrocardiogram, ischemic heart disease
Unrecognized myocardial infarctions (UMI) are myocardial infarctions that go undetected by the patient and the physician at the time of occurrence, only to be incidentally discovered by an abnormal electrocardiogram (ECG) performed some time later. From routine or surveillance electrocardiogram studies, UMIs have an estimated prevalence of 1.2–6.4% in the adult population. 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 Accurately assessing the presence of an UMI is clinically relevant since UMI are reported to have a prognosis similar to that of clinically recognized myocardial infarctions (RMI). 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 The five‐fold difference in UMI rates in the Caucasian population could be due to true differences in UMI prevalence in the cohorts or due to differences in study methodologies, especially differences in the ECG definition of MI criteria used. Currently, there are no studies assessing the basis for this marked variation in UMI prevalence.
We have recently suggested the variation in ECG‐MI criteria as the primary explanation for the variance in prevalence estimates of UMI. 12 For example, some studies on UMI have used only major Q wave criteria (Q wave > 0.04 seconds), others use major and minor Q wave criteria, and yet others use Q wave criteria enhanced by ST segment and T wave abnormalities. Although many newer imaging modalities (echocardiogram and nuclear studies) are in clinical use now, the electrocardiogram continues to be the most commonly available and lowest cost tool in clinical practice to screen for silent coronary artery disease. Therefore, identifying the source of variation and suggesting the best ECG definition for MI and therefore UMI is a clinically relevant and important task, with a direct impact on provision of secondary preventive measures for the individual patient.
Using data from a large ongoing population‐based study of cardiovascular disease (the Olmsted County Heart Study), we examined variations in prevalence estimates of UMI based on variations in ECG definitions for MI. Since the proportion of MIs that goes unrecognized is directly dependent on correctly identifying old recognized myocardial infarctions, we used medical record review to confirm previously documented RMI. The performance of ECG‐MI criteria could only be measured against RMI, as there is no gold standard for the diagnosis of UMI. Therefore, we assessed the ability of each set of ECG‐MI (n = 6) to identify the confirmed RMIs. Finally, we assessed the rates of UMI predicted by each set of ECG‐MI criteria. Final recommendations for selection of most appropriate criteria for clinical practice were based on both the ability to identify confirmed RMIs and rate of identified UMIs.
METHODS
The Olmsted Medical Center and Mayo Foundation Institutional Review Boards approved this study, and all the subjects gave written informed consent.
Study Setting
The Olmsted County Heart Study started enrollment in the Olmsted County, Minnesota on January 1, 1997 and ended its first survey on September 30, 2000. The Olmsted County had a population of approximately 110,000 in 1990, of which 96% residents were white. The unique ability to capture all health‐care related information in this county has been described previously. 13 , 14
Population Sampling, Subject Recruitment, and Enrollment
A random sample of Olmsted County residents, ≥45 years of age, as of January 1, 1997 was identified. A sampling fraction of 7% was applied within each of the gender and age (5 years) specific strata. Of the 4203 subjects invited, 2042 (49%) participated.
Charts of all enrolled participants were reviewed by four trained nurse chart abstractors with an average of 8 years of research experience. An earlier report of this cohort has documented long‐term medical record availability (median 36 years). 15 Each participant completed a Rose chest pain questionnaire, 16 which is a validated instrument for the diagnosis of angina in studies of cardiovascular epidemiology, has been used in UMI studies, 9 , 17 , 18 , 19 and defines angina as chest pain brought on either by walking uphill or hurrying, or by walking at an ordinary pace on the level and is relieved by rest within 10 minutes. In this questionnaire, the subject is also asked for a history of possible myocardial infarction by looking for prolonged (≥half an hour duration) chest pain across the chest, and then further characterization is carried out. All the subjects underwent a standardized physical examination. BMI was calculated by weight (kg)/height (m2). Spirometry was performed on each subject to measure forced expiratory volume in the first second (FEV1) and forced vital capacity (FVC). A standard 12‐lead ECG was performed on all the subjects using Marquette Mac 8 ECG machines and interpreted by the application of the Minnesota Code (MC) with MC‐MEANS (Minnesota Code‐Modular ECG Analysis System). 20
Application of Electrocardiographic Criteria to Identify Myocardial Infarction
Each subject's ECG was assessed using several different published ECG‐MI criteria. In order to compare ECG‐MI criteria systematically and eliminate concerns about inter‐rater reliability, only those criteria based on the Minnesota Code were used. Minnesota coding was available for all study electrocardiograms. This process excluded other ECG‐MI criteria based on manual reading, 3 , 4 , 7 , 8 as well as criteria based on the NOVACODE. 11 , 17
The MC 1‐2‐8 code for MI is based on poor R wave progression across the precordial leads. This code has the potential for giving false positive results in obese and COPD patients, therefore it has been dropped in some studies. 5 , 10 , 18 In order to improve the diagnostic performance of MC 1‐2‐8, we created BMI and FEV1 sensitive ECG‐MI criteria that are a modified version of Reykjavik I ECG‐MI criteria. According to these ECG‐MI criteria, any participant who has severe obesity (BMI > 40) or severe emphysema (FEV1 < 40% predicted) would not be given an ECG‐MI designation if his ECG‐MI designation is based on MC 1‐2‐8 only. All other patients may qualify for ECG‐MI based on the presence of this code only, or any of the MC 1‐1‐X or 1‐2‐X, which represent major and intermediate Q waves.
The six different ECG‐MI criteria used are listed in Table 1. They are presented in order from the most restrictive (Reykjavik II), using only the major MC Q codes for myocardial infarction to the most inclusive (BRHS Possible MI) as it uses all major and minor Q codes. We hypothesized that the specificity of ECG‐MI criteria would increase as we go down the list.
Table 1.
ECG‐MI Criteria used in Different Studies on Unrecognized Myocardial Infarctions
| Study | ECG Abnormality | Minnesota Codes |
|---|---|---|
| 1. Reykjavik Study II. 10 | Major Q waves | MC 1–1–1 to 1–2–5 and 1–2–7 |
| 2. British Regional Heart Study 18 | Major Q waves | Definite MI = MC 1–1–1 to 1–2–6 |
| 3. MRFIT 5 | Major Q waves | MC 1–1–1 to 1–2–7 |
| 4. Weight and BMI sensitive ECG‐MI criteria 12 | Major Q waves | MC 1–1–1 through 1–2–7. 1–2–8 to be used only if that subject's BMI is less than 40 and FEV1 is more than 40% predicted. |
| 5. Reykjavik Study I 9 | Major Q waves | MC 1–1–1 to 1–2–8 |
| 6. Cardiovascular Health Study 2 | Major Q waves as stand alone and minor Q waves if accompanied by ST and T abnormalities | (1–1–1 through 1–2–7. 1–2–8 and 1–3–X were included if the ECG also had ischemic ST segment abnormalities(4–1 to 4–3) or had ischemic T wave abnormalities (5–2 through 5–3). |
| 7. British Regional Heart Study 18 | Major and minor Q waves | Definite and Possible MI = All MC Q–codes i.e., 1–1–X, 1–2–X, and 1–3–X |
ECG criteria are arranged in an order of increasing number of ECG‐MI Minnesota Codes, from the top to the bottom.
Definitions of Recognized and Unrecognized Myocardial Infarction
We considered three methods of detecting a previously recognized myocardial infarction. The WHO and/or Gillum criteria for acute myocardial infarction were considered as the objective gold standard for establishing the diagnosis of an inpatient recognized myocardial infarction. 21 In addition, two subjective methods for detection of a previous RMI were also evaluated against the objective WHO criteria: (1) patient self‐identified history of heart attack and (2) the history of prolonged chest pain as described in the Rose questionnaire. These have been used in published studies. 2 , 18 , 19 , 22 Our primary interest was in their performance as compared to the gold standard (Table 2).
Table 2.
Criteria for Detection of Old Recognized Myocardial Infarctions
| MI Recognition Criteria | Type | Description |
|---|---|---|
| WHO or Gillum 21 | Objective | A designation of RMI was given to a subject if WHO or Gillum criteria (chest pain, elevated enzymes and evolving ECG changes) were fulfilled on nurse medical record abstraction which included all the inpatient and out patient charts. ECG reports were abstracted, as read by the cardiologist during the patient care. Creatine kinase or CK‐MB levels were accepted if abnormal in interest of MI recognition. The strict criterion of CK being more than two times normal was not applied. A subject was considered to have had an UMI if chart review did not reveal a myocardial infarction, but the ECG‐MI criteria were fulfilled on the survey ECG. |
| History of heart attack 2 | Subjective | Myocardial infarction recognition was based on self‐administered patient questionnaire, i.e., the patient was asked, “Has a medical care provider told you that you had a heart attack?” Yes means a history of RMI. |
| History of prolonged chest pain 18 | Subjective | Rose questionnaire for angina includes a question, which classifies the pain as pain of possible infarction. In the self‐administered questionnaire, the patient was asked; “Did you ever have a severe pain across the front of your chest lasting for half an hour or more?” Yes means a history of RMI. |
All other myocardial infarctions identified on surveillance ECG that could not be matched to a previously recognized myocardial infarction were considered UMI. The ability to vary ECG‐MI criteria as well as RMI detection criteria, while using the same ECG, allowed us to judge their impact on UMI prevalence.
The diagnostic performance of ECG‐MI criteria could only be measured for RMI due to the lack of a criterion standard for UMI diagnosis. Therefore, we assumed that ECG‐MI criteria perform equally well for any myocardial infarction, whether recognized or unrecognized.
Statistical Analysis
The diagnostic performance of two subjective criteria for RMI and the six ECG‐MI criteria was compared to the objective gold standard of WHO/Gillum criteria (confirmed by medical review) by calculating the sensitivity, specificity, positive predictive value, negative predictive value, accuracy, and index of merit. The differences in sensitivities and specificities were further evaluated by McNemar test for correlated proportions. UMI prevalence was calculated by applying the six different ECG‐MI criteria on the same survey ECG.
RESULTS
Study Participants
The mean (±SD) age of study participants was 62.8(±10.6) years, the mean BMI was 28.4(±5.4). About half (52%) were women, 8.9% were current smokers, 4.5% had diabetes, and 12.2% had a history of coronary artery disease. Medical records from 500 participants and 500 nonparticipants revealed no statistically significant differences in terms of age, sex, hypertension, coronary artery disease, previous myocardial infarction, diabetes, previous cardiovascular hospitalization, and CHF.
MI Recognition Criteria
Initial chart review revealed a total of 98 RMI. For subjects who had a self‐reported “history of heart attack” or “prolonged chest pain,” but initial chart review had not revealed an MI, a second chart review was carried out. From the second chart review, 7 subjects with a self‐reported history of MI were excluded as insufficient information was available in charts to apply WHO/Gillum criteria. Three other subjects with a self‐reported history of MI, were found to fulfill RMI criteria based on repeat medical record review. This brought the number of identified RMI to 101, yielding a prevalence rate of 4.4%. Subjective criteria for MI recognition showed limited sensitivity and positive predictive value, compared to medical record confirmation of the WHO/Gillum criteria for RMIs. The Rose questionnaire had lower sensitivity (44% vs 93%) and positive predictive value (28% vs 75%) than subject recall for a history of “heart attack” (Table 3). Specificity and negative predictive value were ≥94% for both the subjective criteria.
Table 3.
Comparison of Subjective MI Recognition Criteria (Based on Self‐administered Patient Questionnaire) Against the Gold Standard of Recognized Myocardial Infarction as per Gillum/WHO Criteriaa
| Sensitivity % | Specificity % | Positive Predictive Value% | Negative Predictive Value% | |
|---|---|---|---|---|
| History of heart attackb | 92.9 (91/98) | 97.6 (1871/1918) | 75.2 (91/121) | 99.7 (1871/1876) |
| History of prolonged chest painc | 44.4 (44/99) | 94.1 (1797/1909) | 28.2 (44/156) | 97.0 (1797/1852) |
aShown in % (Frequency/WHO‐MI frequency).
bMyocardial infarction recognition was based on self‐administered patient questionnaire, i.e., the patient was asked, Has a medical care provider told you that you had a heart attack.
cIn the self‐administered Rose questionnaire, the patient was asked; Did you ever have a severe pain across the front of your chest lasting for half an hour or more?
UMI Prevalence
The prevalence of UMI in the cohort varied significantly between 2.4% and 7.6% according to the choice of ECG criteria, while keeping the ECG and MI recognition method constant (Table 4). The number of recognized myocardial infarctions remained the same at 101 as it is determined by independent chart abstraction, applying the WHO/Gillum criteria. The lowest prevalence of UMI was seen with Reykjavik II criteria, which were the most restrictive; and the highest prevalence was seen if all Minnesota Q codes were used to identify unrecognized myocardial infarction. The UMI proportion of all MIs increased from 32% to 61% as the definition of MI became less restrictive. No problems with false positivity of MC 1‐2‐8 were seen as the BMI increased and the FEV1 fell. The “BMI & FEV1 sensitive” ECG‐MI criteria identified the same patients as Reykjavik I ECG criteria. However, the numbers of patients with MC 1‐2‐8 (n = 9), BMI > 40 (n = 68), or with FEV1 < 40% (n = 30) were small. Not a single participant with FEV1 < 40% or BMI > 40 had a UMI by MC 1‐2‐8.
Table 4.
The Effect of Selection of ECG‐MI Criteria on the Proportion of Myocardial Infarctions Identified as Unrecognized in a Population‐Based Studya
| All Myocardial Infarctions (All MI) | Unrecognized Myocardial Infarctions (UMI) | UMI/All MI Proportion (C.I.) | UMI Prevalence | |
|---|---|---|---|---|
| Reykjavik II | 150 | 49 | 0.32 (0.25, 0.40) | 2.4 |
| BRHS | 158 | 57 | 0.36 (0.28, 0.43) | 2.8 |
| MRFIT | 160 | 59 | 0.37 (0.29, 0.44) | 2.89 |
| Weight and BMI sensitive criteria | 168 | 67 | 0.40 (0.32, 0.47) | 3.29 |
| Reykjavik I | 168 | 67 | 0.40 (0.32, 0.47) | 3.29 |
| CHS | 181 | 80 | 0.44 (0.37, 0.51) | 3.93 |
| All Q‐Codes (BRHS) | 256 | 155 | 0.61 (0.55, 0.66) | 7.61 |
aThe number of RMI stays constant at 101 as RMI identification is independent of ECG‐MI criteria.
Selection of Optimal ECG‐MI Criteria
The sensitivity of ECG‐MI criteria, compared to the WHO/Gillum criteria for RMI, varied from 21% to 36%, with highest sensitivity with all MC Q‐codes followed by Cardiovascular Health Study (CHS) criteria. The specificity varied from 92% to 97.5%, while the negative predictive value was approximately 96% for all the ECG criteria (Table 5). There was no statistically significant difference in the sensitivity (20.8–22.8%) of the criteria based on the major Q waves only. This included the ECG‐MI criteria used in MRFIT, Reykjavik II, Reykjavik I, and BRHS studies. The ECG‐MI criteria used in the Cardiovascular Health Study had a higher sensitivity than all other major Q‐code criteria (30% vs 21–23%; P < 0.0083), with a significantly lower specificity (97% vs 96%; P < 0.0072). The sensitivity of BRHS Possible MI criteria was the highest (37%; P < 0.0083) with the lowest positive predictive value (19% vs 31% with Reykjavik II) and specificity (92% vs 96–98%; P < 0.0001) as compared to any other ECG‐MI criteria. Overall, the CHS criteria seemed to present the best possible compromise between sensitivity and specificity with second highest sensitivity (30%) and similar positive predictive value (27%) and specificity (96%) as compared to other criteria that use major Q codes only.
Table 5.
The Ability of Survey ECG‐MI Criteria in Identifying Old Recognized Myocardial Infarctions
| Sensitivity (%) | Positive Predictive Value (%) | Specificity (%) | Negative Predictive Value (%) | Accuracy (TP + TN/n × 100) | Index of Merit (Sensitivity + Specificity/2) | |
|---|---|---|---|---|---|---|
| Reykjavik II | 21.8 | 30.99 | 97.5 | 95.98 | 93.71 | 59.7 |
| British Regional Heart Study | 20.8 | 26.9 | 97.0 | 95.9 | 93.27 | 58.9 |
| MRFIT | 21.8 | 27.2 | 96.9 | 96.0 | 93.22 | 59.4 |
| Reykjavik I | 22.8 | 25.6 | 96.5 | 95.99 | 92.87 | 59.7 |
| Cardiovascular Health Study | 29.7 | 27.3 | 95.9 | 96.3 | 92.58 | 62.8 |
| All Q‐codes (BRHS) | 36.6 | 19.3 | 91.99 | 96.5 | 89.24 | 64.3 |
The majority of survey electrocardiograms of known RMI patients did not fulfill any ECG‐MI criteria. The number of the 101 RMI patients without survey ECG evidence for MI was 79 for Reykjavik II ECG‐MI criteria, 80 for BRHS, 79 for MRFIT, 78 for Reykjavik I, 71 for CHS and 64 for all Q‐codes. The mean time elapsed between MI and survey ECG was 242 months (SD = 217), but no specific pattern of ECG normalization over time could be identified (plot not shown).
DISCUSSION
UMI Prevalence
A variety of previously published studies, employing different ECG criteria for myocardial infarction, have reported that 20–60% of infarcts are unrecognized. By applying the same range of ECG criteria to our study population, we have estimated that 32–61% of myocardial infarctions are unrecognized. Our ability to reproduce the same range of estimates in a single cohort suggests that previously reported variations in ECG criteria for myocardial infarction rather than actual differences in UMI occurrence rates in the studies likely explain the range of estimates. The reported proportion of myocardial infarctions unrecognized is sensitive to ECG‐MI criteria and rises significantly in our study as more inclusive ECG criteria are used to define UMI (Table 4). This underscores the need for further research designed to identify the most accurate ECG‐MI criteria.
ECG‐MI Criteria
All ECG‐MI criteria were relatively insensitive for identifying old RMIs (Table 5). Sensitivity did significantly increase from 21% to 30% (P < 0.01) when minor Q codes with concomitant ST, T changes were added, as was done in the Cardiovascular Health Study. 2 However, even the highest rate is not sufficient for clinical use. The low sensitivity is explained by the tendency of ECG‐MI signs to regress with time and the presence of non–Q‐wave MIs. Over 70% of our patients, who had diagnostic ECG changes during their recognized myocardial infarction, had a nondiagnostic survey ECG. Similar finding of regression of ECG signs of MI was seen in 74% of 292 patients with history of recognized myocardial infarction in the British Regional Heart Study (personal communication with Dr. Fiona Lampe). 18 This is a much higher rate of regression than reported in earlier studies and may be due in part to higher rate of use of thrombolytics and emergent angioplasty in present practice. 23 It is possible that many of these RMI patients still had ECG‐MI abnormalities that were not abnormal enough to fulfill the ECG‐MI criteria used by us. Since many myocardial infarctions lose ECG expression over time, retrospective assessments of prior MI have built in limitations. Many UMI studies have used serial ECG analyses over several years to detect an UMI, something that is not possible in the current study due to the cross‐sectional study design. For UMI, the problem will persist irrespective of the study design, whether prospective or cross‐sectional, as the diagnosis can only be made retrospectively, for UMI by definition goes unrecognized at the time of occurrence. It has been observed in the context of RMI that most of the electrocardiograms that normalize after myocardial infarction, do so within 2 months of the MI. 24 Applying this concept to UMI would mean that in order to detect all incident UMI, frequent serial ECG analyses (every week or so) need to be performed, which is not feasible in a large population‐based study. This is probably the reason why we did not observe a relationship between regression of ECG‐MI signs and time since RMI in our study, as most of our patients' RMI occurred more than 2 years ago and none had an RMI within 2 months of the survey ECG.
As expected, arranging the ECG‐MI criteria in Table 1 in an ascending order of increasing number of ECG abnormalities considered diagnostic for MI, led to a pattern of decreasing specificity and accuracy in Table 5, from top to bottom. Although the differences in specificity in between different ECG‐MI criteria were statistically significant for multiple levels (P values available from authors), the clinical significance can best be gauged by the positive predictive value that drops only a little from 31% for Reykjavik II criteria to 27% for Cardiovascular Health Study criteria. Overall, the CHS criteria (MC 1‐1‐1 through 1‐2‐7; 1‐2‐8 and 1‐3‐X included if 4‐1 to 4‐3 or 5‐2 through 5‐3 were present) appeared to provide the best possible compromise between sensitivity and specificity for correlation with old RMI, and therefore, detection of UMI. If we assume that the ECG‐MI criteria are equally specific and sensitive in RMI and UMI, we must also assume that surveillance ECG identify less then one‐third of the UMI (sensitivity 27–31%). Therefore, using ECGs to identify UMI for tertiary CHD prevention opportunities will miss many people who could potentially benefit. The high specificity suggests that if a UMI is identified by ECG, prevention strategies are appropriate to institute. In etiologic research, where optimal specificity and positive predictive value is favored; the Reykjavik II criteria would out perform others as they produced the highest positive predictive value, specificity, and accuracy values of 31%, 97.5%, and 93.7%, respectively. In public‐health related research, where optimal sensitivity is desired to estimate the magnitude of the population burden of the disease, the application of all MC Q‐code criteria may be preferable due to the highest sensitivity (36.6%) and index of merit (64.3%). This calls attention to the importance of selecting the ECG‐MI criteria that match study aims in future studies of UMI.
BMI and FEV1 sensitive ECG‐MI criteria did not add to the Reykjavik I criteria although they were theorized to improve the diagnostic performance of MC1‐2‐8. These ECG‐MI criteria may still be helpful in other populations with higher prevalence of ECG‐MI based on MC 1‐2‐8.
The Impact of Recall Bias in Diagnosis of RMI
UMI estimated prevalence depends on methods of defining a previous RMI. Three major methods have been used in the literature, i.e., patient self‐reported history of heart attack, a positive Rose questionnaire and medical record abstraction. Table 3 highlights the problems with recall bias in self‐report. It also points out the lack of validity of the Rose questionnaire for identifying previous RMI. This is not surprising since the Rose questionnaire was developed for assessment of angina and intermittent claudication in research settings, for which it is still, considered the gold standard. Our observations highlight the importance of selecting tools appropriately validated for the question being studied.
RMI detection criteria, based on medical record abstraction, such as the WHO/Gillum criteria, appear to be the most objective and are best when available for research work. In clinical settings, when medical records are not obtainable, patient history of heart attack should be used in preference to the Rose questionnaire. Whether in research or clinical setting, use of "history of heart attack" has an additional value of picking up those cases of RMI that were missed due to missing medical records information. In the current study, we reevaluated the charts of subjects who said yes to a history of heart attack, but initial chart review had failed to reveal evidence of RMI. This led to identification of three additional cases of RMI who were not picked in the initial chart review or the information was in the correspondence section.
There are several epidemiological controversies regarding risk factors of UMI. Some epidemiological studies show diabetes 7 , 11 , 17 and hypertension 3 , 4 , 7 to be associated with UMI, but most others do not support this association. 2 , 6 , 8 , 9 The effect of variations in ECG criteria on risk factor associations of UMI has not been previously studied. Many of these controversies may be related to wide variations in ECG criteria for MI and may need to be re‐analyzed using a standard set of criteria. An adoption of uniformly accepted ECG‐MI criteria would provide an important step toward resolution of these epidemiologic controversies.
Strengths
The current study has the advantage of a probability‐based, stratified random sampling of the parent population. This is a significant issue in an epidemiological study that is attempting to determine the prevalence of a disease and represents a significant improvement from some of the previous epidemiological studies on UMI which either did not use probability‐based random sampling 7 , 25 or included only men. 3 , 4 , 5 , 6
This study is the first to employ a range of ECG‐MI criteria, in a population‐based setting, to evaluate the accuracy of ECG‐MI criteria and assess their impact on the prevalence of unrecognized myocardial infarctions. Previous studies evaluating the diagnostic performance of ECG‐MI criteria have used different criterion standards including hospital‐based autopsy case series, hospital‐based RMI case series, or predictive validity by the ability of ECG‐MI to predict future MI or coronary death. 26 , 27 , 28 , 29 , 30 , 31 Each of these methods is subject to inherent limitations.
Potential Limitations
These include cross‐sectional study design, participation bias, lack of racial diversity, limited number of RMI, lack of availability of troponin data to detect RMI, heightened health awareness in the Olmsted County (37% of population is employed in the health‐care profession as compared to 20% of US average) and misclassification by the criterion standard. The concern for potential participation bias is minimized by the similarity of participants and nonparticipants in terms of coronary risk factors. Heightened medical awareness may increase the probability of detecting an MI, thereby lowering the rate of UMIs. Seven of 32 subjects, who reported they had been told by a doctor that they had a heart attack, actually had a UMI, detected during routine medical care. This underscores the incremental value of querying patients who answer affirmatively to a history of previous heart attack, whether it was diagnosed during an episode of chest pain and hospitalization or on an ECG in the office.
CONCLUSIONS
The prevalence of UMI is strongly dependent on the ECG criteria used to define myocardial infarction, as well as on the criterion standard used to identify a previous RMI. The UMI prevalence estimates of 2.4–7.6% in adults over age 45 years indicate that UMI has a significant population expense in terms of morbidity and mortality. Efficacious tertiary prevention therapy is available for persons who had experienced myocardial infarctions; therefore, an accurate method of identifying this event will permit more appropriate allocation of resources. In epidemiological studies, standardization of the diagnostic criteria of unrecognized myocardial infarction should be agreed upon, just as was done for the recognized myocardial infarction two decades ago. 21
Acknowledgments
Acknowledgments: We are indebted to Ms. Carolyn Hain and Ms. Tammy Burns for their secretarial support.
Supported by grants from the Public Health Service (NIH HL 555902‐Rodeheffer and AHRQ 1RO1 HS10239‐Yawn), the Miami Heart Research Institute‐Rodeheffer, the OMC Foundation and the Mayo Foundation.
REFERENCES
- 1. Sheifer SE, Manolio TA, Gersh BJ. Unrecognized myocardial infarction. Ann Intern Med 2001;135: 801–811. [DOI] [PubMed] [Google Scholar]
- 2. Sheifer SE, Gersh BJ, Yanez D, et al Prevalence, predisposing factors, and prognosis of clinically unrecognized myocardial infarction in the elderly (The CHS Study). J Am Coll Cardiol 2000;35: 119–126.DOI: 10.1016/S0735-1097(99)00524-0 [DOI] [PubMed] [Google Scholar]
- 3. Roseman RH, Friedman M, Jenkins CD, et al Clinically unrecognized myocardial infarction in the Western Collaborative Group Study. Am J Cardiol 1967;19: 776–782.DOI: 10.1016/0002-9149(67)90498-5 [DOI] [PubMed] [Google Scholar]
- 4. Medalie JH, Goldbourt U. Unrecognized myocardial infarction: Five year incidence mortality and risk factors (Israeli Heart Study). Ann Intern Med 1976;84: 526–531. [DOI] [PubMed] [Google Scholar]
- 5. Grimm RH, Jr , Tillinghast S, Daniels K, et al Unrecognized myocardial infarction: Experience in the Multiple Risk Factor Intervention Trial (MRFIT). Circulation 1987;75: 116–118. [PubMed] [Google Scholar]
- 6. Yano K, Maclean CJ. The incidence and prognosis of unrecognized myocardial infarction in the Honolulu, Hawaii, Heart Program (Honolulu Heart Study). Arch Intern Med 1989;149: 1528–1532.DOI: 10.1001/archinte.149.7.1528 [DOI] [PubMed] [Google Scholar]
- 7. Kannel WB, Abbot RD. Incidence, precursors and prognosis of unrecognized myocardial infarction (Framingham Study). Adv Cardiol 1990;37: 202–214. [DOI] [PubMed] [Google Scholar]
- 8. Nadelmann J, Frishman WH, Ooi WL, et al Prevalence, incidence and prognosis of recognized and unrecognized myocardial infarction in persons aged 75 years or older: The Bronx Aging Study. Am J Cardiol 1990;66: 533–537.DOI: 10.1016/0002-9149(90)90477-I [DOI] [PubMed] [Google Scholar]
- 9. Sigurdsson E, Thorgeirsson G, Sigvaldson H, et al Unrecognized myocardial infarction: Epidemiology, clinical characteristics and the prognostic role of angina pectoris. The Reykjavik Study. Ann Intern Med 1995;122: 96–102. [DOI] [PubMed] [Google Scholar]
- 10. Jonsdottir LS, Sigfusson N, Sigvaldson H, et al Incidence and prevalence of recognized and unrecognized myocardial infarction in women. The Reykjavik Study. Eur Heart J 1998;19: 1011–1018.DOI: 10.1053/euhj.1998.0980 [DOI] [PubMed] [Google Scholar]
- 11. Shlipak MG, Elmouchi DA, Harrington DM, et al The incidence of unrecognized myocardial infarction in women with coronary heart disease (HERS Research Group). Ann Intern Med 2001;134: 1043–1047. [DOI] [PubMed] [Google Scholar]
- 12. Ammar KA, Kors JA, Yawn BP, et al Defining unrecognized myocardial infarction: A call for standardized electrocardiographic diagnostic criteria. Am Heart J 2004;148: 277–284.DOI: 10.1016/j.ahj.2004.03.019 [DOI] [PubMed] [Google Scholar]
- 13. Kurland LT, Molgaard CA. The patient record in epidemiology. Sci Am 1981;245: 54–63. [DOI] [PubMed] [Google Scholar]
- 14. Melton LJ, III . History of the Rochester Epidemiology Project. Mayo Clin Proc 1996;71: 266–274. [DOI] [PubMed] [Google Scholar]
- 15. Okura Y, Urban L, Mahoney DO, et al.; The agreement between questionnaire data and medical records in the community: Influence of respondent characteristics. J Clin Epidemiol 2004;57: 1096–1103.DOI: 10.1016/j.jclinepi.2004.04.005 [DOI] [PubMed] [Google Scholar]
- 16. Rose G, Mccartney P, Reid DD. Self administration of a questionnaire on chest pain and intermittent claudication. Br J Prev Soc Med 1977;31: 42–48. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Boland LL, Folsom AR, Sorlie PD, et al Occurrence of unrecognized myocardial infarction in subjects aged 45 to 65 years (The ARIC Study). Am J Cardiol 2002;90: 927–931. [DOI] [PubMed] [Google Scholar]
- 18. Lampe FC, Whincup PH, Wanamethee SG, et al The natural history of prevalent ischaemic heart disease in middle‐aged men. Eur Heart J 2000;21: 1052–1062.DOI: 10.1053/euhj.1999.1866 [DOI] [PubMed] [Google Scholar]
- 19. Chadha SL, Radhakrishnan S, Ramachandran K, et al Epidemiological study of coronary heart disease in urban population of Delhi. Ind J Med Res 1990;(B)92: 424–430. [PubMed] [Google Scholar]
- 20. Kors JA, Van Herpen G, Wu J, et al Validation of a new computer program for Minnesota coding. J Electrocardiol 1996;29: 83–88. [DOI] [PubMed] [Google Scholar]
- 21. Gillum RF, Fortmann SP, Prineas RJ, et al International diagnostic criteria for acute myocardial infarction and acute stroke. Am Heart J. 1984;108: 150–158.DOI: 10.1016/0002-8703(84)90558-1 [DOI] [PubMed] [Google Scholar]
- 22. Gopinath N, Kaul U, Chadha SL, et al Asymptomatic coronary heart disease detected on epidemiological survey of urban population of Delhi. Indian Heart J 1992;44(2):95–98. [PubMed] [Google Scholar]
- 23. Fisch C. Electrocardiography In Braunwald E. (ed.):Heart Disease. A Textbook of Cardiovascular Medicine, 5th edition Philadelphia , PA , W. B. Saunders Company, 1997, p. 132. [Google Scholar]
- 24. Pyorala K, Kentala E. Disappearance of Minnesota Code Q‐QS patterns in the first year after myocardial infarction. Ann Clin Res 1974;6: 137–141. [PubMed] [Google Scholar]
- 25. Hulley SB, Newman TB, Summings SR. Choosing the study subjects: Specification, sampling and recruitment In Hulley SB, Cummings SR, Browner WS, Grdy D, Hearst N, Newman TB. (eds.): Designing Clinical Research, 2nd edition Philadelphia , PA , Lippincott Williams and Wilkins, 2001, p. 31. [Google Scholar]
- 26. Parker AB, Waller BF, Gering LE. Usefulness of the 12‐lead electrocardiogram in detection of myocardial infarction: Electrocardiographic‐anatomic correlations. Clin Cardiol 1996;19: 55–61. [DOI] [PubMed] [Google Scholar]
- 27. Horan LG, Flowers NC, Johnson JC. Significance of the diagnostic Q wave of myocardial infarction. Circulation 1971;43: 428–434. [DOI] [PubMed] [Google Scholar]
- 28. Sullivan W, Vlodaver Z, Tuna N, et al Correlation of electrocardiographic and pathologic findings in healed myocardial infarction. Am J Cardiol 1978;42: 724–732.DOI: 10.1016/0002-9149(78)90090-5 [DOI] [PubMed] [Google Scholar]
- 29. Ashley EA, Raxwal VK, Froelicher VF. The prevalence of prognostic significance of electrocardiographic abnormalities. Curr Prob Cardiol 2000;25(1):1–72. [DOI] [PubMed] [Google Scholar]
- 30. Rautaharju PM. Electrocardiography in epidemiology and clinical trials In Macfarlane P, Veitch‐Lawrie TD. (eds.): Comprehensive Electrocardiology. Theory and Practice in Health and Disease, 1st edition New York , NY , Pergamon Press, 1989, pp. 1219–1266. [Google Scholar]
- 31. Pipberger HV, Simonson E, Lopez EA, et al The electrocardiogram in epidemiological investigations. A new classification system. Circulation 1982;65(7):1456–1464. [DOI] [PubMed] [Google Scholar]