A major finding is that there is a statistically significant association between resting ECG abnormalities (ECGA) at baseline as a group and the cumulative incidence of myocardial infarction over 25 years also after adjustment for traditional risk factors. The ORs quantifying the comparisons are all high, between 2.8 and 4.2, regardless of the regression model used.
The benefit of ECGA compared to traditional risk factors following adjustment indicates that ECGA may not follow the same arteriosclerotic pathway as the Framingham Risk Index.
However, the Pseudo-R2, which explains the proportion of the variation, was for ECGA only 2.1% compared to FRI with 8.6%. The model combining ECGA and FRI has a Pseudo-R2 at 11%, still low.
There is no statistically significant association between ECGA and the outcome of death.
In 2018, based on a review of the literature, the US Preventive Services Task Force (USPSTF) published a recommendation not to use ECG for screening in previously healthy men without any other cardiovascular risk factor [22]. Our study and others though indicate that resting ECGs add value for the predictive model of the risk of future myocardial infarction [6, 23,24,25,26,27,28,29,30,31,32].
It is important to realize the difference between using ECG for screening of an assumed healthy population and using it as a tool among others in clinical practice. Using the present data as an example, a population screening would have a sensitivity of 20%, i.e., only one in five would be classified as a potential MI patient. Further, there will be about 75% false positive, unnecessarily requiring some attention. Population screening aims at finding potential cases early and an important requirement also is that there is some action that can be taken, which hardly is the case in the present study.
In clinical work, the assumptions are different with higher sensitivity and less false positive due the higher prevalence of persons with high risk.
It can be clinically relevant to examine healthy 45–50-year-old men with ECG as one component among others when judging the risk for an MI.
Since no remedy exists for unspecific ECG abnormalities, such findings may only be used to emphasize the important prevention of a healthy lifestyle.
Due to the few cases in the subgroups of the ECG abnormalities, it is not meaningful to evaluate statistically the impact of most Minnesota Code subgroups on the specific risk of MI.
Other studies suggest that prolonged QRS complexes (24–27) and negative T-waves (24, 28–29) lead to increased risks. This is not contradicted by our study.
In addition, we also find that STJ depression shows an increased but not statistically significant risk for MI and death.
Prolonged QTc (pulse-corrected QT-time) has been associated with higher risk of MI and death [32,33,34,35]. Prolonged QTc is not classified in the Minnesota Code and has therefore not been studied here.
An unexpected finding was that the category 9 participants (only one EGG taken and without abnormality) had a double risk of death, 28% vs 14%, (p < 0.05). We have no satisfactory explanation for this discovery. Some participants may have been skeptical or ill and refrained from the second follow-up. One thought was that this phenomenon might have been caused by death between the two examinations (in 1993 and 1998) but as seen in Fig. 2, this is not the explanation.
Another finding shown in Tables 5 and 6 is that there is more information in the variables included in FRI if they are studied separately than in the combined risk index (FRI). The Pseudo-R2 for MI goes from 11 to 15% and for death from 3.8 to 8.3%.
Limitations and strengths
Strengths
The original 1000 men were randomly selected from a well-defined population of male workers followed over a very long time (25 years). Very few rejected to participate. The nurses and health professionals were specially trained and well-defined methods were used. The procedures for baseline information acquisition were carefully developed based on current standards. Accurate information about the endpoints was available through national registers with quality control to contain valid data. The internal validity of the study and its results have throughout been carefully secured.
Limitations
The external validity, i.e., validity for other populations with other contextual structures, is clearly limited. The study is confined to males in a certain age range in a particular industry and certain country. The comparably small sample size and, therefore, the small number of outcomes events are limited and are a problem to some extent.
The study is also limited by the use of the Framingham Risk Index in the original Coeur study since this index has been superseded by better predictive tools such as QRISK2 [36] and the use of biomarkers [37]. However, in a 25-year longitudinal study, the analysis is necessarily restricted to the original risk measure. Important to point out is that even QRISk2 at its core include the traditional risk factors from the Framingham study. Another limitation is that the Minnesota Code used for classification does not register QTc.
Confounding
The main research question concerns the statistical association between the risk and results of the ECG investigations. This relation can be confounded by several variables. Many of these were observed at baseline such as age, smoking status, alcohol consumption, weight, height, and systemic blood measurements. Most of these are correlated with each other and with the risk of MI and death. Therefore, the crude correlation between risk and ECGA is confounded by a manifold variable. Among these, some are known and possible to observe. Others are known but not possible to observe and yet others have not even been imagined. The first mentioned group can be adjusted for using different regression models. For this paper, we have tried several models. An important finding is that the OR for ECG abnormality remains reasonably stable, OR about 3.4, regardless of the model. This is an indication that the main results are not seriously confounded by the variables used in this study.
Bias
A bias is a systematic error that has similar size and direction for a group of measurements. Several biases could influence the results. For example, laboratory results might be biased due to incorrect calibrations; questions for self-reporting can be formulated to give biased information. We have tried to address some of these biases by careful quality control and the use of certified laboratories in our study.
In longitudinal studies, it is possible for both confounding and biases to occur since it is difficult to account for changes in an individual’s habits or health status over the whole, in this case, 25-year period.
Confounding factors in this study would include not accounting for the eventuality that participants’ behaviors might change over time, for example smoking less, losing or gaining weight as well as medical treatment of cholesterol and hypertension, making the baseline risk factors obsolete in some cases.
Although we do not have secular, longitudinal data of the 25-year period, the findings presented in another longitudinal Swedish study of 50-year-old men followed for 50 years which show secular changes: smoking less, lower cholesterol but higher BMI, and a more sedentary lifestyle in cardiovascular risk factors over time are likely to apply to our cohort [38]. Furthermore, the addition of medication during this period may have changed the prognosis of some endpoints.