Background Differentiating physiological left ventricular hypertrophy (LVH) in athletes from pathological hypertrophic cardiomyopathy (HCM) can be challenging. This study assesses the ability of cardiac MRI (CMR) to distinguish between physiological LVH (so-called athlete's heart) and HCM.
Methods 45 patients with HCM (71% men and 20% athletic) and 734 healthy control participants (60% men and 75% athletic) underwent CMR. Quantitative ventricular parameters were used for multivariate logistic regression with age, gender, sport status and left ventricular (LV) end-diastolic volume (EDV) to ED ventricular wall mass (EDM) ratio as covariates. A second model added the LV EDV : right ventricular (RV) EDV ratio. The performance of the model was subsequently tested.
Results LV EDM was greater in patients with HCM (74 g/m2) compared with healthy athletes/non-athletes (53/41 g/m2), while LV EDV was largest in athletes (114 ml/m2) as compared with non-athletes (94 ml/m2) and patients with HCM (88 ml/m2). The LV EDV : EDM ratio was significantly lower in patients with HCM compared with healthy controls and athletes (1.30/2.39/2.25, p<0.05). The LV EDV : RV EDV ratio was significantly greater in patients with HCM (1.10) than in healthy participants (non-athletes/athletes 0.94/0.93). The regression model resulted in high sensitivity and specificity levels in all and borderline-LVH participants (as defined by septal wall thickness). Corresponding areas under the receiver operator characteristic (ROC) curves were 0.995 (all participants) and 0.992 (borderline-LVH participants only). Adding the LV EDV : RV EDV ratio yielded no additional improvement.
Conclusions A model incorporating the LV EDV : EDM ratio can help distinguish HCM from physiological hypertrophy in athletes. This also applies to cases with borderline LVH, which present the greatest diagnostic challenge in clinical practice.
- Cardiology prevention
- Exercise physiology
Statistics from Altmetric.com
Cardiac MRI (CMR) is increasingly used in athletes to rule out cardiac disease, often following inconclusive preparticipation screening. Left ventricular hypertrophy (LVH) is frequently observed in athletes, but is also a characteristic finding in hypertrophic cardiomyopathy (HCM), the leading cause of sudden cardiac death (SCD) in young athletes.1
It is therefore essential to be able to determine the cause of LVH, not only in order to prevent SCD but also to avoid unjustified exclusion from sports. Diagnostic uncertainty is greatest when borderline LVH is observed, with septal wall thickness (SWT) values between 12 and 16 mm in men, or between 11 and 16 mm in women.2 ,3 Several factors that can help to differentiate HCM from physiological cardiac adaptation have been described, including information on demographics (gender and ethnicity), family history, gene mutations and changes in ECG .4–7 Imaging features suggestive of HCM that have been reported include a normal or decreased left ventricular (LV) diameter and an enlarged left atrium.6
While the majority of existing data are derived from echocardiographic studies, CMR combines high spatial and temporal resolution with excellent visualisation of the entire heart, enabling accurate function and volume analysis of both ventricles.4 The cardiac apex, the lateral LV wall, and hypertrophy distribution may be poorly visible on echocardiography, which can result in underestimation of LV wall thickness.8
In this study, we use CMR to compare quantitative imaging parameters in healthy athletes with a group of patients with HCM, including some athletes with pathological hypertrophy. Ratio measures of ventricular volume to wall mass are assessed in particular, since they represent the relatively balanced nature of cardiac adaptation, and may therefore be of possible use in distinguishing pathological cardiac changes from physiological cardiac changes.
CMR was performed according to a standardised protocol on 45 patients with HCM (71% men), 554 healthy athletes (63% men) and 180 healthy non-athletic controls (49% men). The healthy athletic and non-athletic participants were recruited in a non-clinical setting, whereas the patients with HCM were all recruited from clinical practice. The healthy participants were recruited through advertisements (in the hospital and in sports clubs), flyers and word of mouth in the general public. Patients diagnosed with HCM that had undergone CMR between November 2004 and April 2011 were included. HCM diagnosis was the cardiologist's clinical diagnosis based on multimodality imaging, ECG and exercise testing, family and patient's history and, if possible, genetic testing, according to current guidelines of the European Society of Cardiology.9 All participants gave written informed consent and the study was approved by the Medical Ethics Committee of the University Medical Center Utrecht, the Netherlands.
Athletes exercised for a minimum of 6 h per week and non-athletic controls for a maximum of 3 h/week. Exclusion criteria for enrolment as a healthy participant included any personal or family history of cardiovascular, pulmonary or metabolic disease, as evidenced by a detailed questionnaire, ECG and CMR. Athletes performed a variety of sports. The athletic cohort consisted of 56 strength-type athletes (low dynamic-high static sports: weightlifting, powerlifting and judo; mean 6.5 h weekly training for an average of 11.4 years), 184 high dynamic-low static type athletes (high dynamic-low static sports: football, hockey, tennis and long distance runners; mean 13.7 h weekly training for an average of 7.3 years) and 314 combined strength-endurance-type athletes (high dynamic-high static sports: cycling, rowing and triathlon; mean 15 h weekly training for an average of 8 years).
The clinical diagnosis of HCM was established using all available data regarding clinical and family history, genetic testing, ECG, echocardiography, exercise testing, Holter monitoring and CMR. HCM represents unexplained LV hypertrophy with non-dilated ventricular chambers and without other cardiac or systemic diseases capable of producing the observed magnitude of hypertrophy, such as aortic stenosis or hypertension. Hypertension (systolic blood pressure >140 mm Hg or systolic blood pressure >90 mm Hg) was an exclusion criterion for all groups.10
LVH was defined on echocardiography and CMR as an interventricular SWT exceeding 11 mm in women or 12 mm in men.2 ,3 In order to provide insight into the subgroup that presents the greatest diagnostic uncertainty in clinical practice, a borderline LVH group was also defined as having an SWT below 16 mm, and over 11 mm in women or 12 mm in men.2 ,3 ,11 ,12 Individuals exhibiting borderline LVH were assessed further for differentiators between HCM and physiological adaptation to exercise. This included a comparison of LV diameter with established reference values and assessment of abnormal ECG parameters.6 ECGs were assessed by experienced cardiologists, using the Sokolow-Lyon criteria for the presence of LVH in all HCM participants, in all healthy athletes with borderline LVH and in a large subset of healthy athletes and non-athletes who were also evaluated in an earlier study.13 ,14
All CMR images were obtained on a 1.5 T MRI scanner (Achieva, Philips, Best, the Netherlands). Imaging parameters have been described in detail previously.15 In short, the imaging protocol included steady-state free precession (SSFP) cine images in a four-chamber, two-chamber, LV outflow tract (three-chamber), with full coverage of both ventricles in the short axis and quantitative flow measurement (Q flow) over all four cardiac valves.
Endocardial and epicardial contours of both ventricles were traced by blinded observers using semiautomated software (View Forum cardiac package V.R5.1V1L2.SP3, Philips, Best, the Netherlands) and an established, reliable and reproducible contour-tracing protocol.16 A blinded observer with experience in CMR checked the tracings before finalising the results.
End-diastolic (ED) and end-systolic (ES) endocardial contours were used to calculate the ED volume (EDV) and ES volume (ESV), derived ejection fraction and stroke volume of both the ventricles. ED ventricular wall mass (EDM) of both the ventricles was calculated using the epicardial and endocardial contours in end-diastole. The LV and right ventricular (RV) outflow tract, papillary muscles and trabeculae were considered part of the endocardial blood volume.17 ED ventricular diameter and SWT were measured on the short-axis SSFP cine images.
All analyses were carried out using the R statistical environment V.2.15.218 using the ‘rms’ (regression modelling strategies) and ‘logistf’ (Firth's bias reduced logistic regression) packages.19 Continuous data are presented as the mean±SD unless otherwise stated. Differences between groups were assessed by analysis of variance tests and a two-sided p value below 0.05 was considered statistically significant. CMR data were indexed to body surface area (BSA), with BSA calculated using the DuBois and DuBois formula: BSA (m2)=0.20247×height (m)0.725×weight (kg)0.425.20
In order to ensure convergence and to reduce bias in this dataset, we employed Firth's modified score procedure to perform multivariate logistic regression analysis, using HCM diagnosis as the outcome and the ratio of LV EDV to LV EDM as a discriminator, along with age, gender and other variables that demonstrated significant differences between groups. Sports status was defined as a dichotomous parameter with all athletes defined as 1 and non-athletic participants as 0. The contribution of each variable was assessed using a p value threshold of 0.15, as recommended for logistic regression models.18 Backward stepwise selection determined whether variables remained part of the model or were omitted. To correct for overoptimism and optimise internal calibration, the model was bootstrapped (1000 resamplings) and both the coefficients and intercept were adjusted accordingly.
The performance of the model was checked by calculating the area under the corresponding receiver operator characteristic (ROC) curve (AUC) in the complete study population. To relate these results to the most challenging group in clinical practice, the model was subsequently applied to the borderline-LVH subgroup. The absolute numbers of false negatives and false positives were determined at different levels of sensitivity. The same analysis was performed with the addition of the ratio of LV EDV to RV EDV, to assess if this could further improve the diagnostic performance.
Baseline characteristics are presented in table 1, showing variations in gender, age, diastolic blood pressure and resting heart rate. Quantitative ventricular parameters and their correlation with HCM diagnosis are listed in tables 2 and 3, respectively. Although healthy athletes had a significantly greater LV wall mass than healthy non-athletes, LV EDM was significantly greater in patients with HCM (p<0.001) than in athletes. In contrast, ventricular volumes (EDV and ESV) in healthy athletes were significantly larger than in patients with HCM. The ratio of LV EDV to LV EDM was significantly lower in patients with HCM compared with both healthy non-athletic controls and athletes.
Multivariate logistic regression analysis yielded the following formula for predicting the risk of HCM diagnosis:
Probability of HCM =1/(1+e−z), where z=10.36 – 7.84×LV EDV : EDM ratio+0.081×age in years − 3.44×sport status (yes=1)+1.48×gender (female=1).
All coefficients in this model were statistically significant (p<0.04). By plotting all possible cut-off values for the probability in an ROC curve, AUC was computed at 0.995, indicating that the model can very accurately discriminate between HCM and non-HCM participants (figure 1). This confirms the model's robustness in the complete population including patients with a most clear-cut HCM with wall thickness measurements exceeding 16 mm. More importantly, when applied in the subgroup of 47 participants with borderline LVH, the model still proved to discriminate very well (AUC 0.992). The model performs better than the LV EDV : EDM ratio alone in the overall study population (figure 1, AUC 0.995 vs AUC 0.966), but especially shows improvement in the borderline-LVH group (AUC 0.992 vs AUC 0.690). Incorporating the ratio of LV EDV : RV EDV in the model showed only slight improvement in all participants (AUC 0.997 vs AUC 0.995) and no improvement in borderline-LVH participants (AUC remains 0.992).
The probability of being diagnosed with HCM can be calculated by entering an individual patient's details into the model. A positive variable coefficient corresponds to an increased probability of HCM diagnosis (ie, with increasing age or female gender), whereas negative coefficients represent a decreased probability (ie, with an increasing LV EDV : EDM ratio and being an athlete). Depending on the selected cut-off value for the outcome of this multivariate model, HCM can be ruled out with different degrees of certainty (table 4).
The absolute numbers of incorrectly classified participants are given for different levels of sensitivity and specificity in table 4. Each row in this table corresponds to a cut-off for the model outcome (which is a probability between 0 and 1), representing a trade-off between high sensitivity and high specificity and not an absolute risk of having HCM. A low model cut-off will not only guarantee the identification of all HCM cases, but will also result in more false positives (lower specificity). Conversely, a higher model cut-off prevents false positives (higher specificity), but will cause HCM cases to be missed (lower sensitivity).
ECG characteristics suggestive of HCM are listed in table 5. There was a greater prevalence of LVH, T wave inversions (TWIs), left axis deviation, left atrial enlargement and ST-segment depression in patients with HCM compared with healthy participants. TWI was most frequently observed in the inferior leads in patients with HCM, and leads V1–V4 in healthy athletes. Conversely, patients with HCM frequently exhibited widespread TWIs involving multiple territories, with the inferior leads being involved most often.
Table 6 serves as an illustration of parameters that may be different in HCM and healthy athletes with borderline LVH. No healthy non-athletes were included in the borderline-LVH group as none exhibited an SWT greater than 11 or 12 mm (women and men, respectively). Of the 39 athletes included in the borderline-LVH group, 36 were combined strength-endurance type athletes (cycling/rowing/triathlon), while only 1 athlete performed an endurance type sport (long distance running) and 2 athletes were strength athletes. TWIs and voltage criteria for LVH appeared to be more frequent in patients with HCM with borderline LVH as compared with the healthy participants with borderline LVH (TWI, 63% vs 15%; voltage LVH, 38% vs 15%). However, these ECG characteristics did not reduce the number of false positives and false negatives as compared with the performance of the model presented.
This CMR study demonstrates that the LV EDV : LV EDM ratio, in conjunction with several simple demographic and clinical parameters (age, gender and sport status), accurately differentiates physiological LV remodelling in athletes from pathological changes observed in HCM. Most importantly, this phenomenon is also observed in the clinically most challenging group of participants with borderline LVH. The discriminatory ability of the LV EDV : EDM ratio illustrates that physiological cardiac changes in athletes are relatively balanced, as has been previously reported.15 ,21–23 Ventricular volume and wall mass increase proportionately in healthy athletes, with similar LV EDV : EDM ratios and LV EDV : RV EDV ratios observed in athletes and non-athletic controls. Conversely, patients with HCM typically develop a selective increase in ventricular wall mass and decrease in LV volume resulting in a lower LV EDV : EDM ratio and, to a lesser extent, a lower LV EDV : RV EDV ratio.
The LV EDV : EDM ratio, as measured by three-dimensional echocardiography (termed the ‘LV remodelling index’ by De Castro et al24), has previously been shown to differentiate between HCM and athlete's heart with a high degree of accuracy. Moreover, Petersen et al25 reported results similar to our own using CMR, albeit in a smaller population of athletes (n=25) and patients with HCM (n=35).
There is little doubt about pathological hypertrophy if the LV wall thickness exceeds the upper limits (>16 mm) in athletes, and a diagnostic model should do nothing less than perform well under these circumstances. However, not all patients with HCM exhibit such a substantial LV wall thickness or LV EDM increase, leading to diagnostic uncertainty in phenotypically mild cases. In a large study by Maron et al,8 10% of patients with HCM demonstrated LV hypertrophy in only 1 or 2 segments, while in a CMR study of 264 patients with HCM by Olivotto et al,26 the LV wall mass was within the 95th percentile of normal in 20% of patients with HCM. Notably, the model developed in the present study performed well in the subgroup of borderline-LVH participants, with a large AUC (0.992) and corresponding high levels of sensitivity and specificity. It should be noted that despite a strong correlation between SWT and HCM diagnosis in the overall population (table 3), the difference in SWT between HCM and non-HCM participants in borderline-LVH participants is comparable in size to the difference in the LV EDV : EDM ratios (table 6). In fact, the latter difference may even be slightly greater in this subgroup. In addition, an altered ratio of LV EDV : LV EDM seems a better reflection of imbalanced cardiac changes as seen in pathological hypertrophy, as demonstrated by the present study (table 6) and in previous works by our group and other groups.25–27 As discrimination is particularly important in borderline-LVH participants, the ratio of LV EDV : LV EDM seemed preferable to use in the regression model.
When applying a cut-off model outcome corresponding to 100% sensitivity, some healthy participants are incorrectly classified as HCM cases. In the overall population, 32 participants were misclassified using our model, but no particular pattern seemed to explain the misclassification of these participants. These participants tended to be older and to have a lower LV volume. In borderline-LVH cases, there was only one misclassified healthy participant when using a 100% sensitivity cut-off for the model outcome. This participant differed from other healthy athletes in having the lowest LV EDV/EDM ratio by quite a margin, based on an above average LV EDV and a particularly large LV EDM (the third largest of all healthy participants). This participant was a male professional cyclist exercising for 36 h a week for over 12 years. His SWT was 12.1 mm and his ECG did not show any abnormalities besides sinus bradycardia, which is considered physiological for well-trained athletes. This particular case was reviewed in detail by a sports cardiologist and not considered to have HCM. As we have demonstrated in a related study, the unreported use of anabolic androgenic steroids may explain a disproportionate LV wall mass increase.28
It is inconvenient for everyday clinical practice to include variables that require additional knowledge beyond the left ventricle if it does not significantly improve the model. Although left atrial dilation is associated with HCM, it is also a feature of athlete's heart, as previously stated by Maron and Pelliccia.6 This may explain why atrial volumes or diameters were not suitable as a differentiating parameter and were subsequently not incorporated in our model. While table 6 serves as an illustration of other parameters that may differentiate between HCM and healthy athletes with borderline LVH, the numbers in these groups are too small to perform statistical analyses.
In accordance with other studies, ECG features suggestive of HCM were more common in patients with HCM, suggesting that TWIs in the lateral and inferior leads are more likely to represent pathology.7 ,29 Although the observed ECG differences could be useful for differentiation between physiological and pathological LVH, we decided not to include ECG parameters in the model. In contrast to the other variables included in the model, ECG data are not always available and could complicate the model's application in clinical practice.
Other CMR parameters of possible importance in identifying HCM in athletes were not investigated in this study. These include late gadolinium enhancement (LGE), strain on tagged MRI30 and the morphology of papillary muscles and mitral valve.4 In our study, the use of a gadolinium-contrast medium in healthy athletes was considered unethical. It was therefore not part of the CMR protocol and we could not assess its use as a distinguishing diagnostic tool. LGE is considered to be a sign of pathological hypertrophy, and is often observed in patients with HCM,31 ,32 most commonly in segments with the greatest magnitude of ventricular wall thickening and at the septal insertion points of the right ventricle.33 ,34 Although LGE has been suggested to be an extra discerning factor to differentiate pathological hypertrophy from physiological hypertrophy in athletes,35 it is not seen in every patient with HCM. In addition, La Gerche et al36 have also described LGE in healthy endurance athletes with many years of training, and therefore the presence of LGE does not always indicate pathological hypertrophy. As LGE imaging was not performed in our healthy participants, we cannot be sure if LGE would have improved the differentiating ability of our model.
Our study included only seven athletes with HCM, of which five had an SWT >16 mm. The other two were male athletes with an SWT between 12 and 16 mm, reflecting a wider paucity of CMR data in athletes with clinical manifestations of cardiac pathology. Earlier studies of ratio measures also lack data regarding athletes with HCM.24 ,25
A further limitation relates to the fact that our study population consisted almost exclusively of Caucasians. The findings may therefore not be readily extrapolated to individuals of black ethnicity who are known to develop a greater degree of physiological LVH.2 ,37 This requires further investigation, and possibly correction of the model for ethnicity. In addition, the recruitment of participants from rather specific populations may have induced selection bias to some extent in both healthy participants and patients with HCM. The patient population was selected on the basis of those who had already undergone CMR as an inclusion criterion for this study. Recruiting some of the non-athletic healthy population in an academic hospital may have led to the over-representation of highly educated individuals, who may have been healthier than the general population. Although we attempted to enrol an entire team of athletes, consent may have been driven by personal reasons.
Finally, the regression model was not developed in a clinical setting where CMR usually follows other investigations such as medical and family history, ECG and echocardiography. The combination of these investigations is likely to have diagnosed unambiguous cases of HCM without the requirement for our model. Most importantly, however, the model also performed with a high degree of accuracy in individuals exhibiting borderline LVH, who present the greatest degree of diagnostic uncertainty in clinical practice. Therefore, it seems likely that the proposed model will be of greatest utility in this particular subgroup.
A CMR-based model incorporating the LV EDV : EDM ratio can assist in the differentiation between physiological cardiac adaptation in athletes and HCM, especially in participants exhibiting borderline LVH, who present the greatest degree of diagnostic uncertainty in clinical practice.
What is already known on this topic?
Physical exercise commonly results in increased left ventricular wall mass, which may create a diagnostic overlap with phenotypically mild hypertrophic cardiomyopathy (HCM). Data from echocardiographic studies indicate that training-induced left ventricular hypertrophy (LVH) is usually accompanied by concomitant chamber dilation, in contrast to the reduced cavity size observed in HCM.
What are the new findings?
The cardiac MRI (CMR) ratio of ventricular volume to wall mass proves to be a useful discriminator between athlete's heart and HCM.
A model is presented that can reliably distinguish between HCM cases and healthy athletes and non-athletes.
This study provides novel insight into the diagnostic ‘grey zone’ represented by cases exhibiting mild LVH. The proposed model also performs with a high degree of accuracy in this clinically challenging subgroup.
How it might impact on clinical practice in the near future?
The CMR ratio of ventricular volume to wall mass can be a useful measurement to help differentiate physiological LVH in athletes from pathological LVH in a clinically challenging group.
Contributors TL performed most of the data acquisition, scanning and was the major author of the manuscript. The other main contributor to data acquisition and scanning was NHJP, who also helped revise the manuscript. CFB played a major role in the statistical analyses and study design. AZ, BKV and MJC helped design the study and were involved in the inclusion process of patients as well as the writing of the manuscript. RR, AZ, AM and BD assessed all ECGs and had a role in the editing of the manuscript. WPTMM supported the entire process and helped with the study design and the writing of the manuscript.
Competing interests None.
Patient consent Obtained.
Ethics approval Medical Ethics Committee of University Medical Center Utrecht.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.