Author + information
- Received February 21, 2017
- Accepted April 1, 2017
- Published online March 5, 2018.
- Jonas Selmeryd, MDa,b,∗ (, )
- Egil Henriksen, MD, PhDa,b,
- Håvard Dalen, MD, PhDc,d,e and
- Pär Hedberg, MD, PhDa,b
- aDepartment of Clinical Physiology, Västmanland County Hospital, Västerås, Sweden
- bCentre for Clinical Research, Uppsala University, Västmanland County Hospital, Västerås, Sweden
- cLevanger Hospital, Nord-Trøndelag Health Trust, Levanger, Norway
- dDepartment of Cardiology, St. Olav’s University Hospital, Trondheim, Norway
- eCardiac Exercise Research Group, Department of Circulation and Medical Imaging, Norwegian University of Science and Technology, Trondheim, Norway
- ↵∗Address for correspondence:
Dr. Jonas Selmeryd, Department of Clinical Physiology, Västmanland County Hospital, SE-72189 Västerås, Sweden.
Objectives This study aimed to derive age-specific multivariate reference regions (MVRs) able to classify subjects into those having normal or abnormal filling patterns and to evaluate the prognostic impact of this classification.
Background The integration of several parameters is necessary to diagnose disorders of left ventricular (LV) filling because no single measurement accurately describes the complexity of diastolic function. However, no generally accepted validated multiparametric algorithm currently exists.
Methods A cohort of 1,240 apparently healthy subjects from HUNT (Nord-Trøndelag Health Study) with measured early (E) and late (A) mitral inflow velocity and early mitral annular diastolic tissue velocity (e′) were used to derive univariate 95% reference bands and age-specific MVRs. Subsequently, the prognostic impact of this MVR-based classification was evaluated by Cox regression in a community-based cohort (n = 726) and in a cohort of subjects with recent acute myocardial infarction (n = 551). Both evaluation cohorts were derived from VaMIS (the Västmanland Myocardial Infarction Study).
Results Univariate reference bands and MVRs are presented graphically and as regression equations. After adjustment for sex, age, hypertension, body mass index, diabetes, prior ischemic heart disease, LV mass, LV ejection fraction, and left atrial size, the hazard ratio associated with abnormal filling patterns in the community-based cohort was 3.5 (95% confidence interval: 1.7 to 7.0; p < 0.001) and that in the acute myocardial infarction cohort was 1.8 (95% confidence interval: 1.1 to 2.8; p = 0.011).
Conclusions This study derived age-specific MVRs for identification of abnormal LV filling patterns and showed, in a community-based cohort and in a cohort of patients with recent acute myocardial infarction, that these MVRs conveyed important prognostic information. An MVR-based classification of LV filling patterns could lead to more consistent diagnostic algorithms for identification of different filling disorders.
Systolic heart failure and asymptomatically reduced left ventricular (LV) systolic function are clinical entities for which a high degree of consensus exists on diagnostic criteria (1). The opposite holds true for heart failure with preserved ejection fraction (HFpEF) and for diastolic dysfunction (DDF), the diagnostic criteria for which are constantly debated and revised (2–6).
Several echocardiographic measurements and derived parameters reflect, in various degrees, different aspects of LV filling. However, these parameters all have their individual shortcomings with regard to accuracy, reproducibility, and/or feasibility (7,8). The proposed solution to the shortcomings of individual measurements has been to use a multiparametric integrative approach (1,8). Several multiparametric classification schemes have been proposed to aid in identifying subjects with HFpEF or DDF, but the concordance among these methods has been shown to be poor, thus indicating that further work is needed (9). Notably, multiparametric diagnostic schemes presented in recommendations are frequently formulated on the basis of expert consensus and theoretical considerations rather than on clinical studies (10). We believe this to be a consequence of the scarcity of studies exploring how multiparametric diagnostic schemes could be constructed. Within the field of clinical chemistry, multivariate reference regions (MVRs) have been developed to facilitate identification of abnormal patterns on multivariate test profiles for arterial blood gases (11), thyroid function (12), and insulin-like growth factors (13). An MVR for interpretation of echocardiographic Doppler data could function as a simple and intuitive test to identify subjects with filling disorders.
In this study, we aimed to derive an MVR with the ability to classify subjects into those having a normal filling pattern (NFP) or an abnormal filling pattern (AFP) and evaluate the prognostic impact of such a classification.
HUNT3 (the third phase of the Nord-Trøndelag Health Study) was used as the derivation cohort. This cohort was described in detail previously (14). Briefly, 94,194 people from the general population were invited to participate in HUNT3, and 50,839 (54%) accepted the invitation. A subsample of subjects free of cardiovascular disease, hypertension, and diabetes was selected by random sampling (n = 1,296) for an echocardiographic substudy. Subjects whose echocardiograms revealed arrhythmia, valvular disease, or any kind of myocardial disease (n = 30) were excluded. The echocardiographic measurements used in the present study were early mitral annular diastolic tissue velocity, measured by pulsed tissue Doppler, averaged for the lateral and septal wall (e′), early mitral inflow velocity (E), and late mitral inflow velocity (A). Subjects with missing values for A, E, or e′ (n = 26) were excluded, thereby leaving 1,240 subjects to form the derivation cohort. The study was approved by the Regional Committee for Medical Research Ethics and was conducted according to the principles of the second Declaration of Helsinki. Written informed consent was obtained from all participants. Data on the intraobserver and interobserver variability of the derivation cohort were published earlier (14).
The evaluation cohorts were derived from VaMIS (the Västmanland Myocardial Infarction Study); the inclusion criteria and study protocol were reported previously (15,16). Briefly, subjects hospitalized for acute myocardial infarction (AMI) between November 2005 and May 2011 were included in VaMIS. For each included patient, a control subject was recruited from the general population. From the Swedish Population Register in which all Swedish citizens are registered, a subject of the same sex with the nearest date of birth and living in the same municipality as the VaMIS patient was identified and invited to participate. All subjects underwent clinical examination, echocardiographic examination, and blood sampling. LV mass was calculated from the LV wall thickness and diameter and was indexed for body surface area (BSA) as left ventricular mass indexed for body surface area (LVMi) (17). LV ejection fraction was categorized from biplane Simpson measurements in subjects with adequate image quality and otherwise by visual estimation into the following categories: ≥55%; 45% to 54%; 35% to 44%; 25% to 34%; and <25%. Left atrial volume was measured by monoplane Simpson in the apical 4-chamber view and indexed for BSA. Reproducibility in the evaluation cohorts was tested by repeat measurement in a random subset of 22 subject by 2 observers (P.H. and J.S.), including 1 who performed the original measurements (P.H.). The intraobserver and interobserver coefficients of variation for E, A, and e′ were <5.2% for all. VaMIS was approved by the Ethics Committee of Uppsala University, Sweden. All participants gave written informed consent.
From the myocardial infarction cohort (n = 1,008), subjects <30 years or >80 years of age (n = 239), with nonsinus rhythm (n = 39), with valvular disease of more than moderate grade (n = 39), or with missing values for AFP or any of the predictors in the Cox regression model (n = 140) were excluded, leaving 551 subjects who formed the AMI evaluation cohort (AMI cohort). From the control group of VaMIS (n = 855), we excluded subjects <30 years or >80 years of age (n = 48), with nonsinus rhythm (n = 29), with valvular disease of more than moderate grade (n = 6), and with missing values (n = 46), thus leaving 726 subjects who formed the community-based evaluation cohort (CB cohort). Subjects excluded from the AMI cohort because of missing values were older (67.0 years vs. 65.3 years; p = 0.056), more obese (body mass index [BMI] 29.2 kg/m2 vs. 27.2 kg/m2; p < 0.001), and more frequently hypertensive (59% vs. 45%; p = 0.003) than the included subjects. In the CB cohort, subjects with missing values included a higher proportion of women (48% vs. 28%; p = 0.006) and subjects with hypertension (61% vs. 35%; p < 0.001) and were older (69.0 years vs. 65.8 years; p = 0.023), more obese (BMI 28.7 kg/m2 vs. 26.6 kg/m2; p < 0.001), and had lower LVMi (91.0 g/m2 vs. 98.7 g/m2; p = 0.024) but similar LV mass (182.3 g vs. 195.2 g; p = 0.117) compared with the included subjects.
Continuous data were expressed as mean ± SD if approximately normally distributed or otherwise as median and interquartile range. Categorical data were expressed as numbers and percentages. Continuous variables were compared using the unpaired Student t test if approximately normally distributed or the Wilcoxon rank sum test for skewed distributions. Categorical variables were compared using the Fisher exact test. The results were regarded as significant when p < 0.05. STATA software version 14.1 (StataCorp LP, College Station, Texas) was used for all statistical analyses.
For a detailed description of the statistical methodology used for derivation of the age-specific MVRs, see the Online Appendixes. E, A, and e′ were transformed and modeled as functions of age by fractional polynomial regression. Assuming normally distributed residuals and constant variance, approximate 95% univariate reference bands were calculated as age-predicted mean ±1.96 times the model’s SD on the transformed scale followed by a back-transformation onto the original scale. Assuming constant variance and covariance of the residuals, an equation for the age-specific multivariate Mahalanobis distance, a statistical measurement of the “unusualness” of an observation, was established. By assuming multivariate normality of the residuals (Online Figure 1), the equation of an age-specific 95% MVR could be derived. This equation describes a 3-dimensional ellipsoid in the space of E, A, and e′. The size and geometry of this ellipsoid changes with age. Observations with filling patterns that fall within these age-specific ellipsoids can be considered normal for age analogously with how univariate reference bands are interpreted.
The age-specific Mahalanobis distance was rescaled into a filling pattern abnormality score (FPAS) so that an FPAS ≤1 was characteristic of an observation within the 95% MVR. An FPAS was calculated for all study subjects in the evaluation cohorts, and subjects with an FPAS >1 were considered to have an AFP (i.e., subjects with a filling pattern that fell outside the MVR for his or her age).
The outcome used for analysis was all-cause mortality. The basis of the follow-up information was the Swedish Population Register. Participants were followed up from their inclusion date to their date of death or censoring date of May 26, 2016. The relationship between filling pattern status and outcome was evaluated by Kaplan-Meier survival curves, and differences in survival were compared with the log-rank test. The additional prognostic value of AFP to clinical and echocardiographic risk markers was further evaluated by Cox regression adjusted for sex, age, hypertension, BMI, diabetes, prior ischemic heart disease, LVMi, LV ejection fraction, and left atrial volume indexed for body surface area.
The basic characteristics of the derivation cohort are outlined in Table 1. Regression equations for univariate reference bands and MVRs valid for ages 30 to 80 years can be found in the Online Appendixes. Graphic representations of univariate reference bands can be seen in Figure 1 and of the MVRs in Figure 2. To facilitate critical evaluation of the goodness-of-fit, separate graphs for each age group, with scatter superimposed, can be found in Online Figures 2 to 4.
The basic characteristics of the evaluation cohorts partitioned according to the presence of AFP are shown in Table 2. The prevalence of AFP in the CB cohort was 6.3% (95% confidence interval [CI]: 4.7% to 8.4%), and in the AMI cohort it was 21.4% (95% CI: 18.1% to 25.1%). In a subgroup of the CB cohort that comprised subjects free of diabetes, hypertension, and ischemic heart disease (n = 376), the prevalence of AFP was 5.3% (95% CI: 3.3% to 8.1%). During a median follow-up time of 7.6 years, 61 subjects in the CB cohort and 117 subjects in the AMI cohort died. Kaplan-Meier survival curves are presented in Figure 3. In both evaluation cohorts, survival was significantly reduced in subjects with AFP compared with subjects with NFP (p < 0.001). In univariate Cox regression analyses, the presence of AFP was associated with an increased mortality risk as indicated by a hazard ratio (HR) of 3.9 (95% CI: 2.1 to 7.3; p < 0.001) in the CB cohort and of 2.6 (95% CI: 1.8 to 3.7; p < 0.001) in the AMI cohort (Table 3). After adjustment for clinical and echocardiographic risk markers, the mortality risk associated with AFP remained significant, with an HR of 3.5 (95% CI: 1.7 to 7.0; p < 0.001) in the CB cohort and an HR of 1.8 (95% CI: 1.1 to 2.8; p = 0.011) in the AMI cohort.
In this paper, we present derived age-specific MVRs for echocardiographic parameters commonly used in the evaluation of LV diastolic function and describe how subjects with AFP can be identified. In addition, we show that subjects with AFP are at an increased risk of death compared with subjects with NFP, both in a CB cohort and among patients with recent AMI.
Univariate reference intervals for common indices of filling have been published previously (14,18,19). However, to use multiple univariate reference intervals side by side to define normality is not advisable, for several reasons. First, specificity is lost for every measurement added (20,21). For instance, if 3 measurements with 95% reference intervals are used, it can be shown that the proportion of false-positive results will be as high as 1−0.953 (i.e., approximately 14%) (20). Second, sensitivity is lost because measurements within univariate reference intervals can display highly unusual patterns when they are analyzed in combination (20,21). For instance, for a man, a height of 185 cm and a weight of 65 kg are both univariately normal, although the combination is extremely unusual. Third, with univariate reference intervals, discrepancies among measurements are common and may result in interpretative difficulties and large proportions of subjects that are considered unclassifiable (1). These shortcomings can be avoided by using MVRs (20,21).
An E/A ratio of approximately 0.6 to 2.0 and an E/e′ ratio <15 have previously been shown to identify subjects with normal mitral inflow (14,18,22). Interestingly, the MVRs presented in Figure 2 coincide well with these cutoffs for subjects 50 to 70 years of age who have an e′ in the range of 7 to 10 cm/s, which are probably characteristic of the majority of patients evaluated by echocardiography on clinical grounds. The E/A and E/e′ ratios can therefore in a sense be seen as simplified multivariate normal delimiters. However, in younger subjects and in subjects with high or low e′ values, these cutoffs appear to be less optimal. For instance, the E/A ratio frequently exceeds 2 in younger subjects and/or in subjects with high e′. The finding of an E/A ratio <1 is rare in younger subjects, whereas ratios <0.6 frequently can be encountered in older subjects with low e′. In older subjects with low e′, an E/A ratio <1 is, in fact, the expected finding, whereas an E/A ratio >1 may indicate an abnormality. Moreover, our data suggest that in older subjects an E/e′ ratio >15 can be regarded as a normal finding if the E/A ratio is <1 and/or e′ is <7 cm/s. Further, it seems that using E/e′ >15 as a cutoff for younger subjects or subjects with relatively high e′ could result in suboptimal sensitivity because the multivariate upper E/e′ limit for such subjects is closer to 8. One could speculate that the size of the gray zone of E/e′ 8 to 15 cm/s, where conclusions on filling pressure cannot be drawn from E/e′ alone (23), could be reduced by using multivariate analysis with the proposed MVRs to account for the impact of age, E, A, and e′ on the expected values of E/e′.
A potential use of the MVRs presented here could be to aid in developing diagnostic algorithms for identification of DDF, HFpEF, and elevated filling pressures. However, as indicated by the discussion in the previous paragraph, it is likely that such a decision tree or scoring system would need to be complex. A simpler approach could be to use the MVRs directly to establish first whether the mitral inflow is normal or abnormal and then use univariate z-scores or absolute values of E, A, and e′ to characterize the deviation. Previously published recommendations on diagnostic criteria for DDF have been interpreted very differently in different studies, thus making interstudy comparison hazardous (16). A strength of using the proposed MVRs is that there is no room for subjective interpretation of a complex diagnostic algorithm. Furthermore, it is possible to obtain the parameters used (E, A, and e′) for almost all patients (14). We believe that an MVR-based definition of AFP would yield more consistent definitions of different filling disorders.
The age-specific MVRs were evaluated in 2 separate cohorts. The prevalence of AFP was 21% in the AMI cohort and 6.3% in the CB cohort. In a subgroup of the CB cohort free of subjects with hypertension, diabetes, and ischemic heart disease, the prevalence of AFP was 5.3%, which is close to the expected 5.0% in subjects with characteristics similar to the derivation cohort. When evaluated against all-cause mortality, the presence of AFP was found to be a strong predictor of all-cause mortality even after adjustment for several clinical and echocardiographic risk markers. This held true among subjects in both the CB cohort and the AMI cohort (Table 3).
It should be stressed that subjects with AFP are a heterogeneous cohort because many different forms of abnormal patterns are represented, as can be appreciated from Table 2. For instance, the subjects with AFP in the CB cohort include subjects with both restrictive filling and impaired relaxation, as reflected by the 13% of subjects with E/e′ >15, the 17% with E/A >2, and the 13% with E/A <0.6. As could be expected, the predominant pattern in subjects of the AMI cohort with AFP was that of restrictive filling and elevated filling pressure with 24% having an E/A >2 and 36% having an E/e′ >15. Further studies are required to characterize more precisely the different patterns among subjects categorized by the proposed methodology as having AFP.
An unresolved question in diagnosing DDF is whether age should be taken into account. As can be seen from Figures 1A to 1C, E and e′ decrease with age, and A increases. By using fixed cutoffs unrelated to age, the prevalence of DDF will appear to increase with age. Because all body tissues, including cardiac tissue, lose elasticity with age, this can be regarded as reasonable, and some of the decreased working capacity of older people can probably be explained by the DDF that develops with age. Furthermore, it is not uncommon in other areas of medicine to use fixed cutoffs, even if the age-specific normal limits change with age. For example, when we use blood pressure to diagnose hypertension we ignore the influence of age. However, a substantial difference between hypertension and DDF is that hypertension can be treated, whereas no treatment is yet available for age-induced DDF. The diagnosis of age-induced DDF is therefore of little value to either the clinician or the patient. Furthermore, an erroneous diagnosis of DDF in an older dyspneic patient could carry the risk of overlooking other serious conditions. Pending effective treatment, we believe that it is better to consider only DDF beyond that which can be explained by age.
First, we have ignored the effect of sex on mitral inflow and tissue Doppler velocities. Mitral inflow velocities and e′ have been shown to be higher in women, but because the differences in the absolute values are small (22) and because the influence of sex is routinely ignored in clinical guidelines (10), the effect of sex was regarded as clinically insignificant and was therefore ignored in the present study. Second, the use of the age-specific MVRs for ages beyond 30 to 80 years is not recommended because the assumption of constant variance and covariance was not met outside this age interval. Third, even though the derivation cohort was free of known cardiovascular disease, hypertension, and diabetes, unknown subclinical disease could have been present. Fourth, the derivation and evaluation cohorts consisted mainly of white subjects, and whether the findings can be extrapolated to other ethnicities is uncertain. Fifth, a slight selection bias could be present in the mortality analysis because difficult-to-examine obese, female, and/or old subjects were overrepresented among the subjects excluded because of missing values. Sixth, we believe that directly trying to use the MVRs, as presented in Figure 2, in a clinical workflow would be perceived as cumbersome and complex. Rather, the regression equations have to be integrated into the echocardiography evaluation software so that 4 clicks (E, A, lateral e′, and septal e′) and the knowledge of age instantaneously inform the user about whether filling is normal or abnormal and to what degree (FPAS). Finally, in the absence of a true reference method (i.e., invasive cardiac catheterization), we could not evaluate the ability of the proposed MVRs to identify HFpEF or DDF. It is important to stress that the finding of an AFP does not equate with the presence of HFpEF or DDF. We strongly suspect that subjects can have HFpEF or DDF despite having an NFP. We also believe that an AFP should be considered a nonspecific finding that could indicate HFpEF or DDF but that could also be caused by many other cardiac and extracardiac disease states (e.g. hypovolemia or hypervolemia, valvular heart disease, bundle branch blocks, impaired systolic function, or impaired left atrial function). As a consequence, it is likely that the mitral inflow MVRs will need to be used in conjunction with auxiliary parameters (i.e., left atrial size and pulmonary pressure), as proposed in recent guidelines (10).
We describe a methodology to classify subjects into those having a normal or an abnormal LV filling pattern, by taking several parameters into account simultaneously, and show that this classification has important prognostic implications in a CB sample as well as in subjects with a recent AMI. The classification, in conjunction with auxiliary parameters, can potentially be used to identify DDF, HFpEF, and elevated filling pressures. Further studies are required to establish the ability of the age-specific MVRs to identify relevant disease and also to characterize the heterogeneous group identified to have abnormal LV filling patterns more precisely.
COMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS: Because no single echocardiographic measurement accurately describes LV diastolic function, clinicians must use complex algorithmic approaches to integrate several parameters. The age-specific MVRs derived in the present study show promise for simplifying the integration of commonly used Doppler measurements by answering a simple question: is the LV filling pattern normal for the patient’s age?
TRANSLATIONAL OUTLOOK: A population of subjects with AFPs, as identified by the proposed age-specific MVRs, is likely to be very heterogeneous. Further studies are required to determine which distinct clinical entities are represented among such subjects and whether it is possible to differentiate among them by echocardiography. In addition, evaluation of the diagnostic performance for identification of HFpEF and DDF of the age-specific MVRs is desirable.
The authors thank the Nord-Trøndelag Health Study (HUNT) Research Center for giving them access to the HUNT3 data. The HUNT Study is a collaboration among the HUNT Research Center (Faculty of Medicine, Norwegian University of Science and Technology), the Nord-Trøndelag County Council, the Central Norway Health Authority, and the Norwegian Institute of Public Health. The authors also thank the research staff at the Centre for Clinical Research at Västmanland County Hospital, Västerås, Sweden for their contributions to VaMIS (Västmanland Myocardial Infarction Study). Finally, the authors are grateful for advice received from Lars Lindhagen (Uppsala Clinical Research Center, Sweden) and user mvw at Stackexchange.com (http://math.stackexchange.com/questions/1592118/).
The VaMIS Study was supported by grants from Sparbanksstiftelsen Nya, the County of Västmanland, Selanders Stiftelse, and the Swedish Medical Association. The HUNT3 study was funded by the Norwegian University of Science and Technology. The sponsors did not take an active role in design or conduct of the study, data collection, analysis, or manuscript preparation. All authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- late mitral inflow velocity
- abnormal filling pattern
- acute myocardial infarction
- body surface area
- confidence interval
- diastolic dysfunction
- deceleration time
- early mitral inflow velocity
- early mitral annular diastolic tissue velocity
- filling pattern abnormality score
- heart failure with preserved ejection fraction
- hazard ratio
- left ventricular
- left ventricular mass indexed for body surface area
- multivariate reference region
- normal filling pattern
- Received February 21, 2017.
- Accepted April 1, 2017.
- 2018 American College of Cardiology Foundation
- Kitzman D.W.,
- Little W.C.
- Burkhoff D.
- Maurer M.S.,
- Spevack D.,
- Burkhoff D.,
- Kronzon I.
- Al-Jaroudi W.A.,
- Thomas J.D.,
- Rodriguez L.L.,
- Jaber W.A.
- Petrie M.C.,
- Hogg K.,
- Caruana L.,
- McMurray J.J.
- Nistri S.,
- Ballo P.,
- Mele D.,
- et al.
- Nagueh S.F.,
- Smiseth O.A.,
- Appleton C.P.,
- et al.
- Dalen H.,
- Thorstensen A.,
- Vatten L.J.,
- Aase S.A.,
- Stoylen A.
- Condén E.,
- Rosenblad A.
- Lang R.M.,
- Badano L.P.,
- Mor-Avi V.,
- et al.
- Ommen S.R.,
- Nishimura R.A.,
- Appleton C.P.,
- et al.