## Background

In pediatric echocardiography, myocardial strain measurements are likely influenced by cardiac size and growth in healthy children. The application of this technique in clinical practice has been hampered by the lack of good normal reference values for the pediatric population. The aim of this study was to determine reference values and *Z *score equations for left ventricular systolic circumferential and longitudinal strain in a healthy pediatric population.

## Methods

Two hundred thirty-three healthy subjects 1 to 18 years of age were prospectively recruited. Left ventricular systolic longitudinal and circumferential strain measurements were recorded using two-dimensional speckle-tracking. Normalization for body size was performed using parametric nonlinear regression modeling. Several analyses were performed to detect potential residual associations with body size, residual heteroscedasticity, or departure from an adequate *Z *score distribution.

## Results

There were weak but statistically significant nonlinear associations between body size and most strain values. Body surface area was superior to adjust for body size compared with age, height, and weight. Most strain values displayed a second-order polynomial relationship with body surface area. *Z *score equations were computed with adequate normal distributions and without residual associations in relation to BSA for most strain parameters.

## Conclusions

There was a weak but significant influence of body size on most left ventricular circumferential and longitudinal systolic strain parameters used in pediatric echocardiography. *Z *scores are presented for strain measurements normalized to body surface area and adjusted for heteroscedasticity. The use of these normalized values may reduce the risk for misclassification caused by normal variation in myocardial strain values during growth.

In recent years, speckle-tracking echocardiography has allowed the measurement of myocardial strain and strain rate, which has been demonstrated to be a useful tool for the evaluation of cardiac function in children with acquired and congenital heart disease. The application of the technique in clinical practice has been hampered by the lack of good normal reference values for the pediatric population. Previous studies have suggested that longitudinal and circumferential strain measurements vary with age, whereas others have failed to detect any significant association.

In the present study, we aimed to further investigate the relationship between body size and strain measurements. Currently available pediatric reference values for myocardial strain are presented as means and SDs stratified for age. To our knowledge, there are no published reference values or *Z* score equations based on parametric modeling of echocardiographic strain measurement for body size. Our objective was to determine echocardiographic reference values and parametric *Z* scores equations for left ventricular (LV) longitudinal and circumferential strain measurements in a prospectively recruited healthy cohort of children and adolescents.

## Methods

## Population and Recruitment

Subjects 1 to 18 years in age, without heart disease, were prospectively recruited from local schools, as volunteers, and also from a random selection of patients attending the Hospital for Sick Children cardiac clinic (Toronto, Ontario, Canada). All subjects had normal results on physical examination (with the exception of a physiologic heart murmur) and unremarkable medical histories. Subjects who agreed to participate underwent full anatomic and functional echocardiography to rule out any abnormalities, and any children with abnormal echocardiographic findings, including minor defects such as small atrial septal defect and patent ductus arteriosus, were excluded from the study. The same cohort of healthy subjects was also previously used to determine reference values for pulsed-wave Doppler and tissue Doppler imaging. The institutional research ethics board approved the study, and written informed consent was obtained from each participant.

Age, sex, height, weight, and blood pressure were recorded for every patient at the time of the echocardiographic study. Body surface area (BSA) was estimated using the equation proposed by Haycock *et al *. The Haycock equation was selected because of its closer correlation with echocardiographic measurements in children compared with other published estimations of BSA. Body-mass index (BMI) was calculated as weight divided by the square of height and adjusted for age and sex according to the World Health Organization standards. Subjects with adjusted BMIs outside the second or 98th percentile were excluded.

## Echocardiographic Image Acquisition

All echocardiographic examinations were performed on a Vivid 7 or Vivid E9 ultrasound system (GE Medical Systems, Wauwatosa, WI). The participants were examined at rest while in the left lateral position during sinus rhythm. No sedation was used, and all image acquisition was done during free breathing. We did not attempt to synchronize image acquisitions according to the participant’s breathing. All echocardiographic images were acquired by two experienced research sonographers (C.S., W.H.) using a standardized echocardiographic functional protocol. The appropriate transducer for optimal imaging was selected depending on subject size and image quality. All participants underwent extensive anatomic and functional echocardiographic examinations, which included the acquisition of grayscale images optimized for speckle-tracking analysis. Grayscale images for offline speckle-tracking analysis were acquired at frame rates between 40 and 90 frames/sec. Parasternal short-axis images were acquired at the basal, mid, and apical ventricular levels and used for the assessment of circumferential strain. Standard apical four-chamber view images were used for longitudinal strain assessment.

## Offline Speckle-Tracking Analysis

The grayscale images were analyzed using EchoPAC version 7.0 (GE Medical Systems). Analysis was performed according to the vendor’s instructions. The endocardial border was manually traced (starting at the midseptum for short-axis images and at the basal septum from the apical four-chamber images). Tracking was automatically performed, and the analysis was accepted after visual inspection and when the software indicated adequate tracking. If the tracking was not satisfactory, manual adjustments were made to the tracking points throughout the cardiac cycle. Only image acquisitions with a minimum of four appropriately tracking segments were accepted for analysis. Circumferential strain curves from the three short-axis views (basal, mid, and apical) and six segments (anterior septum, anterior, lateral, posterior, inferior, and septum) were analyzed. Longitudinal strain curves from the apical four-chamber view were analyzed for six segments (basal septum, mid septum, apical septum, apical lateral, mid lateral, and basal lateral).

We measured ventricular systole from onset of the QRS complex on the electrocardiogram to the termination of flow measured from pulsed-wave Doppler aortic tracing immediately proximal to the aortic valve to obtain valve clicks whenever possible. Figure 1 shows the placement of the maker on the aortic valve click. Figure 2 demonstrates the difference between peak systolic strain and peak strain.

## Modeling and *Z *Scores

*Z *scores were computed using a standardized approach that has been described previously. A detailed stepwise approach to regression modeling and *Z *score validation is available in the Supplemental Material . Briefly, we tested several bivariate prediction models (linear, allometric, polynomial, and gamma function models) on four independent variables (age, weight, height, and BSA). When heteroscedasticity was detected, we used weighted models as described previously. *Z *score equations were computed and tested for each model and each independent variable.

To select the most adequate model, we first assessed the goodness of fit of the independent variable on the echocardiographic measurement by visual inspection and by reviewing various fit diagnostic aids. Second, we inspected the residual values (observed values minus predicted value) against the independent variable to ensure that no residual association was present. If no residual association was present, we assessed the *Z *score distribution to ensure that no significant departure from the standard normal distribution was present. Identification of departure from the normal distribution was done by visually inspecting the distribution histograms and the normal probability plot, and by calculating the Shapiro-Wilk statistic. To compute *Z *score equations that would yield intuitively lower *Z *scores for measurements closer to zero, we used the absolute value of strain measurements with negative values. Other mathematical transformations of the dependent variables (logarithmic or reciprocal transformation) were considered only if the distribution of the normalized echocardiographic measurement suggested significant departure from the normal distribution.

When a final model was selected, the newly computed *Z *scores were further tested for residual association for body size by plotting *Z *scores against age, weight, height and BSA. We also verified associations with sex, resting heart rate, and BMI adjusted for age and sex.

## Statistical Analysis

We used SAS for Windows version 9.4 for all analyses (SAS Institute, Cary, NC). Mean *Z *scores were compared using Student’s *t *test. The *t *statistic was used to estimate *P *values for residual association. Departure from a normal distribution was evaluated by the Shapiro-Wilk statistic. A *P *value < .05 was considered to indicate statistical significance, except for the Shapiro-Wilk statistic, for which a *P *value < .01 was considered to indicate statistical significance.

## Results

## Population

Two hundred thirty-three subjects met the inclusion criteria and were included in the final analysis (52% female subjects). The age distribution is shown in Figure 3 , and subjects’ characteristics are provided in Table 1 . The mean age was 10.8 years, and the mean BSA was 1.25 m ^{2 }. The mean BMI *Z *score was 0.2. Blood pressure was within normal limits for every subject. Overall, image quality was adequate to measure strain in 91% of subjects (varying from 82% to 96% depending on the measured segments). A total of 7,222 measurements were analyzed. Intraobserver and interobserver agreement for global circumferential and longitudinal strain are presented in the Supplemental Material .

Characteristic | Value | Range |
---|---|---|

Sex (male) | 112 (48.1%) | |

Age (y) | 12.1 (8.9) | 1.3 to 18 |

Height (cm) | 151 (48.8) | 77 to 186 |

Weight (kg) | 40.9 (34.6) | 8.7 to 92.0 |

BSA (m ^{2 }) |
1.30 (0.75) | 0.43 to 2.17 |

BMI (kg/m ^{2 }) |
17.8 (4.7) | 13.4 to 27.8 |

BMI Z score |
0.27 (1.4) | −1.9 to 1.9 |

## Regression Models and Dependent Variables

We found weak but significant nonlinear associations between most strain measurements and all body size variables (weight, height, age, and BSA). Adjusting for BSA was better than adjusting for age, weight, and height to account for the effect of body size (i.e., a residual association with body size was more likely to be present when *Z *scores were computed for weight, age, or height compared with BSA; data not shown). BSA was therefore selected to compute final *Z *score equations.

Table 2 shows equations for predicted mean and *Z *scores for each type of model. *Z *scores could be computed for 31 of 33 strain measurements. For two measurements (basal anterior circumferential strain and apical lateral longitudinal strain), none of the models could produce *Z *scores with a normal distribution and without significant residual association with BSA. For the remaining 31 measurements, the selected models and parameters for circumferential and longitudinal strain are shown in Tables 3 and 4 , respectively. Parameters in Tables 3 and 4 may be used in equations displayed in Table 1 to calculate *Z *scores. Second-order polynomial models were selected for 21 measurements (64%). Linear model and allometric models were used for four and two measurements, respectively. Significant heteroscedasticity was often detected, and weighted models were used for 29 measurements. Four measurements did not vary significantly with BSA and were not modeled for BSA (circumferential strain for all three mid segments and for the basal anterior septal segments). We nonetheless provide intercept coefficients in Tables 3 and 4 to compute *Z *scores.

Description | Equation for the predicted mean | Equation for Z score |
---|---|---|

No model | <SPAN role=presentation tabIndex=0 id=MathJax-Element-1-Frame class=MathJax style="POSITION: relative" data-mathml='y=c’>y=cy=c y = c |
<SPAN role=presentation tabIndex=0 id=MathJax-Element-2-Frame class=MathJax style="POSITION: relative" data-mathml='Z=obs−cf×BSA+g’>Z=obs−cf×BSA+gZ=obs−cf×BSA+g Z = obs − c f × BSA + g |

Linear | <SPAN role=presentation tabIndex=0 id=MathJax-Element-3-Frame class=MathJax style="POSITION: relative" data-mathml='y=b×BSA+c’>y=b×BSA+cy=b×BSA+c y = b × BSA + c |
<SPAN role=presentation tabIndex=0 id=MathJax-Element-4-Frame class=MathJax style="POSITION: relative" data-mathml='Z=obs−(c×BSA+d)f×BSA+g’>Z=obs−(c×BSA+d)f×BSA+gZ=obs−(c×BSA+d)f×BSA+g Z = obs − ( c × BSA + d ) f × BSA + g |

Log-linear | <SPAN role=presentation tabIndex=0 id=MathJax-Element-5-Frame class=MathJax style="POSITION: relative" data-mathml='ln(y)=b×BSA+c’>ln(y)=b×BSA+cln(y)=b×BSA+c ln ( y ) = b × BSA + c |
<SPAN role=presentation tabIndex=0 id=MathJax-Element-6-Frame class=MathJax style="POSITION: relative" data-mathml='Z=ln(obs)−(b×BSA+c)f×BSA+g’>Z=ln(obs)−(b×BSA+c)f×BSA+gZ=ln(obs)−(b×BSA+c)f×BSA+g Z = ln ( obs ) − ( b × BSA + c ) f × BSA + g |

Polynomial (second order) | <SPAN role=presentation tabIndex=0 id=MathJax-Element-7-Frame class=MathJax style="POSITION: relative" data-mathml='y=a×BSA2+b×BSA+c’>y=a×BSA2+b×BSA+cy=a×BSA2+b×BSA+c y = a × BSA 2 + b × BSA + c |
<SPAN role=presentation tabIndex=0 id=MathJax-Element-8-Frame class=MathJax style="POSITION: relative" data-mathml='Z=obs−(a×BSA2+b×BSA+c)f×BSA+g’>Z=obs−(a×BSA2+b×BSA+c)f×BSA+gZ=obs−(a×BSA2+b×BSA+c)f×BSA+g Z = obs − ( a × BSA 2 + b × BSA + c ) f × BSA + g |

Reciprocal of polynomial (second order) | <SPAN role=presentation tabIndex=0 id=MathJax-Element-9-Frame class=MathJax style="POSITION: relative" data-mathml='1y=a×BSA2+b×BSA+c’>1y=a×BSA2+b×BSA+c1y=a×BSA2+b×BSA+c 1 y = a × BSA 2 + b × BSA + c |
<SPAN role=presentation tabIndex=0 id=MathJax-Element-10-Frame class=MathJax style="POSITION: relative" data-mathml='Z=(1obs)−(a×BSA2+b×BSA+c)f×BSA+g’>Z=(1obs)−(a×BSA2+b×BSA+c)f×BSA+gZ=(1obs)−(a×BSA2+b×BSA+c)f×BSA+g Z = ( 1 obs ) − ( a × BSA 2 + b × BSA + c ) f × BSA + g |

Log-polynomial (second order) | <SPAN role=presentation tabIndex=0 id=MathJax-Element-11-Frame class=MathJax style="POSITION: relative" data-mathml='ln(y)=a×BSA2+b×BSA+c’>ln(y)=a×BSA2+b×BSA+cln(y)=a×BSA2+b×BSA+c ln ( y ) = a × BSA 2 + b × BSA + c |
<SPAN role=presentation tabIndex=0 id=MathJax-Element-12-Frame class=MathJax style="POSITION: relative" data-mathml='Z=ln(obs)−(a×BSA2+b×BSA+c)f×BSA+g’>Z=ln(obs)−(a×BSA2+b×BSA+c)f×BSA+gZ=ln(obs)−(a×BSA2+b×BSA+c)f×BSA+g Z = ln ( obs ) − ( a × BSA 2 + b × BSA + c ) f × BSA + g |

Allometric | <SPAN role=presentation tabIndex=0 id=MathJax-Element-13-Frame class=MathJax style="POSITION: relative" data-mathml='y=d×BSAe’>y=d×BSAey=d×BSAe y = d × BSA e |
<SPAN role=presentation tabIndex=0 id=MathJax-Element-14-Frame class=MathJax style="POSITION: relative" data-mathml='Z=obs−(d×BSAe)f×BSA+g’>Z=obs−(d×BSAe)f×BSA+gZ=obs−(d×BSAe)f×BSA+g Z = obs − ( d × BSA e ) f × BSA + g |