1 INTRODUCTION

Overweight and obesity (OW/Ob) are globally important health disorders that affect all age-groups. Together, their prevalence has increased by 47.1% in children, compared to 27.5% in adults, between 1980 and 2013, resulting in approximately 23% of children in developed countries being classified as OW/Ob.1 The early-onset diseases that result from childhood OW/Ob place a costly burden on economies, estimated at $14 billion a year for America alone.2 Cardiovascular disease (CVD) contributes most to this, accounting for 68.6% of all obesity-related deaths.3 Abnormalities of left ventricular diastolic function (LVDF) are major contributors to such CVD, with >80% of diastolic heart failure (HF) patients being overweight or obese.4

LVDF describes the ability of the left ventricle (LV) to fill with blood during diastole, which completes in four stages: (I) isovolumic relaxation; (II) rapid filling; (III) diastasis; and (IV) left atrium (LA) contraction (Figure 1). Numerous physiological parameters influence LVDF, including the rate of early myocardial lengthening in diastole, and thus filling, which is determined by a combination of active (energy-utilizing) and passive forces. Impaired LVDF includes a number of pathological processes, including impaired relaxation, increased myocardial stiffness, and elevated LA pressure (LAP).

Stages of diastole and echocardiography measures of diastolic function. Stage I—isovolumic relaxation (IVR) which occurs after aortic valve closure and before mitral valve opening, as left ventricle (LV) pressure falls rapidly until it reaches left atrial pressure, prompting mitral valve (MV) opening; Stage II—rapid filling, where the MV is open and blood is suctioned towards the apex of the LV from the left atrium (LA), which occurs as the myocardium lengthens during falling LV pressure; Stage III—diastasis, after initial filling where LA and LV pressures equalize and flow ceases; Stage IV—LA contraction, which generates an additional pressure gradient that drives more blood into the LV. A wave indicates late mitral inflow peak velocity; a′, late diastolic tissue peak velocity; DT, E wave deceleration time; E wave, early mitral inflow peak velocity; e′, early diastolic tissue peak velocity; IVRT, isovolumic relaxation time

An array of interrelated indices obtained by echocardiography or cardiovascular magnetic resonance (CMR) exist to indirectly assess LVDF. Although these are all subject to influence by the factors described above, there is evidence that some measures may be better at differentiating particular elements of LV diastolic dysfunction (LVDD) than others. For example, some measures are less influenced by LAP.

Tissue Doppler imaging (TDI) measures longitudinal myocardial motion at the basal septum or lateral ventricular wall in response to inflow. TDI measures differ from conventional Doppler ultrasound measures of transmitral blood flow velocity because they assess longitudinal rather than global compliance of the ventricle, and are less influenced by LV loading conditions.5-7 It has been suggested that these differences may improve their ability to detect early LVDD,7 particularly in conditions such as obesity where volume overload occurs.

The chronic volume overload and metabolic abnormalities of obesity are associated with progressive LVDD and eventual diastolic HF, through impairment of myocardial relaxation and passive LV properties.8 Detection methods for LVDD in adults are well-established,9 but it is unclear which measures best detect the earliest stages of LVDD and would, therefore, be most suitable in adolescents and particularly in those with OW/Ob. Reliable early detection is important as LVDD reversibility is still potentially achievable and lifestyle habits may be less fixed than in adulthood.

Although there are many studies of LVDF in children with obesity, diverse methods and group definitions have made it difficult to adequately summarize findings by conventional meta-analysis, with study heterogeneity being identified as the main limiting factor in earlier attempts.10, 11 Nevertheless, statistical synthesis should be possible, given that most studies share common measures of adiposity and LVDF. This would confirm whether or not OW/Ob is associated with impaired LVDF at a young age, which is yet to be clearly established.

Although most studies focus on measures of adiposity, a significant number have addressed other cardiometabolic risk factors (CMRFs), such as insulin resistance (IR),12-14 that often accompany OW/Ob and may directly impair LVDF. Where data allow, we aimed to assess the relationships of these other CMRFs with LVDF.

In this systematic review and meta-analysis, our aims were to determine in a population of children and adolescents: (1) the extent to which OW/Ob is associated with various measures of LVDF; (2) measures of LVDF which are most strongly associated with OW/Ob and; (3) the associations of IR and other CMRFs with LVDF measures.

2 METHODS

The protocol for this review was registered with PROSPERO International Prospective Register of Systematic Reviews (identifier CRD42020177470). This review was completed in accordance with the PRISMA 2020 (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines.15

3 CRITERIA FOR CONSIDERING STUDIES FOR THIS REVIEW 3.1 Types of studies

Cross-sectional studies, controlled intervention studies, and pre-post studies that examined the association of childhood and adolescent OW/Ob with LVDF were included. Inclusion was limited to full-text articles reported in English and published in peer-reviewed journals. We excluded studies published in grey literature sources and conference or meeting abstracts without a full text.

3.2 Types of participants

Individuals aged <18 years were included in accordance with the international definition of childhood. Additionally, individuals aged 10–24 years were included and defined as adolescents in accordance with the widest accepted definition to ensure that articles that used this definition were not rejected.16 It has been suggested that this definition of adolescence corresponds best with contemporary features of adolescent growth and social role transitions.16 Almost every study used individual criteria to define their OW/Ob and control groups. Furthermore, the pathological group in some studies was obesity only while others included overweight in this group. In other studies, overweight was grouped with normal weight as a control group. Therefore, study-specific group definitions are reported in the results, but were ignored in the meta-analysis as this heterogeneity did not allow for meaningful group-based comparisons.

3.3 Types of outcome measures

Primary outcomes were measures of LVDF (see Supporting Information for a comprehensive list of individual LVDF measures) (Figure 1). Where both septal and lateral TDI measures were reported without their commonly reported mean, this was calculated using the recommended Cochrane method (Supporting Information).17 TDI measurements were sometimes reported without mention of the site of measurement. These were assessed separately and are identified as “cannot determine” measures. There were insufficient data for some measures of LVDF, such as pulmonary vein peak velocities and diastolic strain rate, to be included in the meta-analysis. To address this, a systematic review was completed to ensure that all measures of LVDF were summarized (Supporting Information).

4 SEARCH METHODS FOR IDENTIFICATION OF STUDIES 4.1 Electronic searches

Search terms were devised by one author and checked by another. Common terms and key words such as obesity, children and diastolic function were combined in search hedges (Supporting Information) and were applied in PubMed.gov (1958 to present), Cumulative Index to Nursing and Allied Health Literature (CINAHL; 1992 to present), Cochrane Central Register of Controlled Trials, and ClinicalTrials.gov (1997 to present), Embase (1974 to present), and Web of Science (1987 to present). The reference lists of included studies, as well as pertinent reviews,10, 11, 18 were also searched, yielding four further studies.19-22 The final search was completed on 11th July 2020.

5 DATA COLLECTION AND ANALYSIS 5.1 Selection of studies

Four authors independently reviewed results of the search to include/exclude studies for full-text screening. Inclusion and exclusion criteria for progression to the full-text screening are documented in the Supporting Information. One author completed a preliminary screen of all potential papers for full-text review to ensure that all included articles reported LVDF.

Two independent full-text screens were completed to include/exclude studies for the review. Criteria for inclusion/exclusion to the full-text screening are documented in the Supporting Information. When the same data were apparently reported in separate/duplicate publications, the article with the greatest number of subjects was selected and the other(s) excluded. However, if the article with fewer subjects reported additional LVDF measures, these measures were included as a separate study. Consensus on disagreements was achieved by discussion between reviewing authors or with the inclusion of a fifth author.

5.2 Data extraction and management

Data were extracted by one author using a pre-defined form and verified for completeness and correctness by two other authors. The following data were extracted: (1) study characteristics and methods; (2) subject/group demographics; (3) homeostatic model assessment of insulin resistance (HOMA-IR) results; (4) measures of LVDF and their results, and where applicable; (5) correlation statistics with adiposity and/or CMRF measures.

5.3 Assessment of risk of bias in included studies

Four authors independently executed quality assessment of the included studies and any discrepancies were resolved by discussion. Modified versions of the Study Quality Assessment Tools by the National Heart, Lung, and Blood Institute (NHLBI) were used to assess study quality and risk of bias.23 Scores of “good” (least risk of bias), “fair” (susceptible to some bias) and “poor” (significant risk of bias) were given to each study based on study design and implementation. Studies that were scored as “poor” overall but were otherwise methodologically sound (e.g. correctly reported LVDF measures and reported BMI, age, and sex) were included in the meta-analysis. A sensitivity analysis was completed to ensure that these studies did not influence the results. Furthermore, details of how these tools assess quality and risk of bias are given in the Supporting Information.

5.4 Data synthesis

Measures of LVDF were transformed into standard units of measurement where necessary. Mean ± standard deviation (SD) were calculated from alternative descriptions of central tendency and dispersion e.g. median, using the recommended Cochrane tools (Supporting Information).17, 24

5.5 Statistical analysis

Analysis was completed using STATA (version 16.1, StataCorp, College Station, TX). Although group data are reported in study descriptions, the marked heterogeneity in the mean BMI of control and OW/Ob groups across studies limited our ability to do a conventional, group-based meta-analysis reliably. To overcome this, mean (SD) BMI values for all groups, regardless of how those groups were defined by authors, were used to assess continuous associations of BMI with LVDF measures, using weighted, random-effects linear regression. These models were adjusted for age and sex to account for their known effect on BMI. These models also took account of the fact that some group means (e.g. a normal and an obesity group) were drawn from the same study. Each study was treated as a unique level in the random-effects regression, allowing the pairwise differences within studies to be captured by the model without reliance on specific group definitions. This enabled estimation of the linear relationships of BMI with multiple measures of LVDF and their relative strength, giving insight into which measurements may be most useful for early detection of impaired LVDF.

HOMA-IR values were similarly used to assess continuous associations with LVDF measures, using weighted, random-effects linear regression. These models were also adjusted for age and sex.

To account for individual study size and measure variance, each group estimate in the random-effects regression models was weighted using the inverse-variance method (1/standard error [SE]2).17 The SE of each measure was calculated using the SD and N reported for each group.

Histogram plots were used to assess normality of variables. Any non-normally distributed variables were transformed into normal distributions using the Tukey Ladder of Powers using the transformation with the smallest chi-squared value. Measures were further transformed to their z-scores (LVDFz, BMIz and HOMA-IRz), to allow correlation coefficients (r) to be calculated. Fisher's z-test was used to compare the strength of these correlations with the strongest association as a reference (Supporting Information). Robust z-scores, which do not depend on parametrically distributed data, were also calculated (Supporting Information) and the analyses were repeated to check that non-parametric distributions were not responsible for the findings. The brand of echocardiography machine was further included as a variable in repeat analyses to determine whether differences in technologies influenced the strength of relationship to LVDF measures. A sensitivity analysis was completed by repeating the analysis but excluding any studies reported as “poor”. A further sensitivity analysis was completed by excluding any study that included participants older than the American Academy of Paediatrics definition of adolescence (11–21 years).25

The r2 and adjusted r2 were reported for each model, and effect sizes, SE, 95% confidence intervals (CI), z-statistic, and P-value were reported for each independent variable in the models. P < 0.05 was considered statistically significant.

6 RESULTS 6.1 Study characteristics

Searches identified 7311 studies. After duplicate removal, 4365 were screened and 4254 were excluded, leaving 111 full-text articles to be assessed (Figure 2). A total of 70 studies (Table S1; sample size = 20–799; representing 9983 participants) were eligible, with 51 studies in the systematic review, 55 studies in the BMI meta-analysis (sample size = 20–650; representing 6782 participants), and 31 studies in the HOMA-IR meta-analysis (Figure 2 and Table S1; sample size = 20–650; representing 3878 participants). All included studies assessed LVDF by echocardiography.

Flow diagram of study identification, screening, eligibility and inclusion/exclusion. Echo indicates echocardiography; HOMA-IR, Homeostatic Model Assessment of Insulin Resistance; n, number of studies; OB, obese; OW, overweight. aExclusion criteria and reasons can be found in the Supporting Information

One study that assessed LVDF by CMR was identified but was excluded from the systematic review due to different ages of participants between groups. The number of studies and number of participants for each LVDF measure are reported in Table 1. Of these, 6 studies were scored as good, 48 as fair, and 16 as poor for quality and risk of bias (Table S1). Mean age, percentage of males, and mean BMI ranged from 8.9 to 18.4 years-of-age, 0–100%, and 15.8–60.0 kg/m2, respectively. There was marked heterogeneity in group-definitions, with >20 definitions identified for groups with OW/Ob groups and > 20 for control groups (Table S1). Furthermore, there was marked overlap of BMI between control and OW/Ob groups and marked dispersion of BMI within groups across studies, presenting a major challenge to the reliable use of conventional group-based meta-analysis (Figure 3).

TABLE 1. Total number of studies and participants available for meta-analysis with BMI for each LVDF measure Measure Total number of studies Number of studies in meta-analysis Total number of participants Number of participants in meta-analysis E wave 38 33 4056 3660 A wave 39 34 4200 3754 E/A 52 42 7795 5668 DT 19 14 2203 1383 IVRT 24 17 4190 1890 e′ 41 33 4464 3491 a′ 28 22 3308 2534 E/e′ 38 30 5048 4075 e′/a′ 23 18 3874 2782 Abbreviations: A wave, late mitral inflow peak velocity; a′, late diastolic tissue peak velocity; BMI, body mass index; DT, E wave deceleration time; E wave, early mitral inflow peak velocity; e′, early diastolic tissue peak velocity; E/A, E wave/A wave ratio; E/e′, E wave/e′ ratio; e′/a′, e′/a′ ratio; IVRT, isovolumic relaxation time.

Distribution of body mass index (BMI) in control (red) and overweight/obese (blue) groups included in the meta-analysis. Groups were defined as per the definitions in individual studies. A normal distribution curve was generated using the reported sample size (N), mean BMI, and BMI standard deviation. Significant overlap of BMI distributions between groups and marked variability of distributions within groups highlights that it was not possible to perform traditional group-based meta-analysis reliably

6.2 Objective 1—The association of OW/Ob with measures of LVDF

Objective 1 was to determine the association of OW/Ob with LVDF. This was done by meta-analysis and by systematic review. A small subset of papers addressed this question directly as a study outcome and the findings of these are summarized in Table S2.

6.2.1 Meta-analysis

The associations of BMI with measures of LVDF, after adjustment for age and sex, are given in Tables 2 and S3. There was evidence of reduced myocardial motion indicated primarily by strong associations of septal early diastolic tissue peak velocity/late diastolic tissue peak velocity (e′/a′) ratios and septal a′ peak velocities with BMI. For example, a typical child/adolescent with a BMI of 35 might have a septal e′/a′ ratio of 2.3 and a septal a′ peak velocity of 6.6 cm/s, compared to 2.0 and 5.5 cm/s, respectively, for controls with a BMI of 20. BMI was associated with all other measures of LVDF, apart from early mitral inflow peak velocity deceleration time (DT). Independent associations of LVDF with age are reported in Table S3. There were no independent effects of sex distribution in the studies.

TABLE 2. Associations of BMI with each left ventricular diastolic function measure, ranked by strength of association (r) Measure (units per 10 point change in BMI) Number of studies References Correlation coefficient (r) b 95% CI Fisher's z-test e′/a′ sep (1/kg/m2) 13 19, 22, 26-36 −0.689 −0.240 −0.299, −0.180 0.000 a′ sep ((cm/s)/kg/m2) 16 19, 22, 26-39 0.621 0.743 0.522, 0.965 0.239 e′/a′ lat (1/kg/m2) 12 19, 27-33, 35, 40-42 −0.593 −0.366 −0.525, −0.208 0.318 a′ lat ((cm/s)/kg/m2) 14 19, 27-33, 35, 38-42 0.432 0.877 0.558, 1.195 0.883 E/e′ sep (1/kg/m2) 16 19, 22, 26, 30, 32, 33, 36-39, 43-48 0.431 0.814 0.593, 1.035 0.902 e′ sep ((cm/s)/kg/m2) 19 19, 22, 26-39, 46, 48, 49 −0.413 −0.747 −1.057, −0.437 1.012 E/e′ average (1/kg/m2) 16 19, 30, 32, 33, 35, 38, 39, 43, 44, 46, 47, 50-55 0.387 0.666 0.552, 0.781 1.046 a′ average ((cm/s)/kg/m2) 14 19, 21, 27-33, 35, 38, 39, 50, 51 0.343 0.589 0.255, 0.924 1.176 e′/a′ average (1/kg/m2) 11 19, 27-33, 35, 54, 56 −0.306 −0.155 −0.262, −0.048 1.208 e′ average ((cm/s)/kg/m2) 20 19, 21, 27-33, 35, 38, 39, 46, 49-53, 55, 56 −0.294 −0.912 −1.302, −0.522 1.463 e′ lat ((cm/s)/kg/m2) 20 19, 27-33, 35, 38-42, 46, 49, 57-60 −0.247 −1.161 −1.571, −0.752 1.649 E/e′ lat (1/kg/m2) 18 19, 21, 29-33, 35, 38, 39, 42-44, 46, 47, 58,