Language abilities in preschool children with critical CHD: a systematic review

Abstract Context: Children with critical CHD are at risk for neurodevelopmental impairments, including delays in expressive and receptive language development. However, no study has synthesised the literature regarding language abilities in children with this condition. Objective: We summarised the literature regarding expressive and receptive language in preschool children with critical CHD. Data sources: MEDLINE, Embase, Scopus, Child Development and Adolescent Studies, ERIC, PsycINFO, and CINAHL. Study selection: We included studies published between January, 1990 and 1 July, 2021, focused on children aged ≤5 years with critical CHD requiring a complex cardiac procedure at age <1 year. Language ability was documented using standardised, validated tools assessing both expressive and receptive language outcomes. Data extraction: Data (study, patient and language characteristics, and results) were extracted by two reviewers. Results: Seventeen studies were included. Among children 2–5 years old with critical CHD, there were statistically significant deficits in overall (standardised mean difference: –0.46; 95 % confidence interval: –0.56, –0.35), expressive (standardised mean difference: –0.45;95 % confidence interval: –0.54, –0.37), and receptive (standardised mean difference: –0.32; 95 % confidence interval: –0.40, –0.23) language compared to normative data. Results reported as medians were similar to meta-analysis findings. Subgroup analysis showed that children with univentricular physiology had lower language scores than children with biventricular physiology. Conclusions: Preschool children with critical CHD had statistically significantly lower language outcomes compared to expected population norms. Healthcare professionals should test early and often for language deficits, referring to individually tailored supports.

of 2021, the same search strategy was repeated to identify any subsequently published studies. The search strategies were performed using the constructs of preschool children, critical CHD, and language outcomes to formulate the search, with adaptations to the search strategy according to each database. The search strategy is available from the authors upon request.

Inclusion criteria
This review included studies published in English from 1990-2021 that examined the receptive and expressive language outcomes of children aged 5 years or younger with critical CHD who required a complex cardiac procedure within the first year of life. Complex cardiac procedure was defined as having undergone surgery with cardiopulmonary bypass or catheter-based intervention. Studies had to involve direct assessment of a child's expressive and receptive language ability through standardised testing using a validated tool to be included in the review. Study designs included in the review were cross-sectional, case-control, cohort, as well as randomised controlled trial.
We excluded studies of children: (1) who did not require surgery or (2) who had their initial heart surgery after one year of age or (3) non-bypass surgeries. Studies that assessed language abilities using screening tools or parent-completed questionnaires were also excluded.
The protocol was registered and submitted to Prospero, 13 an international prospective register for systematic reviews (CRD42020192505).

Study selection
The study selection was completed through a two-step process. Two reviewers (Reviewer 1, Reviewer 2) independently screened titles and, where available, abstracts. The reviewers categorised each study as "include," "unsure," or "exclude." The full text of potentially relevant studies, the "include" or "unsure" categories, was obtained. The formal a priori inclusion criteria were independently applied to each potentially relevant study by Reviewer 1 and Reviewer 2 Discrepancies were resolved by a third reviewer (Reviewer 3).
The authors of articles were contacted if the expressive and receptive outcomes were assessed but the results were not reported. If the author was able to provide the required data, the study was included.

Data extraction
A standardised form to facilitate data extraction was developed based on the Cochrane Handbook for Systematic Reviews of Interventions, Systematic Reviews: CRD's guidance for undertaking reviews in health care, and clinical acumen by healthcare professionals and researchers. 14,15 General and demographic information extracted included article title, author names, date of publication, country of study, study design, single or multicentre study, sample size, population age, population sex, and cardiac diagnoses. Extraction of perioperative variables included procedures performed, number of cardiac surgeries under cardiopulmonary bypass, length of hospital stay, and comorbidities. Language data included age at language assessment, language tool used, and language outcome results for children with critical CHD and control data, if available. Data extraction was first performed independently by two reviewers (Reviewer 1, Reviewer 4) and then reviewed together to resolve any discrepancies.

Quality assessment
Quality assessment was completed using the Revised Cochrane Risk-of-Bias tool for randomised trials 16 and Risk Of Bias In Non-randomised Studies of Interventions 17 assessment tool and template, as per the study design. The Revised Cochrane Riskof-Bias tool for randomised trials tool assesses randomised studies through 5 domains for potential bias: randomisation process, deviations from intended interventions, missing outcome data, measurements of the outcome, and the selection of reported results. The Revised Cochrane Risk-of-Bias tool for randomised trials tool then classifies the randomised studies as low, some concerns, or high risk of bias. 16 The Risk Of Bias In Non-randomised Studies of Interventions tool assesses non-randomised studies through 7 different domains of potential bias: confounding, selection of study participants, classification of interventions, deviations from intended interventions, missing data, measurement of outcome, and the selection of reported results. The Risk Of Bias In Nonrandomised Studies of Interventions tool then classifies each non-randomised study as low, moderate, serious, or critical risk of bias. 18 Two reviewers (Reviewer 1 and Reviewer 2) independently assessed each article. The reviewers first pilot-tested 3 articles to ensure they operationalised each domain similarly based on the detailed guide and tool provided. Any disagreements were resolved by discussion or a third reviewer (Reviewer 3).

Statistical analyses
Review Manager 19 software (version 5.4) was used to pool the study results into a standardised mean difference for overall, expressive, and receptive language outcomes when individual study results provided the mean and standard deviation, and it was statistically and clinically appropriate. Both fixed and random effects meta-analyses were performed. A standardised mean difference pooled result was calculated and displayed as a random effects model with a 95% confidence interval since different language outcome tools were used by different studies. Four studies 24,31,35,37 reported results using the median, interquartile range, or range as described in Table 1. If studies reported critical CHD subgroups without a score for the entire critical CHD cohort, such as children with normal hearing as compared to those with hearing loss, 20 a combined summary statistic was calculated through the formulae provided by the Cochrane Handbook for Systematic Reviews of Interventions. 14,21 For studies that compared univentricular to biventricular critical CHD, a subgroup analysis compared language scores. Statistical heterogeneity between studies was measured using the I2 statistic as suggested by the Cochrane Collaboration, in which values of 0-40% may be considered unimportant, 30-60% as moderate, 50-90% as representing substantial heterogeneity, and 75-100% as considerable heterogeneity (overlapping proportions are intentional). 14,21 If the fixed and random effects results were similar, the random effects models were reported. 14 Publication bias was assessed through visual interpretation of funnel plot symmetry and formally with the Egger test 22 using STATA software 23 where p < 0.05 indicated likely publication bias.

Methodological quality of included studies
Of the 15 observational studies using the Risk Of Bias In Nonrandomised Studies of Interventions tool, 7 (47%) 20,24-29 studies were at serious risk of bias, and eight (53%) 30-37 had a critical risk of bias. Sources of bias included confounding, the selection of study participants, and the measurement of outcome due to lack of blinding. Of the 2 randomised studies that were assessed using the Revised Cochrane Risk-of-Bias tool for randomised trials tool, one study 38 had some concerns and the other 39 was rated as high risk of bias. Sources of bias included the randomisation process and the selection of the reported results. Specific details are provided in Table 2.
Four studies (31%) 32,33,36,38 reported an expressive language score statistically significantly lower than the normative population data (but still within one standard deviation of the population mean).
Nine studies 20,[25][26][27]29,31,33,36,37 reported an overall language score: 2 (22%) 29,36 were statistically significantly lower than the normative mean (but still within one standard deviation of the general population mean).  Figure 2. Forest plot and standard mean difference for expressive (2a), receptive (2b) and overall (2c) language scores of children 2-5y with critical congenital heart disease The standard mean difference for each study is represented by a square with confidence interval bars. The size of the box indicates the relative weight of the study, The total metaanalysis result is represented by the diamond. Negative values indicate lower scores for the critical congenital heart disease.

Subgroup analysis based on cardiac physiology
Only 4 studies (24%) 24,27,29,36 described language outcomes for children with univentricular as compared to biventricular cardiac physiologies. The I 2 statistic assessing statistical heterogeneity within the 2 studies was substantial (I 2 = 64%) for expressive language scores, precluding a formal meta-analysis. 14 Two studies reported significantly lower overall language scores for children with univentricular critical CHD as compared to biventricular physiology. 29,30,36 One study 24 found expressive language values to be significantly lower for children with univentricular versus biventricular physiology. Two studies 24,36 reported significantly lower receptive language values for univentricular physiology as compared to biventricular physiology.

Language outcomes at 12 months
The I 2 statistic assessing statistical heterogeneity for all language outcomes was significant, therefore, precluding a meta-analysis. However, all 6 studies 25,28,33,35,37,39 reported overall, expressive, and receptive language scores on the Bayley-III language tool within one standard deviation of the normative mean. One study 33 reported statistically significantly lower expressive and receptive language scores for children with critical CHD when compared to the general population mean. Overall language was not determined to be statistically significantly different in any study. 25,28,33,37,39 Language outcomes at 6 months The I 2 statistic assessing statistical heterogeneity for all language outcomes was significant; therefore, a meta-analysis was not performed. Both studies 25,34 reported expressive and receptive language scores below the mean. Brosig Soto et al (2011) 34 reported a language outcome significantly lower than the normative population mean at 96.8 (SD: 12.7, p = .005), while Noeder et al (2017) 25 reported a language score greater than one standard deviation below the population mean at 84.1 (SD: 15.0).

Publication bias
Publication bias could only be assessed for language outcomes for children 2 to 5 years of age. Funnel plots showed little evidence of publication bias for overall (p = 0.14), expressive (p = 0.89), and receptive (p = 0.62) language outcomes.

Discussion
Findings of this systematic review and meta-analysis indicate that preschool children with critical CHD have statistically significantly lower expressive, receptive, and overall language abilities when compared to their peers and that they struggle more with expressive than with receptive language skills. Additionally, although statistical heterogeneity precluded determining a pooled overall effect to quantify the difference, children with univentricular physiology appear to have higher rates of language delay than children with biventricular physiology. The importance of studying language outcomes relies on the significant role language plays in a child's development; language is essential for communication and is a key component of academic functioning. Even modest deficits in language abilities are known to significantly impact a child's day-to-day function, communication, and to negatively influence social interaction. 48 The results of this study are consistent with previous reports and recommendations. The Cardiac Neurodevelopmental Outcome Collaborative indicates expressive language delays are a common concern for children with critical CHD and should be monitored. 49 The guidelines developed by American Heart Association also highlight the need to assess language development and to refer to speech-language pathology when language deficits are identified. 49,50 Findings that children with univentricular critical CHD struggle more in certain areas of neurodevelopment than children with biventricular critical CHD have been previously reported and determined to be statistically significant . [51][52][53][54] This review found similar findings to those of studies looking at the longer-term language outcomes of school-aged children with critical CHD. In an article by Bellinger et al (2003), 55 school-aged children within the critical CHD cohort were found to have significantly lower scores than the expected mean. Similarly, in an article by Mahle et al (2000), 56 children with hypoplastic left heart syndrome at school age had expressive and receptive skills that were statistically significant and below the norm. In a study by Hövels-Gürich (2006), 57 expressive and receptive language values in children at school age were found to be within one standard deviation of the normative mean within the cohort studied and were statistically significant findings as compared to the normative population mean. In a second study by Hövels-Gürich (2008), 58 schoolaged children were more impaired in expressive language testing than overall or receptive language testing. Such studies then highlight the importance of long-term follow-up as children with critical CHD appear to continue to be at high risk for language delays (and particularly expressive language delays). A lack of significant language difference found at 12 months compared to those differences found at 2-5 years suggests a need for continuous follow-up that is supported by the literature. 50 The results of this language-focused systematic review are also consistent with reviews of studies in motor and cognitive neurodevelopmental delay in children with critical CHD. 7-9 Reviews of both motor and cognitive abilities of preschool children with critical CHD have shown that children with this condition score significantly lower than their non-critical CHD peers. [7][8][9] Proposed explanations of such delays in multiple developmental domains include chronic brain hypoxia, increased incidences of pre-and post-natal brain injury, brain immaturity, and other clinical and environmental factors. [59][60][61][62][63] Importantly, some of the literature suggests the Bayley-III overestimates language ability for both healthy developing and children with critical CHD; 27,64-66 which could mean that children with critical CHD have even worse language skills than this systematic review and meta-analysis reports. Anderson et al (2017) 65 found an increase in language scores on the most recent edition of the Bayley-III compared to the Bayley-II. Moreover, Goldstone et al (2020) 66 found this increase to be significant in children with critical CHD. Notably, our review included publications that used other language assessment tools and determined that those findings were consistent with the results of studies that utilised the Bayley-III. This consistency suggests that the language abilities of preschool children with critical CHD are typically below average.

Limitations
There are several limitations to our systematic review and metaanalysis. First, the methodological quality of the included studies was rated quite poorly. This rating is unsurprising given that the
detailed guide of the Risk Of Bias In Non-randomised Studies of Interventions tool indicates that " : : : it will be rare that an NRSI [non-randomised studies of the effects of interventions] is judged as at low risk of bias due to confounding, we anticipate that most NRSI will be judged as at least at moderate overall risk of bias." 18 Often, the overall risk of bias was rated as serious or critical and was typically due to the first domain regarding confounders, missing statements of possible confounders, selection of participants, and lack of blinding. However, the results of the two randomised controlled trials that focused on 2 to 5 year-old-children, which should have a balance of known and unknown confounders between the 2 groups, were similar to the non-randomised controlled trial results. The expressive and receptive scores led to the consistent conclusion that children with critical CHD at 2-5 years of age are below the normative means. While many studies were classified at critical risk, basing the use of articles solely on the quality assessment of studies that are not amenable to randomised control trials may prevent the inclusion of critical results in many different areas of research. 67 Comparing outcomes to normative population data without further adjustments may introduce bias into the results as any difference found between children with critical CHD and the normative population may seem to be causal, when in reality differences such as socioeconomic status, support interventions, or other differences may have large effects in children with critical CHD that may not be accounted for.
Finally, only two studies 27,36 reported language outcome data comparing univentricular and biventricular cardiac physiologies. Although heterogeneity precluded pooling the individual results, reports of lower language scores and other neurodevelopmental domains support the findings of our review. 51,52,54 Likewise, although the presence of a genetic anomaly among children with critical CHD is known to impact developmental outcomes, this review was unable to examine language outcomes based on the presence or absence of genetic anomalies as the included studies did not consistently stratify their results by this variable. Future research to determine the exact clinical and statistical significance of differences in language outcomes between these two cardiac physiologies and the stratification of language abilities in children with additional genetic anomalies is recommended.

Conclusion
This systematic review and meta-analysis is the first to review and assess the results from the literature on overall, expressive, and receptive language abilities in preschool children with critical CHD. The findings indicate that preschool children with critical CHD have significantly lower language abilities when compared to the general population and may be more affected in the expressive language domain than in their receptive language skills. Future research should focus on determining language outcomes among older children with critical CHD as well as on testing interventions to improve language skills in this population.