Characterization of Language Abilities and Semantic Networks in Very Preterm Children at School-age

doi:10.21203/rs.3.rs-4130846/v1

It has been widely assessed that very preterm children (< 32 weeks gestational age) present language and memory impairments compared to full-term children. However, differences in their underlying semantic memory structure have not been studied yet. Nevertheless, the way concepts are learned and organized across development relates to children’s capacities in retrieving and using information later. Therefore, the semantic memory organization could underlie several cognitive deficits existing in very preterm children. Computational mathematical models offer the possibility to characterize semantic networks through three coefficients; average shortest path length (i.e., distance between concepts), clustering (i.e., local interconnectivity), and modularity (i.e., vocabulary enrichment). Here we assessed these coefficients in 38 very preterm schoolchildren (aged 8–10 years) compared to 38 full-term schoolchildren (aged 7–10 years) based on a verbal fluency task. Using semantic network analysis, very preterm children showed a lower interconnectivity at a local level than full-term children. However, we found no differences between very preterm and full-term children regarding their average shortest path length between concepts and their modularity at a global level. These findings provide preliminary evidence that very preterm children demonstrate subtle impairments in the organization of their semantic network, encouraging the adaptation of the support and education they receive.

Biological sciences/Neuroscience/Cognitive neuroscience

Biological sciences/Neuroscience/Learning and memory

Biological sciences/Psychology/Human behaviour

very preterm birth

semantic network

language

Very preterm births (i.e., before 32 gestational weeks) are known to cause various adverse neurodevelopmental outcomes ^1,2. Between 3–5 years, very preterm children present more difficulties in multiple cognitive domains compared to full-term children ^3,4. Among these domains, language is often reported as a problematic ⁵, with long-lasting impairments commonly reported in preterm children (for reviews see for example Vandormael et al. ⁶ or van Noort-van der Spek et al. ⁷). With the exception of those without severe neonatal medical complications ⁸, language impairments seem to affect many subdomains. While early childhood delays are caught up in low-level language abilities (i.e., phonological processing and sight word reading) by early adolescence, high-level abilities (i.e., syntax, semantics, phonemic decoding, and verbal language memory) remain more often impaired in very preterm adolescents ^9,10. Consequently, they need about eight times more academic support in writing and reading compared to their full-term peers ⁵. One aspect of these long-lasting impairments concerns their vocabulary. Where Kunnari et al. ¹¹ found no differences in expressive vocabulary for children aged 2, Luu et al. ⁵ showed poorer receptive and expressive vocabulary in preterm children compared to full-term controls aged 8 and 12. However, later at the age of 16, they found that the deficit in receptive vocabulary was no longer present ⁹. Moreover, while 3-year-old very preterm children uttered the same number of questions and negations, and had comparable lexical variability (i.e., number of nouns, verbs, adjectives, prepositions) compared to full -term children, they produced less varied sentence structure with, in addition, shorter and fewer words as well as fewer words roots ¹². On the other hand, preterm children also exhibit impaired reading skills and comprehension as well as delayed grammatical development (e.g., Guarini et al. ¹³). These skills rely in particular on the word comprehension ⁶. Altogether these quantitative measures show that preterm children have poorer receptive and expressive language, semantics, and receptive grammar, reflecting a lower overall comprehension capacity (e.g., sentences, words, structure, verbal information) ¹⁴.

Moreover, most studies report that these impairments also affect verbal memory. At 3.5 years old, preterm children already present poorer phonological working memory (i.e., memory of words and non-words) than full-term children. Omizzolo et al. ¹⁵ discovered that despite preterm children having a good long-delay verbal recall, their short-delay verbal recall remains impaired at 7 years compared to full-term controls. This persists during adolescence, with 9-16-year-old preterm children showing weaker linguistic working memory, based on sentence recall and the establishment of semantic relationships than full-term of the same age ¹⁶.

Whereas language abilities and verbal memory have been well studied in very preterm children, the underlying complexity of semantic organization has not been explored yet. The way concepts are learned and progressively organized in memory predicts various higher-order abilities ¹⁷. Consequently, language difficulties in very preterm schoolchildren may mirror the underlying semantic organization. Indeed, children reinforce and expand their knowledge in their early years. During their first 7 years, they acquire on average 800 root words per year and then learn 1,000 new words per year until the age of 12 ¹⁸. Vocabulary and concepts are stored in the semantic memory network ¹⁹. Alongside this acquisition, children also weave links between these words and gain a deeper understanding and meaning of concepts, allowing them to use abstract and conditional reasoning ²⁰. While gaining knowledge and expertise in a particular domain, the semantic structure becomes more cohesive and associative, thus they have better access to specific knowledge ¹⁷. Semantic memory structure is considered to be built and influenced by experience ²¹, but there is a gap in knowledge concerning children who have early developmental adversity.

Recently, some studies have begun to focus on semantics in the form of networks using methods based on mathematical models and computational research (e.g., Siew et al. ²²). Concepts are represented by nodes (i.e., basic units of the network) which are interconnected (i.e., represented by the edges) ²³. These associations are based on different similarities, creating a network. This latter is structured by three mechanisms, the modularity (i.e., Q, subnetworks creation), the average shortest path length (i.e., ASPL, moving from one node to others), and the clustering (i.e., CC, closeness, or isolation of the nodes). These parameters influence the way we perceive and retrieve concepts. For example, the more distant the concepts are, the more difficult it is to make associations between them ²².

By investigating semantic networks in very preterm children, we aimed to examine if their verbal difficulties could relate to a problem with the organization of the semantic memory and the encoding and storage of concepts. One common and simple way to investigate semantic networks is the verbal fluency task where people are asked to give as many words as possible in a specified category (e.g., animal, food, words beginning with a given letter) in a fixed time (usually 60 or 90 seconds). Given the language and memory difficulties present in the population of very preterm children, we expected to find differences in their semantic memory organization compared to full-term children. More specifically and based on the existing literature, we hypothesized (1) higher ASPL, (2) lower CC and (3) higher Q. Differences in their semantic memory structure may point to differences in learning and encoding. This could help to adapt current teaching and specialized educational support, as some pedagogical practices are known to foster a more flexible semantic network ²⁴.

Ethics

The present study is part of a larger project titled Long-term impact of early nutritional and pain management in very preterm infants on brain health and function and approved by the University ethical committee (CER-VD). All methods were carried out in accordance with the STROBE 2007 guidelines and with the principle of the Declaration of Helsinki. Each parent signed informed consent for their child, and oral consent from each child was obtained.

Design of the study

The present study was nested in an observational cohort study of very preterm children who were recruited from birth at the Clinic of Neonatology of the University Hospital of Lausanne between 2011 and 2013. They were iteratively followed up at the Development Unit of the same hospital. Further details on this cohort study have been previously described ²⁵.

For comparison, we used a reference group of full-term children of the same age. They were recruited by one of the authors in the French-speaking side of Switzerland between 2018 and 2021 as part of a parallel study about schooling and neurocognitive development ²⁴.

Participants

Very preterm children

The initial cohort comprised 51 very preterm neonates. Throughout the study, 2 (3.9%) patients died during the follow-up, 2 (3.9%) were excluded due to complex social situations, 2 (3.9%) were lost at previous appointments, 2 (3.9%) were unreachable and 2 (3.9%) refused to participate. Finally, 3 (5.9%) children were not able to perform the verbal fluency task (for details see Supplementary Materials).

Therefore, in total, 38 (75% of the initial cohort) very preterm children born between 25 and 31 gestational weeks and aged 8–10 years took part in the present study. They were tested separately in quiet rooms during their follow-up appointment at the Developmental Unit between October 2019 and October 2021.

Full-term children: reference group

Thirty-eight full-term children aged 7–10 years were selected based on age and gender. They performed a verbal fluency task about animals alongside different tests during their participation in the parallel study ²⁴ and were tested at the same hospital or at their school.

Material

Verbal Fluency

This task is a subtest of the NEPSY-II ²⁶, a neuropsychological battery where children must generate words. It is divided into two sub-categories, the Semantic Word Generation score and the Initial Letter Word Generation score. In the Semantic Word Generation, children had sixty seconds to name all the words belonging to the category (i.e., animals and foods/drinks) they could think of. Then, in the Initial Letter Word Generation, they were given a letter (i.e., F and S) and had to produce all the words starting with this letter they could in sixty seconds. Words were immediately transcribed. The total number of correctly generated words was used. This subtest also comprises a comparison score calculated based on the scores of the two subcategories. It is aimed to compare fundamental processes (i.e., semantic) and more complex ones (i.e., initial letter).

Based on previous research ^24,27, we selected the animal category to compute the semantic networks and compare the two groups.

Socioeconomic status (SES)

Regarding the very preterm children, SES was assessed through the Largo ²⁸. It is composed of two scores, the mother’s education (or current work if applicable), and the father’s current job, both scoring from 1 to 6 (1 representing higher SES). Full-term children’s SES was determined by both parent’s educational level and current job (both scoring from 1 to 4, 1 representing lower SES). These scores were summed and averaged between both parents. To compute the comparison between both groups, the very preterm score was reversed, and all scores were set as a percentage.

Non-verbal intelligence

The Matrix Reasoning subtest from the Wechsler Intelligence Scale for Children 5th Edition (WISC V) ²⁹ was used to measure the intelligence of very preterm children. They were asked to choose the missing piece that completed the sequence of coloured visual pattern matrices. Full-term children filled the black and white Raven's Progressive Matrices³⁰. They were also asked to select the missing piece of black and white visual pattern matrices. The two scores were converted in success rate.

Data Analyses

Group Comparison

The socioeconomic background and intelligence having a significant impact on verbal abilities, both variables were tested between groups using t-tests to assess group comparability and then used as covariates in the semantic network analysis.

To ensure the validity of the use of the Matrix Reasoning subtest success rate for very preterm children, Pearson’s correlation was made between success rates and the Full Scale IQ T-scores from the WISC-V. We found strong correlations between both scores (r (36) = 0.729, p < .001) allowing us to use the success rate as a measure of non-verbal intelligence. We kept the Matrix Reasoning subtest to be as close as possible to the Matrix task done by the full-term children.

Verbal Fluency

All three scores (i.e., Semantic Word Generation, Initial Letter Word Generation, Comparison scores) were expressed as standard scores ranging from 1 to 19. Comparison score is calculated based on Semantic Word Generation and Initial Letter Word Generation standard scores. Lower scores are representative of a profile with good capacity for vocabulary production and less effective strategies for retrieving non-categorical information, while the highest scores are the opposite.

Semantic network analysis

The semantic network is decomposed in nodes and edges. Each node represents a category exemplar (i.e., in the present study an animal). Edges are the associations between two category exemplars. They represent the capacity of the groups to generate the second exemplar when they have produced the first one (e.g., generate both lion and tiger).

The fluency data were preprocessed and the semantic network components were measured using the public-available SemNA pipeline in R ³¹ using the R packages SemNetCleaner, SemNetDictionaries, and SemNet³¹, NetworkToolbox ³², readxl ³³, RDS ³⁴,

Network estimation. The data were first pre-processed and cleaned with the SemNetDictionaries ³⁵ and SemNetCleaner ³⁶ tools. Repeated answers, non-categorical (i.e., others than animals), and fictional animals were excluded. Misspelled words and root variations (e.g., plural or adults and their offspring) were corrected and changed to their root (respectively, e.g., cats → cat). A binary response matrix was constructed with these cleaned responses. Its columns represent all the different unique exemplars given by the sample and its rows represent participants. This matrix is filled out by 1 (if an exemplar was generated by that participant) and 0 (if that exemplar was not).

Independent t-tests and chi-squared were performed on average number of these cleaned responses and on the total unique cleaned responses (i.e., all different words generated by a group).

Then, we used a bootstrapping approach to simulate and compare semantic networks of both groups. We applied the case-wise bootstrap method, meaning that for each group, we resampled with replacement from the participants in their respective groups. Each bootstrap sample has as many participants as the original sample but with some participants being included more than once and others not being included at all. We resampled the same participants' covariate measures (i.e., SES and non-verbal intelligence) with each bootstrap sample. The bootstrap handled finalizing (i.e., we used only responses given by at least 2 participants) and equating (i.e., networks of both groups in each sample are compared using the same nodes ³⁷) with each replicate sample. We used the similarity to avoid negative associations between nodes in the network (i.e., all values are positive ranging from 0 (two responses do not co-occur) to 1 (two responses always co-occur)). In addition, we applied the Triangulated Maximally Filtered Graph (TMFG; ³⁸) as a filtering method on the word association matrix. It minimizes the noise and potential spurious associations by constructing a sub-network to capture only the most relevant information (i.e., removal of spurious associations and keeping the highest correlations) within the original network. The TMFG method keeps the number of edges the same between groups, therefore newly created networks will be comparable as having the same number of nodes and edges. This eliminates the risk of confounding different network structures being due to a different number of edges ^39,40.

Networks are thus estimated and their measures (i.e., ASPL, CC, Q) are computed. This is repeated iteratively for 1,000 times.

Network analyses. We computed the 3 network coefficients for both groups (see Fig. 1 for visualization). The Average Shortest Path Length (ASPL) refers to the average shortest number of steps (i.e., edges) we need to get between any pair of nodes in the network. It measures the global organization. While higher ASPL means that the semantic network is more spread out (i.e., less interconnected) and people are less likely to make associations due to greater distances between associations ⁴¹. Lower ASPL represents a more interconnected network with shorter distances between associations. Thus, lower ASPL increases the possibility of reaching more distant associations faster, like in creativity ^22,42. The Clustering Coefficient (CC) refers to the extent that two neighbours of a node will be neighbours themselves. It illustrates the manner in which the semantic network is organized at a local level ³¹. The higher this coefficient is, the more the nodes that are near neighbours to each other will tend to also be connected and the network will be more interconnected ²². The third coefficient, the Modularity Coefficient (Q) represents the extent to which the network can be compartmentalized into small sub-networks. A sub-network can represent an animal category (e.g., sea animals, savannah animals, farm animals, pets). The higher the coefficient, the more connections there are within a sub-network and the fewer there are with nodes in other subnetworks making sub-networks more distinct from each other. It illustrates vocabulary enrichment.

Thanks to a specially adapted version of the code (i.e., adapted version of the pipeline from Christensen & Kenett³¹, we then applied an ANCOVA on the three networks coefficients (i.e., ASPL, CC, Q) with SES and non-verbal intelligence as covariates to examine the difference across both groups.

Bootstrapping small samples can over-representing outliers. Given our groups are relatively small, a lot of variation in the results could occur because of these cases in our analyses. Therefore, we performed the bootstrapping analysis twenty times and reported the proportion of times ASPL, CC, and Q are significant to show consistency in the results and that the results are not due to a one-off chance.

As the visualization of the semantic network cannot be computed with the addition of the covariates, and cannot be pooled over the 20 tests, bar plots of the coefficients average of the 20 tests were produced to compare the two groups.

1. Demographics of Very Preterm Children

The very preterm children from our cohort were born between 25 and 31 gestational weeks (M = 27.1, SD = 1.48). their birth weight was between 517 and 1590 grams (M = 908, SD = 248). Comparing to the general population (M = 100, SD = 15), their general IQ assessed with the WISC was in the norms (M = 103.61, SD = 11.31).

2. Group Comparison

Very preterm and full-term children demographics were assessed to establish their comparability (see Table 1for details).

Table 1

Demographic and non-verbal Intelligence comparison between very preterm and full-term groups.
	Group
	Very preterm children	Full-term children	t-test or ꭓ²	p value	Cohen’s d or φ
N (% girls)	38 (52.63%)	38 (52.63%)	0	1	0
Age in years	8.78 (0.458)	8.82 (0.715)	0.32	0.75	/
Socioeconomic status	58.6 (25.1)	59 (18.1)	0.07	0.94	0.02
Non-verbal Intelligence	51.8 (11.1)	88.7 (9.82)	73	< .001	/

3. Verbal Fluency of the Very Preterm Children

Table 2

Results of the Word Generation subtest in very preterm children. *Note*. The scores were expressed as standard scores.
	Very preterm Children		Norms
	Very preterm Children	min, max	Norms
Semantic	10.4 (2.54)	5–15	10 (3)
Initial Letter	7.22 (1.95)	4–11	10 (3)
Comparison	6.95 (2.24)	3–11	/

Very preterm children performed in the norms for the semantic score with a potential deficit for the initial letter score (see Table 2 for details). The lower comparison score (M = 6.95, SD = 2.24) indicated a good capacity for vocabulary production but less effective strategies for retrieving non-categorical information.

4. Semantic network analysis

4.1 Words Generation

Overall, very preterm children showed lower quantitative semantic abilities compared to full-term children (see Table 3for details).

Table 3

Word generation scores separated by group.
	Group
	Very Preterm Children	Full-term Children	Welch’s t-test or ꭓ²	p value	Cohen’s d or φ
Mean total number of responses (Mean, SD)	13.05 (3.48)	17.92 (5.16)	4.82	< .001	/
Unique responses (N)	136/242	202/242	37.7	< .001	0.39
Responses produced only by one child (N)	57	96	/	/	/
Responses not provided by the other group (N)	40	106	/	/	/

4.2 Network coefficients

Twenty case-wise bootstrap network analyses demonstrated that despite small samples, very preterm children (M_CC = 0.693) had a lower Clustering Coefficient than full-term children (M_CC = 0.705), 16 times out of 20. But despite very preterm children (M_ASPL = 3.35) had longer Average Shortest Path Length than full-term children (M_ASPL = 3.14), it was not significant (7 tests out of 20). Likewise, regarding the Modularity Coefficient, there was no significant difference (4 times out of 20) between very preterm children (M_Q = 0.602) and full-term children (M_Q = 0.591) (for the ANCOVA results of the twenty case-wise bootstrap analyses, see the Supplementary Material and see Fig. 2 for a visual representation).

Schoolchildren born very preterm are known to present language weaknesses, measured through quantitative tools. However, qualitative aspects of language are less investigated. In the present study, we examined the organization of concepts in verbal memory of very preterm schoolchildren. Using a verbal fluency task, we depicted their semantic network (i.e., the encoding and storage of concepts) and compared it with the semantic network of schoolchildren born at term, based on regular network metrics. Overall, very preterm children shared similar ASPL (i.e., distance between concepts) and Q (i.e., compartmentalization into small sub-networks) than full-term children. However, lower CC (i.e., local interconnectivity) was observed. Here, we discuss this finding through the lens of known processing speed and language use in very preterm children and how pedagogical practices could be seen as a preventive action specifically addressing this issue.

Very preterm schoolchildren are known to have language impairments, impacting semantics, expressive vocabulary, verbal language memory, and grammar^10,13. While 12 very preterm children of our cohort were undergoing speech therapy, they overall performed in the norms at the Word Generation subtest. However, they presented subtle difficulties in retrieving information when not organized into categories, with a consequently low comparison score. It may reflect either weaknesses in search strategies due to poorer working memory and inhibition capacity (i.e., executive functions) or a deficit in orthographic fluency or sound awareness. In other words, they have encoded and stored semantic representations in long-term memory and have access to them, but they have not stored orthographic/sound representations or have difficulty retrieving them. However, in terms of categories, very preterm children generated fewer animal names, less unique ones, and poorer diversity compared to the full-term children in the present study. Consequently, if we consider only quantitative measures, we might think that very premature children have vocabulary difficulties and therefore suggest educational support based on learning additional vocabulary. Yet, the underlying difficulties might lie elsewhere. While their global semantic network (ASPL) and vocabulary enrichment (Q) were similar to those of full-term schoolchildren in the present study, the semantic network of very preterm children was less interconnected (CC) than observed in full-term children. Given that less embedded concepts will be harder to be retrieved and used concomitantly with other concepts ¹⁷, this qualitative perspective provides new understandings on their linguistic and cognitive difficulties, namely the way they organize and connect information. With a less organized network, it is more difficult to retrieve information. We can observe this difficulty in young childhood, at 18 months, as very preterm infants take more time to respond to spoken demands ⁴³. These authors linked this slowness to the amount of vocabulary the children had independently of their prematurity status. However, it can be hypothesized that this could also be related to the semantic network organization and the difficulty for children to access information. Across development, this may have a cascading effect on learning processes ⁴⁴. In fact, very preterm children have a lower processing speed in various tasks (e.g., Brydges et al.⁴⁵). This could reflect an inadequate semantic network organization with the difficulty to relate the information received to known concepts, as well as to retrieve and select concept to respond. Furthermore, while very preterm children show memory impairments in various domains (for a review see Anderson ⁴⁶), there is evidence that a less interconnected word will be less easily recalled during a memory cued recall task ⁴⁷. Therefore, their less organized network could be partly responsible for their neurodevelopmental deficits.

Very preterm children are known to experience multisensory difficulties ^48,49, that are responsible for higher-level cognitive impairments (for a review see Wallace et al. ⁵⁰) including language and memory ⁵¹. Therefore, their inability to organise new concepts optimally could be explained by multisensory difficulties during the learning process. In view of this vulnerability, it might be appropriate to integrate interventions promoting multi-sensoriality into the learning process. Pedagogies using didactic materials engaging multiple senses, in active-based form, such as the Montessori pedagogy ⁵², were found to positively impact semantic networks’ structure. Indeed, Montessori schoolchildren compared with traditionally-schooled children, have a higher CC and lower ASPL and Q ²⁴. Therefore, it might be relevant to see pedagogy as a preventive measure for very preterm children. Providing access to educational settings and learning materials that reinforce and train complementary sensory channels will foster qualitative language abilities. Focusing more on meaning and integrating new words while engaging multiple senses rather than teaching isolated concepts, might allow them to build a more interconnected and organized network.

Some limitations need to be acknowledged. First, the experimental settings for the very preterm and full-term children were not similar. For the former, data were collected in the framework of a routine clinical assessment in the context of a longitudinal study, while for the latter, data were collected in the framework of a voluntary-based experiment. Consequently, contexts may not be similarly stressful. Second, bootstrapping in small sample sizes being severely affected by outliers, given our modest sample-size, it was not robust. Nevertheless, our results showed some consistency suggesting that although both groups were small, there did not appear to be issues related to outliers. Yet, it would be relevant to reproduce this study with a bigger cohort to confirm these results. Third, we only evaluated animal-related semantic network. It could be interesting to repeat this task with other semantic categories, such as food or hobbies. In fact, due to its advantages (e.g., generational, cultural, linguistically independent), the semantic category of animals is the most widely used ⁵³. But depending on the child's interest in the category, the organisation of the network may change. Fourth, we acknowledge the non-verbal intelligence measure was not the same for both groups, nevertheless, we took the most similar measure and attested to its validity for the very preterm children. Finally, we only have a cross-sectional perspective. As longitudinal studies show that the clustering coefficient diminishes with age in healthy population ⁵⁴, it would be interesting to see if it is also the case with very preterm cohorts. In addition, given that very preterm children catch up with some of their weaknesses in adolescence ^9,16, we could investigate whether their semantic networks also normalize or follow the same pattern as those of full-term children while keeping these differences.

In conclusion, this study is one of the first to highlight qualitative aspects of language organization of very preterm children compared to full-term children, revealing subtle impairments. In fact, their semantic network appeared to be less connected, probably related to their processing speed and language difficulties. Given the importance of semantic network structure in higher cognitive functions, these findings encourage larger, longitudinal studies in this field. This will lead to improved support and teaching practices for very preterm children.

Acknowledgments

We would like to thank all the motivated children and parents who took part in the main longitudinal study and who have agreed to take part in this final stage, thus contributing to the realization of various studies. We also thank Dr. Jérôme Rime who has helped us with the interpretation of the NEPSY results.

Author Contributions

SD, JS, and MD conceived of this specific study and its research question. MD performed the statistical analyses and wrote the original draft. APC coded the adapted version of the SemNA pipeline and revised the Method section. SD collected the Verbal Fluency task of the full-term children and revised the manuscript. CHV and LB collected and scored the Verbal Fluency task of the very preterm children and revised the manuscript. JS and CF designed the main longitudinal study, were responsible for funding acquisition, and revised the manuscript. All authors read and approved the final manuscript.

Competing Interests

The authors declare that they have no known competing financial interests or personal relationships that are relevant to the content of this article.

Funding

This work was supported by the Fondation W. et E. Grand d’Hauteville in Lausanne (Switzerland).

Data Availability

The anonymised experimental data collected, the specific xlsx files for analysis, and the analytic code written in R are publicly available on the Open Science Framework repository (https://osf.io/3qb95/?view_only=2f8013862ae94770a78cde3605e9819e).

Doyle, L. W., Spittle, A., Anderson, P. J. & Cheong, J. L. Y. School-aged neurodevelopmental outcomes for children born extremely preterm. Arch. Dis. Child. 106, 834–838 (2021).
Marlow, N. et al. No change in neurodevelopment at 11 years after extremely preterm birth. Arch. Dis. Child.-Fetal Neonatal Ed. 106, 418–424 (2021).
van Baar, A. L., van Wassenaer, A. G., Briët, J. M., Dekker, F. W. & Kok, J. H. Very Preterm Birth is Associated with Disabilities in Multiple Developmental Domains. J. Pediatr. Psychol. 30, 247–255 (2005).
Woodward, L. J. et al. Very preterm children show impairments across multiple neurodevelopmental domains by age 4 years. Arch. Dis. Child.-Fetal Neonatal Ed. 94, 339–344 (2009).
Luu, T. M. et al. Lasting Effects of Preterm Birth and Neonatal Brain Hemorrhage at 12 Years of Age. Pediatrics 123, 1037–1044 (2009).
Vandormael, C., Schoenhals, L., Hüppi, P. S., Filippa, M. & Borradori Tolsa, C. Language in Preterm Born Children: Atypical Development and Effects of Early Interventions on Neuroplasticity. Neural Plast. 2019, 6873270 (2019).
van Noort-van der Spek, I. L., Franken, M.-C. J. P. & Weisglas-Kuperus, N. Language Functions in Preterm-Born Children: A Systematic Review and Meta-analysis. Pediatrics 129, 745–754 (2012).
Pérez-Pereira, M. Prevalence of Language Delay among Healthy Preterm Children, Language Outcomes and Predictive Factors. Child. Basel Switz. 8, 282 (2021).
Luu, T. M., Vohr, B. R., Allan, W., Schneider, K. C. & Ment, L. R. Evidence for Catch-up in Cognition and Receptive Vocabulary Among Adolescents Born Very Preterm. Pediatrics 128, 313–322 (2011).
Vohr, B. Speech and language outcomes of very preterm infants. Semin. Fetal. Neonatal Med. 19, 78–83 (2014).
Kunnari, S., Yliherva, A., Paavola, L. & Peltoniemi, O. M. Expressive Language Skills in Finnish Two-Year-Old Extremely- and Very-Low-Birth-Weight Preterm Children. Folia Phoniatr. Logop. 64, 5–11 (2012).
Sanchez, K. et al. Conversational Language in 3-Year-Old Children Born Very Preterm and at Term. J. Speech Lang. Hear. Res. JSLHR 63, 206–215 (2020).
Guarini, A. et al. Long-term effects of preterm birth on language and literacy at eight years. J. Child Lang. 37, 865–885 (2010).
Barre, N., Morgan, A., Doyle, L. W. & Anderson, P. J. Language abilities in children who were very preterm and/or very low birth weight: a meta-analysis. J. Pediatr. 158, 766–774.e1 (2011).
Omizzolo, C. et al. Neonatal brain abnormalities and memory and learning outcomes at 7 years in children born very preterm. Memory 22, 605–615 (2014).
Lee, E. S. a, Yeatman, J. D. b, Luna, B. c & Feldman, H. M. a. Specific language and reading skills in school-aged children and adolescents are associated with prematurity after controlling for IQ. Neuropsychologia 49, 906–913 (2011).
Gobbo, C. & Chi, M. How knowledge is structured and used by expert and novice children. Cogn. Dev. 1, 221–237 (1986).
Biemiller, A. & Slonim, N. Estimating root word vocabulary growth in normative and advantaged populations: Evidence for a common sequence of vocabulary acquisition. J. Educ. Psychol. 93, 498 (2001).
Tulving, E. Episodic and semantic memory. in Organization of memory xiii, 423–xiii, 423 (Academic Press, Oxford, England, 1972).
Markovits, H., Fleury, M.-L., Quinn, S. & Venet, M. The development of conditional reasoning and the structure of semantic memory. Child Dev. 69, 742–755 (1998).
Vales, C., States, S. L. & Fisher, A. V. Experience-Driven Semantic Differentiation: Effects of a Naturalistic Experience on Within- and Across-Domain Differentiation in Children. Child Dev. 91, 733–742 (2020).
Siew, C. S. Q., Wulff, D. U., Beckage, N. M. & Kenett, Y. N. Cognitive Network Science: A Review of Research on Cognition through the Lens of Network Representations, Processes, and Dynamics. Complexity 2019, e2108423 (2019).
Goñi, J. et al. Switcher-random-walks: A cognitive-inspired mechanism for network exploration. Int. J. Bifurc. Chaos 20, 913–922 (2010).
Denervaud, S., Christensen, A. P., Kenett, Y. N. & Beaty, R. E. Education shapes the structure of semantic memory and impacts creative thinking. Npj Sci. Learn. 6, 1–7 (2021).
Schneider, J. et al. Evolution of T1 Relaxation, ADC, and Fractional Anisotropy during Early Brain Maturation: A Serial Imaging Study on Preterm Infants. Am. J. Neuroradiol. 37, 155–162 (2016).
Korkman, M., Kirk, U. & Kemp, S. NEPSY, bilan neuropsychologique de l’enfant [adaptation française 2003]. Montreuil Ed. ECPA (1997).
Barker, J. E. et al. Less-structured time in children’s daily lives predicts self-directed executive functioning. Front. Psychol. 5, (2014).
Largo, R. H. et al. Significance of prenatal, perinatal and postnatal factors in the development of AGA preterm infants at five to seven years. Dev. Med. Child Neurol. 31, 440–456 (1989).
Wechsler, D. WISC-V: Manuel technique et d’interprétation. PsychCorp Ed Don Mills Pearson (2016).
Raven, J., Raven, J. C. & Court, J. H. Manual for Raven’s progressive matrices and vocabulary scales. Section 1: General overview. (2003).
Christensen, A. P. & Kenett, Y. N. Semantic network analysis (SemNA): A tutorial on preprocessing, estimating, and analyzing semantic networks. Psychol. Methods (2021).
Christensen, A. P. NetworkToolbox: Methods and Measures for Brain, Cognitive, and Psychometric Network Analysis in R. R J. 10, 422–439 (2018).
Wickham, H. & Bryan, J. readxl: Read Excel Files. https://readxl.tidyverse.org, https://github.com/tidyverse/readxl (2023).
Handcock, M. S., Fellows, I. E. & Gile, K. J. RDS: Respondent-driven sampling. R Package version 09–9 (2023).
Christensen, A. P. SemNetDictionaries: Dictionaries for the’SemNetCleaner’package. (2019).
Christensen, A. P. SemNetCleaner: An automated cleaning tool for semantic and linguistic data. Package Version 1, (2019).
Kenett, Y. N. et al. Semantic organization in children with cochlear implants: Computational analysis of verbal fluency. Front. Psychol. 4, 543 (2013).
Massara, G. P., Di Matteo, T. & Aste, T. Network Filtering for Big Data: Triangulated Maximally Filtered Graph. Preprint at http://arxiv.org/abs/1505.02445 (2015).
Borodkin, K., Kenett, Y. N., Faust, M. & Mashal, N. When pumpkin is closer to onion than to squash: The structure of the second language lexicon. Cognition 156, 60–70 (2016).
Christensen, A. P., Kenett, Y. N., Aste, T., Silvia, P. J. & Kwapil, T. R. Network structure of the Wisconsin Schizotypy Scales–Short Forms: Examining psychometric network filtering approaches. Behav. Res. Methods 50, 2531–2550 (2018).
Christensen, A. P., Kenett, Y. N., Cotter, K. N., Beaty, R. E. & Silvia, P. J. Remotely Close Associations: Openness to Experience and Semantic Memory Structure. Eur. J. Personal. 32, 480–492 (2018).
Benedek, M. et al. How semantic memory structure and intelligence contribute to creative thought: a network science approach. Think. Reason. 23, 158–183 (2017).
Marchman, V. A. et al. Predictors of early vocabulary growth in children born preterm and full term: A study of processing speed and medical complications. Child Neuropsychol. 25, 943–963 (2019).
Rose, S. A., Feldman, J. F. & Jankowski, J. J. Modeling a cascade of effects: The role of speed and executive functioning in preterm/full-term differences in academic achievement. Dev. Sci. 14, 1161–1175 (2011).
Brydges, C. R. et al. Cognitive outcomes in children and adolescents born very preterm: a meta-analysis. Dev. Med. Child Neurol. 60, 452–468 (2018).
Anderson, P. J. Neuropsychological outcomes of children born very preterm. Semin. Fetal. Neonatal Med. 19, 90–96 (2014).
Nelson, D. L., Bennett, D. J., Gee, N. R., Schreiber, T. A. & McKinney, V. M. Implicit memory: Effects of network size and interconnectivity on cued recall. J. Exp. Psychol. Learn. Mem. Cogn. 19, 747 (19931201).
Bröring, T., Oostrom, K. J., Lafeber, H. N., Jansma, E. P. & Oosterlaan, J. Sensory modulation in preterm children: Theoretical perspective and systematic review. PLOS ONE 12, e0170828 (2017).
Bröring, T. et al. Sensory processing difficulties in school-age children born very preterm: An exploratory study. Early Hum. Dev. 117, 22–31 (2018).
Wallace, M. T., Woynaroski, T. G. & Stevenson, R. A. Multisensory integration as a window into orderly and disrupted cognition and communication. Annu. Rev. Psychol. 71, 193–219 (2020).
Quak, M., London, R. E. & Talsma, D. A multisensory perspective of working memory. Front. Hum. Neurosci. 9, (2015).
Marshall, C. Montessori education: a review of the evidence base. Npj Sci. Learn. 2, 1–9 (2017).
Ardila, A., Ostrosky-Solís, F. & Bernal, B. Cognitive testing toward the future: The example of Semantic Verbal Fluency (ANIMALS). Int. J. Psychol. 41, 324–332 (2006).
Dubossarsky, H., De Deyne, S. & Hills, T. Association Networks across the Lifespan: The Large-Scale Structure of Free Association Networks, Evidence for U-Shaped Developmental Changes and Semantic Saturation. S38 (2014).

No competing interests reported.

supplementarymaterials.docx

Characterization of Language Abilities and Semantic Networks in Very Preterm Children at School-age

Status:

Version 1

Abstract

Figures

Introduction

Method

Ethics

Design of the study

Participants

Very preterm children

Full-term children: reference group

Material

Verbal Fluency

Socioeconomic status (SES)

Non-verbal intelligence

Data Analyses

Group Comparison

Verbal Fluency

Semantic network analysis

Results

1. Demographics of Very Preterm Children

2. Group Comparison

4. Semantic network analysis

Discussion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1