Reading abilities have aroused great interest in the scientific community, and particularly the identification of those factors which predict the acquisition of reading competence. This identification can have important consequences for the promotion of reading ability, which is of crucial importance for children’s educational development.

Phonological awareness and rapid automatized naming (RAN) were identified as important predictors of reading abilities in typically developing children (TD) (Swanson et al., 2003; Melby-Lervag et al., 2012; Norton and Wolf, 2012). PA is related with conscious access to the phonological structure and components of words. RAN is the ability to quickly name aloud series of familiar letters, numbers, colors, or objects, which is related to speed processing, sustained attention and response inhibition, and lexical retrieval.

Evidence exists that PA and RAN uniquely contribute to different aspects of reading, and that the combination of deficits in both of them (double deficit hypothesis) produces more pervasive and severe reading impairments than single deficits in either RAN or PA (Wolf and Bowers, 1999; Schatschneider et al., 2002; Kirby et al., 2003; Papadopoulos et al., 2009; Vander Stappen and Reybroeck, 2018). Phonological awareness, letter-sound knowledge and alphanumeric RAN were all found to be strong independent predictors of reading development in two longitudinal studies (Caravolas et al., 2013; Clayton et al., 2019). Schatschneider et al. (2002), however, have pointed out the difficulty in establishing the relative impact of RAN deficits on reading ability independent of deficits in PA. In contrast to the evidence accumulated, Swanson et al. (2003) in a meta-analysis have suggested that the importance of RAN and PA measures in accounting for reading ability has been overstated.

In addition to PA and RAN, other studies have indicated that other precursors may have an impact on reading. Oral language development was found to be a strong predictor of reading ability in TD as well, and of reading comprehension in particular. A great number of studies have indicated the similarities between children with specific language impairment (SLI) and dyslexia (Bishop and Snowling, 2004), and have also indicated that children with SLI or language delay have a significantly higher probability than TD of showing subsequent reading impairments (Conti-Ramsden et al., 2001; Catts et al., 2002; Rescorla, 2002; Joye et al., 2019). Children diagnosed with dyslexia may have not only difficulties in phonological processing, but also in semantics, syntax and discourse (Bishop and Snowling, 2004; van Rijthoven et al., 2018). Deficits in phonological skills were found to be strong predictors of reading difficulties (Russell et al., 2018), although other linguistic abilities were also found to predict reading difficulties. Among them, expressive vocabulary, receptive vocabulary, and syntax have been mentioned as predictors of reading comprehension (Muter et al., 2004; Swanson et al., 2008; Lervag and Aukrust, 2010; Kieffer, 2012; Durand et al., 2013).

Similar to SLI children, children with reading impairment also show problems in working memory and other executive functions (Brosnan et al., 2002; Reiter et al., 2005). These affect phonological processing and phonological awareness, which, in turn, are strongly involved in the reading process. In studies carried out with TD children and children with dyslexia, verbal working memory and complex visuospatial memory were predictors of reading comprehension (Smith-Spark et al., 2003; Soriano and Miranda, 2010; Menghini et al., 2011; López-Escribano et al., 2013; Wang and Gathercole, 2013). Arnell et al. (2009) concluded that working memory encoding underlies part of the relationship between RAN and reading ability.

In a finding that is especially relevant for the aims of our study, Knoop-van Campen et al. (2018), found a mediation effect of PA on the relation between working memory and word reading efficiency in children with dyslexia: working memory affected word reading through PA. It is also theoretically sensible that phonological awareness mediates in the influence of language development on reading.

Most studies on reading abilities in preterm children (PT) were carried out with very preterm (VPT) or extremely preterm (EPT) children (gestational age < 32 weeks), and the results indicate that school-aged PT children obtain significantly lower results than full-term children (FT) in decoding abilities (Anderson et al., 2003; Johnson et al., 2012, 2016; Taylor et al., 2016; Alanko et al., 2017; Guarini et al., 2019), reading comprehension (Lee et al., 2011) or in both decoding abilities and reading comprehension (Pritchard et al., 2009, 2014; Johnson et al., 2011; Leijon et al., 2016, 2018). Similar results were found in two meta-analyses (Aarnoudse-Moens et al., 2009; Kovachy et al., 2015).

If the predictive variables of reading indicated before (PA, RAN, language, and EFs) are delayed in PT children, it is logical to think that PT children will show reading problems, given their role as precursors of reading abilities. The studies on these abilities in PT children have been mostly carried out with EPT or VPT children, as well. The few studies carried out with PT children of a wider GA range shed doubts on the idea of a general deficit in language development or EFs of PT children. In this regard, it is important to remember that EPT and VPT children represent only 20% of the total number of the PT population, and, therefore, there is a risk of making overgeneralizations from the investigation of VPT and EPT children to the general population of PT children.

In the same way, different studies with VPT or EPT children pointed to the existence of a deficit in PA and RAN in this population at 8 years of age (Alanko et al., 2017; Leijon et al., 2018).

Language delay has been commonly reported in EPT and VPT children (see Barre et al., 2011 for a meta-analysis). In contrast, healthy PT children with a wider GA range seem to progress in language similarly to TD in two studies using the same sample as reported here (Pérez-Pereira et al., 2014; Pérez-Pereira and Cruz, 2018).

Abundant investigation supports the conclusion that EPT and VPT children have deficits in EFs, such as working memory, attention, inhibition or flexibility, as compared to FT children (see van Houdt et al., 2019 for a meta-analysis). However, a study carried out with a sample of healthy PT of wider GA range (mean GA = 32.6, SD = 2.5) (the same sample as in this study) did not observe significant differences with FT children in working memory, inhibitory control and sustained attention (Pérez-Pereira et al., 2019).

The studies on the predictors of reading ability in PT children are scarce. Wocadlo and Rieger (2007) found that low performance in RAN increases the probability of difficulties in academic skills, including reading, in VPT children.

Guarini and Sansavini (2012) and Guarini et al. (2010) found that language (vocabulary, grammar, and PA) and short-term verbal memory had a predictive role on literacy at the age of 8 years for VPT children.

Anderson et al. (2018) studied the effect of the implementation of a working memory training program (Cogmed) on academic achievement (including word reading, spelling, sentence comprehension and mathematics), as well as on working memory, attention and executive behavior, in a sample of 7-year-old EPT children. No positive effect of the training was observed 24 months later.

Rose et al. (2011) using Structural Equation Modeling (SEM) found negative effects of prematurity on reading fluency (but not on letter-word identification), and these effects were mediated by processing speed and executive functions, working memory in particular. The authors found a cascade effect, in which prematurity negatively influenced processing speed, which then influenced EFs, which in turn influenced academic achievement (including reading). Working memory influenced reading independently of inhibition and shifting.

Borchers et al. (2019) studied the effect of a series of variables (PA, language, executive function, and non-verbal IQ) assessed at 6 years of age on text reading skills measured at 8 years of age in a group VPT children and a control group of FT children. VPT children had lower scores than FT children on all measures. Linear regressions analysis revealed that PA and language abilities predicted reading in both groups (accounting for 19.9 and 25.0% of variance, respectively, p < 0.001). Executive function and non-verbal IQ predicted reading only in children born preterm.

The novelty of the present investigation is that a great number of possible explanatory (exogenous) variables of reading abilities are studied in a longitudinal design in order to search for a good fit path analysis model which depicts dependencies among those variables. In addition, the sample differs from most previous studies, since it is composed of a wide range of GA variety (26-36 weeks), and the children did not show serious additional medical conditions, which makes it reasonable to think that it is a low risk sample.

The aims of the study are:

The hypotheses of the study are:

The participants form part of a longitudinal sample of children followed since birth. The children were recruited from four different neonatal units of hospitals in Galicia (Spain) at birth. Parents’ consent was previously obtained, as well as the authorization of the Galician Ethics Committee of Clinical Research.

The initial sample was 151 PT children and 49 FT children. The group of PT children had a mean GA of 32.60 (SD = 2.43, range 26–36), and the FT group had a mean GA of 39.84 (SD = 1.44, range 37–42). The mean Apgar scores (1 min) of the PT and FT children were similar: PT mean = 7.87, SD = 1.43, and FT mean = 8.08, SD = 1.25 (t (197) = −0.909, p > 0.05). The group of PT children did not show additional serious complications. Excluded on discharge from the hospital were those PT children who presented periventricular leukomalacia, intraventricular hemorrhage higher than II, hydrocephalus, genetic malformations, chromosomal syndromes, metabolic syndromes associated with intellectual disability (such as phenylketonuria, galactosemia, or homocystinuria), cerebral palsy or severe motor impairments (as diagnosed up until 9 months of age; no children were excluded between the time of hospital discharge and the following assessment), sensorial impairments, or Apgar scores lower than 6 at 5 min.

Data were collected by trained researchers who visited the children’s homes on three occasions within a 6-year interval. The first wave of data collection was carried out when the children were four years of age, and they were assessed on two EFs (working memory and inhibitory control). The second wave took place when the children were five years of age. They were assessed on RAN, language (morphosyntactic production, comprehension of syntactic structures and vocabulary comprehension), and PA (syllabic awareness and phonemic awareness). The third wave took place at nine years of age, and the children were assessed on RAs (letter name, word reading, pseudoword reading, and text comprehension). At this time, the PT sample consisted of 108 children, and the FT sample of 34 children. Both groups were similar in terms of distribution by gender (χ² (1) = 0.036, p > 0.05) and mother’s education (χ² (2) = 1.78 p > 0.05). The distribution of the children by GA groups (as shown in Table 1) was as follows: 23.9% between 26 and 31 weeks (very and extremely preterm children), 23.2% between 32 and 33 weeks (moderately preterm children), 28.9% between 34 and 36 weeks (late preterm children), and 23.9% above 36 weeks. The formation of the PT children’s groups was conditioned by the number of children available. We tried to have groups with a similar number of participants.

Out of a total of 200 children initially recruited, 142 participated in the present study. The reduction from the original number in the sample was due to experimental drop out. There was no substantial change in the characteristics of the sample, which remain very similar. For instance, the distribution of the children by GA groups in the initial sample was as follows: 24.5% between 26 and 31 weeks, 18.5% between 32 and 33 weeks (moderately preterm children), 32.5% between 34 and 36 weeks (late preterm children), and 24.5% above 36 weeks. The PT and FT groups of the initial sample also had a balanced distribution according to gender and mother’s education (χ2(1) = 0.000, p = ≥ 0.05, and χ2(2) = 8.66, p > 0.05, respectively). The mean Apgar scores of the initial sample and those of the sample used in this study were very similar (EPT and VPT: 6.90 and 7.24, respectively; MPT: 8.38 and 8.27, respectively; LPT: 8.31 and 8.20, respectively; and FT: 8.08 and 8.18, respectively).

The Spanish version of Childhood Executive Functioning Inventory (CHEXI, Thorell and Nyberg, 2008) was used to assess working memory and inhibitory control in daily life in children between 4 and 12 years old. CHEXI is completed by children’s parents and it includes 24-items, 5-point Likert-type format (1 = absolutely uncertain, 5 = very true). Parents rate how much each assertion is a true description of the behavior of the child (e.g., “Cuando se le pide que haga varias cosas, sólo recuerda la primera o la última”: When the child is asked to do several things, he/she only remembers the first or the last). Higher scores indicate greater difficulty in working memory and inhibitory control, and lower scores indicate fewer difficulties in working memory and inhibitory control.

The Rapid Automatized Naming (RAN) subtest of the Spanish version of Clinical Evaluation of Language Fundamentals (CELF-4, Semel et al., 2006) was used to assess naming speed (sustained attention and inhibitory control, and fast processing) in persons between 5 and 21 years of age. Children are asked to name rapidly a set of colors (e.g., “rojo”: red), a set of shapes (e.g., “cuadrado”: square) and a set of combining shapes and colors (e.g., “triángulo azul”: blue triangle) that are each presented in a 6 × 6 matrix. Scores for accuracy in naming (RAN-err) were calculated in the matrix of combining shapes and colors, counting the number of errors committed by the child. The number of errors the child committed evidences the degree to which he/she can sustain self-monitoring (accuracy).

The production subscale of the Test de Sintaxis de Aguado (TSA, Aguado, 1999) was used to assess the morphosyntactic production in children between 3 and 7 years of age. It consists of 29 items. The first twenty-five items contain two figures each. In each item, the researcher says two sentences (e.g., “La chica mira los perros”: The girl looks at the dogs; “la chica mira al perro”: the girl looks at the dog), one after the other, without pointing to any picture. Immediately after speaking, the researcher points to one of the images and he/she waits for the child to repeat the match sentence. Then, the other image is pointed to, so that the child repeats the other sentence. The last four items are items of grammatical closure. The production score is obtained by considering the participant’s use of articles, adverbs, prepositions, passive sentences, negations, reflexive sentences, relative clauses, etc. The child receives one point for each correct sentence given.

The Comprensión de Estructuras Gramaticales (CEG, Mendoza et al., 2005) was used to assess the comprehension of syntactic structure in children between 4 and 11 years of age. It consists of 80 sheets that include four pictures each. In each item, the researcher pronounces a sentence (e.g., “El niño que mira a la niña está comiendo”: The boy who looks at the girl is eating) and the child points to the image that matches the target sentence. The other three images act as (lexical or grammatical) distractors. The total number of correct answers was used for the analysis.

The Peabody Test de Vocabulario en Imágenes (PPVT-III, Dunn et al., 2006) was used to assess vocabulary comprehension in people between 2.5 and 90 years of age. It comprises 192 sheets arranged in order of increasing difficulty. Each sheet includes four pictures. In each item, the researcher pronounces a word (e.g., “vaca”: cow) and the child is required to point to the image that best matches that word. The total score is obtained by subtracting the number of errors from the ceiling item.

The phonological awareness scale of Del Lenguaje Oral al Escrito-Evaluación (LOLEVA, Peralbo et al., 2015) was used to assess syllabic and phonemic awareness in children between 3 and 8 years of age. It comprises thirteen tasks: rhyme recognition, initial syllable identification, final syllable identification, initial syllable addition, final syllable addition, initial syllable omission, final syllable omission, initial phoneme identification, final phoneme identification, initial phoneme addition, final phoneme addition, initial phoneme omission and final phoneme omission. Each task includes instructions and two examples that are presented in audiovisual format, except for the omission and addition items. All subscales consist of five items, with the exception of the rhyme recognition subtest, which contain ten. The child receives one point for each correct answer given (out of a possible 70 points).

The Batería de evaluación de los procesos lectores, revisada (PROLEC-R, Cuetos et al., 2007) was used to assess reading capacity. This test can be used with children between 6 and 12 years of age. It consists of nine tasks: identification of letters, same–different, word reading, pseudoword reading, grammatical structures, punctuation, sentence comprehension, text comprehension and listening. In the present study, only the scores of the subscales of identification of letters, word reading, pseudoword reading and text comprehension were used. The identification of letter task consists of a list of 20 letters (e.g., “g”); the word reading task consists of a list of 40 words that vary in length, frequency of use, and the complexity of their syllabic structure (e.g., “peine”: comb); the pseudoword reading task consists of a list of 40 invented words (e.g., “pueña”). In these three tasks, the researcher points to the item (letter, word or, pseudoword) and the child reads it out loud. In each task, the child receives a precision score, measured as the sum of the correct answers, and a speed score, measured as the time taken to complete the task. A combined score (efficiency) is calculated by dividing the precision score by the speed score and multiplying the result by 100. The efficiency score was used for the analyses in this study.

Last, the text comprehension task consists of two narrative and two expositive texts. For each text, the children have to respond to four written questions, 16 total responses (e.g., “¿Para qué sacó varias monedas de la hucha?”: Why did she take a few coins out of the piggybank?). The child receives one point for each correct answer given.

In all cases raw scores were used for the analyses, not percentile or scalar scores.

Firstly, a set of ANOVAs were carried out in order to analyze differences in EFs, language, PA, and reading abilities as regards GA. Partial eta square was used as the estimator of the magnitude of differences between groups. Secondly, zero-order correlations were computed aimed at the examination of inter-relationships among all the study variables. Finally, the effect of EFs, language and PA on reading, as well as the mediating role of PA in the relation between EFs and language on reading, was examined by means of path analysis, which permits the simultaneous modeling of several related regression relationships. Path analysis was selected because it allows for the examination of more complex models including the analysis of the relationships with a set of observed dependent variables as well as mediation effects. The effect of gender and gestational age was controlled. The model was estimated by the Maximum likelihood (ML) method and the following goodness of fit indexes were used for the assessment of the model fit: comparative fit index (CFI), Tucker-Lewis index (TLI), root-mean-square error of approximation (RMSEA), and standardized root mean squared residual (SRMR). According to Hu and Bentler (1999) suggestions, RMSEA and SRMR values lower or equal to 0.06, and TLI and CFI values of 0.95 or higher were considered indicators of excellent model fit. Given that the variables in the model were directly observed and all direct and indirect effects were freely estimated however, the simple mediation path model would be just-identified leading to a perfect model fit. Descriptive analysis and zero-order correlations were conducted on IBM SPSS Statistics 24, whereas path analysis was carried out in MPLUS 7.4 (Muthén and Muthén, 2011).

Descriptive information and differences on EFs, language, PA, and reading abilities as regards GA are presented in Table 1. In order to delve into the specificities of the development of children born prematurely, preterm children were categorized into three different groups according to their weeks of gestation (i.e., < 32 weeks, 32-33 weeks, and 34-36 weeks), whereas the full-term children correspond to the gestation group of more than 36 weeks. The formation of the PT children’s groups was conditioned by the number of children available. We tried to have groups with a similar number of participants. The results showed significant differences between GA groups on inhibitory control and syntactic structure comprehension. The Tukey’s HDS post hoc test evidenced higher scores (which indicate unfavorable performance) in inhibitory control in children born with 34-36 weeks of gestation than those born at 32-33 weeks, as well as higher scores (which indicate favorable results) in syntactic structure comprehension of full-term children as compared with children born at 34-36 weeks. Even then, the results of the ANOVAs revealed a lack of significant differences between preterm and full-term children in the remaining variables, including EFs, language, PA, and reading.

Zero-order correlations between all the variables of study are presented in Table 2. The corresponding findings indicated a significant positive relation of working memory with inhibitory control. We also found significant negative associations of working memory with morphosyntactic production, syntactic structure comprehension, text comprehension, and letter names, as well as between inhibitory control and syntactic structure comprehension, and pseudoword reading. These results mean that the lower the working memory and the inhibitory control problems, the higher the language and reading scores, and viceversa. A significant negative association was also found between RAN and morphosyntactic production, RAN and vocabulary comprehension, and RAN with phonemic awareness. On the other hand, the results showed significant positive associations of morphosyntactic production with syntactic structure comprehension, syllabic awareness, phonemic awareness, and letter names; significant correlations of syntactic structure comprehension with vocabulary comprehension and PA, both syllabic and phonemic awareness; significant inter-relations of syllabic awareness with phonemic awareness and word reading; and highly significant associations among letter names, word reading and pseudoword reading.

In order to assess both the effect of EFs, language and PA on reading in a longitudinal study, as well as the mediating role of PA on the relationship between EFs and language with reading, a mediating path analysis model was implemented, controlling for the effect of gender and GA (Figure 1). Given the lack of differences found between preterm and full-term children in most of the variables, the whole sample was included in the analysis. The path analysis model evidenced a perfect model fit (CFI = 1.00, TLI = 1.00, RMSEA = 0.00, SRMR = 0.00), because the model was just-identified. The results indicated significant positive direct effects of working memory on syllabic and phonemic awareness as well as significant negative direct effects on text comprehension and letter names. At the same time, inhibitory control showed a significant negative effect on syllabic awareness whereas the latter significantly positively predicted the ability of word reading. Likewise, significant direct effects were found as regards morphosyntactic production on syllabic awareness and syntactic structure comprehension on text comprehension, in a positive and negative way, respectively.

The potential mediating effect of PA, both syllabic and phonemic, on the relationship between EFs and language on reading was also analyzed as part of the path analysis model. The bootstrapping results showed only a single significant indirect effect of syllabic awareness on the relationship between working memory and word reading ability (β = 0.18, p < 0.05, 95% CI = 0.058, 0.315).