Does preterm birth increase a child's risk for language impairment

Early Human Development 86 (2010) 765–772 Contents lists available at ScienceDirect Early Human Development j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / e a r l h u m d e v Does preterm birth increase a child's risk for language impairment? Alessandra Sansavini a,⁎, Annalisa Guarini a, Laura M. Justice b, Silvia Savini a, Serena Broccoli c, Rosina Alessandroni d, Giacomo Faldella d a Department of Psychology, University of Bologna, Bologna, Italy School of Teaching and Learning, The Ohio State University, United States Department of Statistical Sciences, University of Bologna, Bologna, Italy d Unit of Neonatology, Department of Woman, Child, and Adolescent Health, S. Orsola Hospital, University of Bologna, Bologna, Italy b c a r t i c l e i n f o Article history: Received 29 April 2010 Received in revised form 16 August 2010 Accepted 24 August 2010 Keywords: Preterm birth Language impairment Predictive indexes Lexicon development Grammar development a b s t r a c t Background: Although premature birth is associated with lags in language acquisition, it is unclear whether preterms exhibit an elevated risk for language impairment (LI). This study determined whether preterms, without frank cerebral damage, at 2;6 and 3;6 exhibited a higher rate of risk for LI as compared to full-terms, and also sought to identify predictors of risk. Method: Sixty-four Italian very immature preterms were assessed longitudinally at 2;6 and 3;6; age-matched full-terms served as controls at 2;6 (n = 22) and 3;6 (n = 40). Each completed individualized assessments of cognition and language ability. At each time point, using cut-offs specific to each of the language measures, children were differentiated into two groups (at risk for LI, not at risk). Results: The percentage of full-terms at risk for LI at 2;6 (9.1% to 13.6%) and 3;6 (7.5%) was consistent with prior estimates of LI at these ages. The percentage of preterms at risk for LI at 2;6 (16.1% to 24.1%) and 3;6 (34.4%) was higher at both ages and statistically significant at 3;6 (difference = 26.8%, 95% CI = 12.3% to 41.4%). The best model predicting risk status at 3;6 was preterms' mean length of utterance (MLU) at 2;6, (sensitivity 72.73%, specificity 85%) when adjusting for maternal education. Conclusion: Preterms exhibit a heightened risk for LI in the preschool years, since about one in four preterms at 2;6 and one in three preterms at 3;6 experiences significant lags in language acquisition. Findings argue the importance of early identification of language difficulties among preterms coupled with implementation of systematic language-focused interventions for these youngsters. © 2010 Elsevier Ireland Ltd. All rights reserved. Accumulating evidence shows that premature birth, characterized by low (b37 weeks, LGA) or very low gestational age (≤32 weeks, VLGA) and/or low (b2500 g, LBW) or very low birth weight (≤1500 g, VLBW), even in the absence of significant cerebral damage, has negative impacts on a variety of important developmental outcomes [1–3]. For instance, results of a 2002 meta-analysis showed a mean difference of about 10 standard score points at school-age between LBW and VLBW preterms and full-terms on measures of general intellect (i.e., IQ) [1]. Relatedly, preterms exhibit an elevated rate of significant developmental difficulties and disabilities as compared to full-terms [4]. For example, nearly one-third of LBW preterms are rated by their teachers as having poor reading skills [5] and one-fifth exhibit learning disabilities [6]; moreover, about one-half of VLBW preterms have at least one identified neurodevelopmental disability [7]. The present study was conducted to contribute to this expanding literature by examining risk for language impairment (LI) among preterms at 2;6 and 3;6, a timeframe in which children's language ⁎ Corresponding author. Department of Psychology, University of Bologna, viale Berti Pichat 5, 40127 Bologna, Italy. Tel.: + 39 051 2091879; fax: + 39 051 243086. E-mail address: [email protected] (A. Sansavini). 0378-3782/$ – see front matter © 2010 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.earlhumdev.2010.08.014 skills are rapidly expanding and at which risks for LI are often recognized by parents and health professionals. Epidemiological surveys find that about 19% of 2-year-olds exhibit significant lags in language development, referred to as late language emergence (LLE) [8]. LLE, a term that is synonymous with the concept of ‘late talkers,’ is typically characterized by significant lags in lexical (e.g., delays in accumulating new words) and grammatical development (e.g., delays in combining words to create multi-word utterances) in absence of any salient developmental abnormality (e.g., autism, severe hearing loss, severe cognitive disability). Many children who have late language emergence will normalize their language skills by 3 and 4 years, and those who do not are identified as having LI. LI refers to an absolute impairment in the domain of language that occurs in absence of frank sensory or neurological impairment (e.g., autism, hearing loss); it is estimated to affect about 7 to 10% of children [9,10]. Children with LI are typically identified on the basis of unexpectedly poor performance on norm-referenced measures of language [10]. Children with LI whose cognitive skills are within the normal range (i.e., N−1 SD of the mean) are typically described as having a specific language of impairment (SLI) whereas those whose cognitive skills are below the normal range (i.e., between −2 and −1 SD of the 766 A. Sansavini et al. / Early Human Development 86 (2010) 765–772 mean) are described as having a nonspecific language impairment (NSLI). In the present study, we use the term LI as an umbrella term that includes both groups of children, similar to other research on these populations [10]. Presence of LI in the later preschool years is associated with a range of future adverse outcomes for children, including poor reading [11] and math achievement [12] as well as difficulties with social competence [13]. To date, research on the language development of preterms has largely attended to determining whether and to what extent specific language abilities are affected [14]. Interestingly, individual studies of the lexical and grammatical development of preterms do not offer converging evidence of the presence of systematic delays in these aspects of language. Such discrepancies in findings reflect, at least in part, methodological variability among studies in how preterms are selected for inclusion and exclusion, the assessment tools used to examine children's language abilities, the timing of language assessments, and the native language that children are acquiring. Studies of preterms' development of lexical skills provide one such example of a lack of convergence. Cross-sectional studies of Finnish infants using the MacArthur–Bates Communicative Development Inventory (MB-CDI) showed no significant difference in vocabulary size when comparing VLBW preterms and full-terms at 2 years [15,16]. Similar findings were found in a recent Italian study involving VLBW preterms (age 2;6) that excluded children with severe cerebral damage; interestingly, however, in this study, male preterms with a birthweight ≤ 1000 g were found to exhibit significantly poorer lexical abilities as compared to male and female preterms with a birthweight N 1000 g [17]. Such findings suggest that a sub-set of preterms may be at particular risk for language difficulties [14]. Nonetheless, other cross-sectional studies using the MB-CDI have shown that, as a group, preterms do tend to have lags in lexical development as compared to full-terms [18,19]. One study, for instance, studied the vocabulary size of preterms varying in gestational age using the MB-CDI at 2 years. Preterms with extremely low gestational age (ELGA b 28 weeks) had a significantly smaller lexicon compared to preterms with very low gestational age (28– 32 weeks), who in turn had a reduced lexicon as compared to fullterms [18]. Some of the discrepancy in findings across such studies of early lexical development may relate to whether preterms with severe cerebral damage are included among the preterm samples; this is unspecified in all but the work of Foster-Cohen and colleagues [19], who found significant differences in the expressive lexicon of New Zealander preterms and full-terms on the MB-CDI at 2 years. Most recently, research on Finnish preterms examined lexical size at 2 years for preterms with and without concomitant neurological disability, using the Finnish version of the MB-CDI [20]. Preterms with neurological disability had significantly smaller lexicons compared to full-terms at 2 years, whereas preterms with no such disability did not differ from full-terms in their lexical skills. An interesting aspect of this work was that preterms' language skills were assessed at 2 years using both indirect (MB-CDI) and direct measures (Reynell Developmental Language Scales [21]), which resulted in divergent findings. That is, although preterms without neurological disability did not differ from full-terms in lexical skills as measured by the MB-CDI, they exhibited significantly poorer receptive and expressive language skills than the full-terms at 2 years as measured by the Reynell. Although the MB-CDI has reasonable concurrent and predictive relations with direct measures of language skill [22], direct assessment is recognized as the gold standard for documenting children's language abilities [22] and may be more sensitive to identifying meaningful differences in group performance on behavioural measures. Beyond the lexicon, grammatical development also appears to be affected by preterm birth, and results are more convergent as compared to the research on the lexicon. Finnish VLBW preterms exhibit a shorter mean length of utterance (MLU) at two years as compared to full-terms [15], with similar findings also reported in studies of French and New Zealander preterms [18,19]. We have previously reported findings of significant grammatical difficulties in an Italian sample of male preterms (gestational age b 31 weeks) at 2;6 [17]. Also, in a study of slightly older (3;6) Italian preterms, Sansavini and colleagues [3] found that MLU was significantly lower for VLBW preterms as compared to full-terms converging with findings of lags in grammatical competence for French preterms at the same age [23]. Particularly concerning is longitudinal evidence showing that these early lexical and grammatical lags do not resolve over time during preschool-age [24] and school age [25]. In sum, the available research suggests, albeit somewhat inconsistently, that preterm birth may be associated with early lags in lexical and grammatical development, and longitudinal studies suggest that preterms at school-age may exhibit learning-related problems that are often attributable to underlying poor language and literacy abilities [5,25]. However, few studies have systematically examined risk for LI among preterms during early childhood, when LI is typically first identified and when early interventions to facilitate language growth are typically introduced. We note here three exceptions. First, Briscoe and colleagues [14] used a direct assessment of language ability in a study involving 26 VLBW preterms at 3 years, finding that 8 of 26 (31%) scored below an a priori empirically derived cut-off demarking a child's risk for LI. The researchers reported that risk for LI is over-represented among preterms, although the small sample size – particularly for those identified as being at risk – raises concerns about the confidence we can place in this finding. Second, Singer and colleagues [26] likewise assessed the language skills of VLBW preterms at 3 years, differentiating those with bronchopulmonary dysplasia (BPD; n = 90) from those without (VLBW, n = 65) as compared to full-terms (n = 91).The language measure utilized, the Battelle Developmental Inventory [27], is a semi-structured tool that integrates observation, interview, history, and direct tasks to arrive at standard scores. These researchers found that 43%, 31%, and 28% of the BPD, VLBW, and full-term samples, respectively, exhibited risk for LI using a b−1 SD of the mean cutpoint; there was little evidence of a difference in rate of risk between the VLBW and full-term samples. The high number of full-terms identified as at risk, which far exceeds prevalence estimates for LI among typical children, raises questions about the validity of the tool used to identify risk; consequently, the usefulness of this study's findings for estimating risk for LI among VLBW is questionable. Of additional concern is that the VLBW sample included a number of children with significant developmental complications that may have compromised language acquisition, including hydrocephalus, hearing loss, and intra-ventricular hemorrhage. Finally, a recent study of New Zealander preterms determined the percentage of children at age 4 years who exhibited language delay; language delay was identified based on direct assessment of language skills using a − 1 SD cut-off [28]. The percentage of preterms identified as language delayed at age 4 was 31%, which was significantly higher than what occurred for a comparison group of full-terms (15%). This study offers an important contribution to the understanding of risk for LI among preterms; nonetheless, it is also necessary to note that the preterm sample included a substantial number of children with poor cognitive abilities (i.e., 34% had concomitant cognitive delay) as well as health complications (e.g., cerebral palsy). Consequently, it is unclear whether this rate of risk for LI is elevated somewhat due to sample characteristics. The available literature therefore suggests that close to one-third of VLBW preterms may exhibit a significant risk for LI during the early childhood years, although these estimates are based on studies that have some limitations or on samples involving a considerable number of preterms with cognitive limitations. The present study was therefore conducted to determine: (1) Do very immature preterms exhibit a higher rate of language difficulties at 2;6 and 3;6 as compared to full-terms? (2) To what extent can language difficulties A. Sansavini et al. / Early Human Development 86 (2010) 765–772 at 3;6 among preterms be predicted from direct and indirect measures of language and cognition collected at 2;6? 1. Method 1.1. Participants This study involved 126 Italian children (64 very immature preterms, 62 full-terms) comprising three sub-samples. A preterm sample consisting of 64 children was assessed longitudinally at 2;6 and 3;6, whereas two control samples of full-terms were ascertained to provide cross-sectional references for the preterms at 2;6 (n = 22) and 3;6 (n = 40). 1.1.1. Preterm sample Sixty-four monolingual Italian very immature preterms were recruited from the Unit of Neonatology of Bologna University, which is one of the main tertiary care level units equipped with assisted ventilation of the Emilia–Romagna Region. Birth dates for the preterms ranged from May 1995 to October 2000. Cranial ultrasound scan (US) was carried out for all neonates within the first 4 days of life and then repeated weekly during the first month of life. Those neonates with abnormal US in the first month of life were reexamined weekly until normalization, and then two times per month until discharge. After discharge, all preterms returned for reexamination using the US at the presumed date of birth and again at 3 months (corrected age); they then entered into a medical followup at the Day-Hospital of Neonatology. At 2;6, those preterms who met the criteria explained below and whose parents accepted that their child took part in this longitudinal research involving two points of developmental assessment, at 2;6 and 3;6, were enrolled. The preterms were recruited into the study if, at birth, they had met three primary criteria: (a) gestational age ≤ 33 weeks and a birthweight ≤ 1600 g, (b) absence of major cerebral damage [i.e., periventricular leukomalacia (PVL), intra-ventricular hemorrhage (IVH N II grade), hydrocephalus, retinopathy of prematurity (ROP N II grade)] and congenital malformations, and (c) no indication of visual or hearing impairment. Residence in the city of Bologna or close to it was added as a criterion, since infants living far from Bologna were followed by medical structures close to their residence. Furthermore, since the present prospective longitudinal study aimed at analyzing which measures of language and cognition collected at 2;6 were predictive of language difficulties at 3;6 among preterms, only those infants who attended both the 2;6 and 3;6 assessment at the scheduled corrected ages (first assessment at 2;6, maximum range from 29 to 33 months; second assessment at 3;6, maximum range from 41 to 45 months) were included in this study. As in many studies of the development of preterms, age was corrected for the preterms so as to take into account their level of neurobiological maturation [17,29]. At 2;6, the preterms' corrected age in months was 30.3 months (SD = 0.92, range 29 to 33) and, at 3;6, was 42.2 months (SD = 0.69, range 41 to 45). The preterms (32 males, 32 females) enrolled in this study had a mean gestational age at birth of 30.4 weeks (SD = 2.1, range 24.5 to 33); mean birth weight was 1192.5 g (SD = 261.9; range 600 to 1600). In this sample, gestational age and birth weight were highly correlated, r = 0.63. At ascertainment, we did allow for preterms with some history of medical complication to be enrolled, which included small for gestational age (SGA, n = 28), respiratory distress syndrome (RDS, n = 44) for which 12 had needed mechanical ventilation, bronchopulmonary dysplasia defined as the need of supplemental oxygen at 36 weeks of postconceptional age (BDP, n = 5), IVH of Grade I or II (n = 4), ROP of Grade I or II (n = 4), and hyperbilirubinemia treated with phototherapy (n = 38). In addition, 16 children had had persistent hyperechogenicity (HE) of white matter (≥14 days) as 767 indicated by US; however, none of these children had developed PVL because, in all instances, the HE had been completely resolved at 3 months. The sample of preterms is best described as representing the general range of socioeconomic status (SES) strata, as estimated from mothers' highest level of educational attainment: 14 mothers (22%) had a primary educational level (completed basic education), 31 (48%) had a secondary level (completed high school), and 19 (30%) had a college level (University/Master's degree or beyond). 1.1.2. Control groups Sixty-two healthy full-term children were recruited to serve as reference groups using a cross-sectional research design; specifically, 22 full-terms were recruited to serve as the 2;6 year control group, and a separate group of 40 full-terms were recruited to serve as the 3;6 year control group. The criteria to select the full-term children are explained below. All full-term children should have experienced normal birth (gestational age N 37 weeks and birth weight N 2800 g), and had no history of major cerebral damage and/or congenital malformations or visual or hearing impairments. Furthermore, all fullterm infants, as the preterms, had to be born within the same period as the preterms and living in the city of Bologna or close to it. An equal distribution of gender similar to that of the preterms was also required. In addition, to be compared with the preterms, full-term children were required to be between 29 and 33 months at the time of the 2;6 assessment and between 41 and 45 months at the time of the 3;6 assessment. The 2;6 year control group of 22 full-terms (11 males, 11 females) was recruited from three nursery schools in Bologna. Birth dates for the full-terms ranged from February 1997 to August 1998. The average age at the moment of the assessment was 31.1 months (SD = 1.02). As with the preterms, these children's background spanned lower- to higher-levels of SES, based on mothers' highest level of education: 7 mothers (32%) had a primary educational level, 12 (55%) had a secondary level, and 3 (13%) had a college level (University/Master's degree or beyond). However, there was a smaller proportion of full-term mothers educated to college level than the 2;6 year preterm group (13% versus 30%). The 3;6 year control group of 40 full-terms (17 males, 23 females) was recruited from three kindergartens in Bologna and their birth dates ranged from June 1995 to January 1999. The mean age at the time of assessment was 42.1 months (SD = 0.66). These children's background spanned lower- to higher-levels of SES, based on mothers' highest level of education: 7 mothers (17%) had a primary educational level, 20 mothers (50%) had a secondary level, and 13 mothers (32%) had a college level. The full-term and the preterm samples had a similar distribution of maternal level of education. 1.2. Measures Standardized assessment tools were used to assess children's language and cognitive abilities at 2;6 and 3;6. At 2;6 years, three measures were implemented. The first measure was an indirect (parent-report) measure of children's lexical and grammar skills, the Il Primo Vocabolario del Bambino (PVB) [30,31]. The PVB is the Italian version of the MacArthur–Bates Communicative Development Inventory (MB-CDI) [32,33], designed for use with infants from 18 to 36 months. In this study, two parts of the PVB were used. Part I, a measure of lexical expression, comprises a checklist of 670 words across various categories. Parents are instructed to mark every word their child uses, to arrive at a total score representing word use (PVB Total Words). Part III, a measure of grammatical expression, consists of a checklist of 37 pairs of sentences, one written in telegraphic style and the other as a complete grammatical sentence. Parents are instructed to mark the item in each pair that represents how their child would say the sentence. The present study used analyzes the 768 A. Sansavini et al. / Early Human Development 86 (2010) 765–772 concerning the educational and social background of the families and the children's health was obtained during a parent interview at the time of each child's assessment. The study adhered to ethical guidelines concerning protection of human subjects, including adherence to the legal requirements of the study country, and all parents of the preterms and full-terms gave informed written consent for participation to the study, data analysis, and data publication. number of items for which children used the complete grammatical form of a sentence results (PVB Complete Sentences). Also at 2;6, children completed direct assessments of both language and cognitive ability. For the former, the Prova di Ripetizione di Frasi (PRF) [34,35] was administered as a direct index of language ability. This Italian test of sentence repetition assesses Italian children's grammatical ability from 2;0 to 4;0. The construct validity of the PRF is supported by prior work showing correlations of this measure to related measures of speech and language. In addition, test–retest comparisons have shown that children's performance on this measure is extremely stable [34,35]. In administrating this test, children are asked to repeat 27 sentences of different length and grammatical complexity while looking at illustrations conveying meaning. The mean length of children's utterances (MLU) can be calculated based on performance across the 27 sentences (PRF MLU). For the latter, the Italian version of Form L-M of the Stanford-Binet Intelligence Scale [36] was administered; this general measure of cognition has been used in other Italian studies of at risk populations [3,37]. This measure was used because the 4th version has not been translated or standardized on Italian children. At 3;6, the two direct assessments (PRF and Stanford-Binet) were re-administered. The PVB was not re-administered, as it is used only through 3;0 years. 1.4. Methods for identifying children's risk for LI Procedures derived from the extant literature were used to identify those children who exhibited risk for LI at both 2;6 and 3;6. The procedures involved applying specific cut-offs to language scores at 2;6 and 3;6 to differentiate groups of children (at risk for LI, not at risk for LI). Note that IQ was not employed in grouping decisions, as our focus was on estimating risk for LI to include both nonspecific (LI concomitant with lower IQ) and specific language impairment (LI concomitant with average or better IQ). In actuality, very few children in this sample had low IQ scores. At 2;6, among the preterm sample, 8 children (12.5%) had an IQ b 85 (range 63.3 to 83.3), whereas among the full-terms 2 children had an IQ b 85 (9.1%, range 83.9 to 84.4). At 3;6, four preterms (6.3%) had an IQ b 85 (range 73.9 to 83.3), whereas no full-terms had an IQ b 85. At 2;6, children's parents completed the PVB, which is the Italian version of the MB-CDI. Identification of risk for LI among 2-year-olds commonly utilize estimates of lexical size [38], most prominently through parent-completed vocabulary checklists. The 10th percentile is commonly used to identify young children who exhibit late language emergence at this age [8,22]. Therefore, at 2;6, we identified children at risk for LI using the 10th percentile cut-point of the PVB Total Words using reference to the Italian normative values [31]. Additionally, as a complement to this indirect measure, we also identified risk for LI at 2;6 based on the MLU metric derived from the PRF, administered at both 2;6 and 3;6. MLU is a valid and reliable index of language skill among children prior to 5 years; it also has the desirable feature of being strongly correlated with more complex measures of language [34,35,39]. For the present purpose, and consistent with procedures used in both basic and clinical research [22], we used a cut-point of b−1.25 SD (which corresponds to the 10th percentile cut-off used with the PVB) of the mean for the PRF MLU measure at 2;6 and 3;6 to identify children at risk. Although some studies identify children with LI using a less-stringent cut-off (−1 SD of the mean), the −1.25 SD cut is more sensitive and specific [10]. Means and standard deviations were based on scores for the typically developing full-terms, based on procedures in Bishop and Edmundson [40], collected at the two time points (see Table 1). Therefore, children were identified as exhibiting risk for LI if they received a PRF MLU score b 0.9 at 2;6 or b 3.5 at 3;6. 1.3. Procedure When children were 2;6 years, parents of all preterms and fullterms (the respective reference group) completed the PVB questionnaire (corrected age for preterm infants). Parents returned the completed questionnaires in person. Information concerning the educational and social background of the families and the children's health was obtained during a parent interview at the time of each child's assessment. The PRF and the Stanford-Binet Intelligence Scale were individually administered to children in a quiet room of the Unit of Neonatology for preterms and in a quiet room of their nursery for full-terms. All assessments were videotaped in their entirety so that the PRF could be scored subsequently by two trained observers working independently to ensure reliability. Inter-rater agreement was appropriately high (N90% agreement) and all disagreements were resolved via consensus procedures. In a few instances, there was missing data at the 2;6 assessment point (e.g., the child would not complete a specific task, parents did not complete all portions of the questionnaire); therefore, when presenting analyses, we provide the exact n. When children were 3;6 years, the preterms and full-terms (the respective reference group) were individually administered the PRF and Stanford-Binet following the same procedures as the 2;6-year assessment. Preterms were assessed at the Unit of Neonatology and the full-terms were assessed at their kindergartens. Information Table 1 Comparison of preterms and full-terms on language and cognitive measures. Age Preterms Full-terms t test ES Difference 95% CI Construct (measure type) n M SD n M SD t d 2;6 PVB Total Words PVB Complete Sentences PRF MLU Stanford-Binet IQ 58 57 62 64 389.2 16.8 2.1 102.6 178.5 14.9 1.4 16.1 22 22 22 22 446.9 20.9 2.4 108.3 156.7 13.7 1.2 14.5 − 1.33 − 1.12 − 0.74 − 1.45 0.33 0.28 0.22 0.36 − 57.7 − 4.1 − 0.2 − 5.6 − 143.91 − 11.36 − 0.90 − 13.32 28.50 3.16 0.41 2.07 0.19 0.26 0.46 0.15 3;6 PRF MLU Stanford-Binet IQ 64 64 3.7 103.5 0.9 11.5 40 40 4.2 114.7 0.6 9.2 − 3.78 − 5.22 0.62 1.04 − 0.5 − 11.2 − 0.84 − 15.48 − 0.26 − 6.96 b0.001 b0.001 Note: d calculations made using Hedges' g to correct for bias. Min p Max 769 A. Sansavini et al. / Early Human Development 86 (2010) 765–772 Table 2 Percentage of preterms and full-terms at risk for LI at 2;6 and 3;6. Age Measure Cut-point Preterms Full-terms Difference 95% CI Min 2;6 3;6 a PVB Total 10th perc 24.1% Words PRF MLU −1.25 SD 16.1% PRF MLU −1.25 SD 34.4% pa Max 13.6% 10.5% − 8.4% 29.4% 0.37 9.1% 7.5% 7.0% 26.8% − 10.3% 24.4% 0.72 12.3% 41.4% 0.002 Fisher's Exact Test. 2. Results Descriptive information about the language abilities of preterms relative to full-terms at 2;6 and 3;6 appear in Table 1. At 2;6, independent-samples t tests showed no significant differences (all psN 0.15) between preterms and full-terms on the language or cognition measures, as we have reported previously [17]. The effect of group was not significant for PVB Total Words, PVB Complete Sentences, PRF MLU and the Stanford-Binet IQ, when adjusting for maternal level of education [2-way MANOVA test of between-subjects effect: respectively F (1,76) = 2.20, p = 0.14; F (1,75) = 1.69, p = 0.20; F (1,80) = 0.65, p = 0.42; F (1,82) = 3.27, p = 0.07]. Although the full-terms exhibited an advantage across all measures at 2;6, these differences were consistent with small effect sizes (ds between 0.22 and 0.36). On the contrary, at 3;6, there was a statistically significant difference between preterms and full-terms on both direct assessments: PRF MLU, t (102) = −3.78, p b 0.001 (d = 0.62) and the Stanford-Binet, t (102) = 5.23, p b 0.001 (d = 1.04). The effect of group remained significant both for PRF MLU and Stanford-Binet IQ when adjusting for maternal level of education [2-way MANOVA test of between-subjects effect: respectively F (1,100) = 11.40, p = 0.001; F (1,100) = 28.02, p b 0.001]. These differences show there to be appreciable betweengroup differences in the language and cognitive skills of the preterms relative to full-terms, consistent with prior findings [3]. To address the first research question, we determined the percentage of preterms and full-terms at risk for LI at 2;6 and 3;6 compared to full-terms (see Table 2). At 2;6, about one-fourth (n = 14; 24.1%) of preterms were at risk for LI as compared to about one-seventh (n = 3; 13.6%) of full-terms based on the indirect lexical measure (10th percentile cut-off on the PVB Total Words). For the direct assessment using PRF MLU, 16.1% (n = 10) of preterms and 9.1% (n = 2) of full-terms were at risk for LI. In neither case was the differential rate of risk between preterms and full-terms significant. It is interesting to note that 8 preterms were at risk for LI on both PVB and MLU measures, while 8 preterms and 5 full-terms were at risk for LI only in one of the two language measures. Among preterms and full-terms at risk for LI based on MB-CDI and/or MLU, 7 preterms and 1 full-term had low scores (b85) on the Stanford-Binet. At 3.6, identification of risk for LI was based exclusively on the PRF MLU using the −1.25 SD cut-point. Results showed that about one-third (n = 22; 34.4%) of preterms exhibited risk for LI as compared to 7.5% (n = 3) of full-terms, a difference that was statistically significant. Of the 22 preterms found to be at risk for LI, only 2 had low IQ scores (b85). Findings therefore show that the risk for LI among preterms is appreciable, with about one-third of preterms affected. Of note is that the rate of risk found for LI among full-terms, at 7.5%, is similar to that found previously in a sample of 1364 healthy 3-year-olds (7%) [10]; likewise, the rate of risk for LI found for the preterms, at 34.4%, is similar to that found previously in a small sample of 26 preterms (31%) [14]. The second research question concerned whether risk for LI at 3;6 among preterms can be predicted from language and cognitive skills as measured at 2;6 (for a comparison on these and other variables, see Table 3). For these analyses, note that only data from the preterm sample were considered. Logistic regression was used given the single dichotomous outcome, namely presence or absence of risk for LI at 3;6. Predictor variables comprised the three language measures collected at 2;6 (PVB Total Words, PVB Complete Sentences, and PRF MLU) and the direct assessment of cognition (Stanford-Binet). Maternal education (coded categorically based on highest level completed: primary, secondary, college degree, with highest level as the reference) was also included in the model, as prior research has shown there to be possible relations to the outcome variable [8]. As only 22 children were at risk of LI at 3;6 it was not possible to perform a single model using all the predictive variables. Therefore, four separate multivariate logistic regression models were run, where the dependent variable was risk for LI at 3;6 and the independent variables were maternal education (for all models), and PVB Total Words for model 1, PVB Complete Sentences for model 2, PRF MLU at 2;6 for model 3 and Stanford-Binet IQ for model 4. The omnibus test was statistically significant for all models (all ps b 0.05) and the overall classification accuracy ranged from 71.88% for model 4 (45.45% sensitivity, 84.71% specificity) to 80.65% for model 3 (72.73% sensitivity, 85.00% specificity). In terms of individual predictors, all of the languages measures at 2;6 and the Stanford-Binet IQ were associated with the outcome. Maternal education showed a trend for the secondary/college degree contrast in three of the four models. Parameter estimates are presented in Table 4, to include odds ratios (OR) with 95% confidence intervals (CI). In all analyses, the odds ratio represents the factor by which one's odd for LI at 3;6 increases for every one-unit change in the predictor variable. Table 3 Comparison of preterms based on risk status for LI at 3;6. Not at risk At risk for LI Age Measure — M (SD) or % n 42 n 22 Birth Gestational Age Birthweight Matern. educ. — primary Matern. educ. — secondary Matern. educ. — college PVB Total Words PVB Complete Sentences PRF MLU Stanford-Binet IQ PRF MLU Stanford-Binet IQ 30.7 (1.8) 1222.3 (220.7) 19.0% 42.9% 38.1% 449.9 (149.8) 21.1 (14.9) 2.7 (1.1) 107.8 (12.3) 4.3 (0.3) 107.4 (10.1) 29.9 (2.5) 1135.7 (324.8) 27.3% 59.1% 13.6% 282.3 (178.0) 8.8 (11.2) 1.1 (1.2) 92.8 (18.0) 2.6 (0.7) 96.1 (10.4) 2;6 3;6 Difference − 0.8 − 86.6 8.2% 16.2% − 24.5% 167.5 − 12.3 − 1.6 − 15.0 − 1.7 − 11.3 pa CI 95% Min Max − 1.98 − 243.86 − 13.8% − 10.1% − 45.8% − 255.32 − 19.30 − 2.19 − 23.68 − 2.00 − 16.63 0.46 70.70 30.2% 42.6% − 3.1% − 79.73 − 5.21 − 0.97 − 6.26 − 1.39 − 5.88 0.22 0.27 0.13 b0.001 0.001 b0.001 0.001 b0.001 b0.001 Note: birthweight reported in grams; gestational age reported in months; maternal education documented when children were 2;6 to represent highest level of education attained; data at 2;6 on the PVB Total Words represents 37 children not at risk and 21 children at risk, on the PVB Complete Sentences represents 37 children not at risk and 20 children at risk, and on PRF MLU represents 40 children not at risk and 22 children at risk. a t-test for independent samples for continuous variables and chi-squared test for discrete variables (maternal education). 770 A. Sansavini et al. / Early Human Development 86 (2010) 765–772 Table 4 Prediction of risk for LI at 3;6 among preterms adjusting for maternal education. Independent variables 95% CI McFadden's Odds Ratio R2 Min Model 1 PVB Total Words Maternal education (primary) Maternal education (secondary) Model 2 PVB Complete Sentences Maternal education (primary) Maternal education (secondary) Model 3 PRF MLU at 2;6 Maternal education (primary) Maternal education (secondary) Model 4 Stanford-Binet IQ Maternal education (primary) Maternal education (secondary) p Max 0.22 0.99 3.88 4.08 0.9899 0.9978 0.59 25.65 0.84 19.73 0.002 0.16 0.08 0.93 4.36 5.11 0.88 0.68 1.04 0.98 27.80 25.07 0.004 0.12 0.05 0.28 5.24 6.21 0.14 0.75 1.09 0.56 36.44 35.28 b0.001 0.09 0.04 0.94 2.72 2.60 0.90 0.47 0.57 0.98 15.87 11.95 0.003 0.27 0.22 0.20 0.34 0.19 Note: results come from four separate multivariate logistic models where the dependent variable is Risk for LI at 3;6 and the independent variables are maternal education, for all models, and PVB Total Words for model 1, PVB Complete Sentences for model 2, PRF MLU at 2;6 for model 3 and Stanford-Binet for model 4. Maternal education was coded categorically based on highest level completed: primary, secondary, college degree, with highest level (college) as the reference. Consideration of the odds ratios indicates that a child's risk for LI at 3;6 increases more than threefold (1/0.28 = 3.57) for every one-unit decrease in MLU, increases 10% for every ten-unit decrease in PVB Total Words (1/(0.9910) = 1.1), increases about 2 times for every tenunit decrease in PVB Complete Sentences (1/(0.9310) = 2.07) and in Stanford-Binet IQ (1/(0.9410) = 1.86). Regarding maternal education, children whose mother had only primary schooling (OR = 5.24, 95% CI = 0.75 to 36.45) or secondary schooling (OR = 6.21, 95% CI = 1.09 to 35.28) as their highest level of education were more likely to exhibit risk for LI at 3;6 compared to children whose mother had a college degree. As a post hoc consideration, we questioned whether children's risk status at 3;6 could be reliably predicted from their risk status at 2;6, on the basis of either the PVB Total Words or PRF MLU criteria. Two post hoc models were conducted; for each, risk for LI at 3;6 was the dependent variable. In the first, risk for LI based on PVB Total Words (10th percentile cut-off) at 2;6 served as the predictor variable with maternal education included as a control. The omnibus test of the model was statistically significant, χ2(3, N = 58) = 13.62, p = 0.003, and the overall classification accuracy of this model was 72.4% (38.1% sensitivity, 91.9% specificity). Children's risk status based on PVB Total Words at 2;6 (OR = 0.13, 95% CI = 0.03 to 0.54) was significantly related to their risk status at 3;6 (p = 0.005); children who were below the 10th percentile on the PVB Total Words at 2;6 were nearly 8 times more likely to be at risk for LI at 3;6. In the second model, risk for LI based on the PRF MLU (−1.25 SD cut-off) at 2;6 served as the predictor variable, again with maternal education included as a control. The omnibus test was statistically significant, χ2(3, N = 62) = 20.21, p b 0.001 and the overall classification accuracy of the model was 77.42% (40.91% sensitivity, 97.5% specificity). Children's risk status based on the PRF MLU at 2;6 (OR = 30.98, 95% CI = 3.15 to 304.94) was significantly related to their risk status at 3;6 (p = 0.003); children identified as at risk at 2;6 using the PRF MLU were nearly 45 times more likely to be identified as at risk for LI at 3;6. 3. Discussion An ample literature on the circumstances surrounding preterm birth shows that these youngsters face an increased risk for a number of adverse outcomes that transcend both physical (e.g., cerebral palsy) [41] and cognitive dimensions of development [1]. Of relevance to this study, a number of research findings have suggested that preterms exhibit a substantially heightened risk for LI during early childhood [14,26]. As we discussed previously, LI refers to a specific or nonspecific impairment in language ability relative to normative (age-based) references that is not the consequent of significant sensory or neurological impairment (e.g., autism, hearing loss); it appears to be a strong contributing factor to children's later academic achievement, particularly in the area of reading [11]. Briscoe and colleagues [14], for instance, found that 31% of preterms exhibited risk for LI at 3 years, characterized by significantly lower language abilities in relation to other preterms as well as full-terms. However, there are some limitations to the extant work on this topic that make it important to further investigate this issue, such as, depending on the studies, small sample size, lack of full-term reference group, inclusion of preterms with significant cerebral damage, and measurement imprecision. In the present work, our goal was to systematically examine the rate of risk for LI among very immature preterms, without frank cerebral damage, at 2;6 and 3;6 and also to determine whether risk for LI at 3;6 could be predicted from measures collected at 2;6. The first major finding of interest was that a greater percentage of preterms, at both 2;6 and 3;6, exhibit risk for LI as compared to fullterms, although the difference in rate of risk was statistically significant only at 3;6. At 2;6, when children are speaking primarily in two- and three-word utterances and show a rapid advance in lexical skills, 16.1 to 24.1% of preterms show risk for LI compared to 9.1 to 13.6% of full-terms, depending upon the domain of language examined (grammar, lexicon) and the assessment approach utilized (direct, indirect). The percentage of full-terms exhibiting risk in this study at 2;6 (9.1% on the PRF MLU) aligns well with previous findings on the risk for LI among 3-year-olds; for instance, La Paro and colleagues [42], in applying a slightly more stringent criteria that in the present work (− 1.33 SD of the mean) found 7% of 3-year-olds to exhibit LI. Based on the current estimate of risk for LI at 2;6 at about 9% for healthy full-terms, the present findings show that risk for LI is over-represented among preterms even at 2;6; nearly twice as many preterms as full-terms show risk for LI at this age, a compelling finding albeit one that did not achieve statistical significance. Between the ages of two and three, many children who are “late talkers” will catch up with their peers, particularly with respect to the lexicon [43]. Consequently, it is not surprising to see that at 3;6, only 7.5% of the full-terms in this study showed risk for LI, a figure that is highly consistent with prior research on the rate of risk for LI among children during early childhood [10]. By comparison, we found that 34% of preterms exhibited risk for LI at 3;6; these very immature preterms experienced risk for LI at a rate of about 5 times that of the full-terms. While this rate of risk is concerning, it is not entirely unexpected given that it converges well with prior research. Briscoe and colleagues [14], as we previously discussed, found that 8 of 26 3-year-old VLBW preterms (33%) exhibited risk for LI using a standardized measure of linguistic complexity; Singer and colleagues [26] similarly found that 31% of VLBW preterms at 3 years exhibited risk for LI on a multi-faceted assessment of language ability (although the high rate of identification among the reference group of full-terms raises questions about the sensitivity of the measure used). Most recently, Woodward and colleagues [28] reported that 31% of preterms exhibited language delay. With such convergence across studies, we can conclude quite strongly that very immature preterms, even among preterms who do not exhibit significant cerebral damage or cognitive disability, experience risk for LI at a rate that substantially exceeds that of healthy full-terms. The second major finding warranting discussion is the finding that preterms exhibiting risk for LI at 3;6 show a history of significantly poorer language and cognitive skills compared to those preterms not at risk, similar to findings reported previously in smaller-scale A. Sansavini et al. / Early Human Development 86 (2010) 765–772 research [14]. For instance, parental estimates of children's lexical size at 2;6 using the PVB Total Words measure show an appreciable difference between preterms identified at 3;6 as at risk (282 words) and not at risk (450 words). Similarly, preterms identified as at risk for LI at 3;6 also exhibited poorer cognitive skills at 2;6 (M = 92.8) as compared to preterms not at risk (M = 107.8), although it is important to note that the majority of preterms were not clinically impaired in the area of cognition (i.e., only a few had an IQ of less than 85). This is a potentially important finding as it implies that the language difficulties apparent at 3;6 among one-third of the preterms should not be conceptualized as occurring in isolation, but may reflect more generalized lags in cognitive development, albeit these are not severe delays. Such an argument was made recently by researchers assessing the rate of “serious” language problems (performance ≤ − 2 SD, 16% rate of identification) among extremely premature children at 6 years [44]; these researchers reported that significant language problems did not tend to occur in isolation of more general cognitive deficits. Important to note, as we have elsewhere, is that many of the studies of language difficulties among preterms involve children with cognitive scores which remain in the normal range. It has been hypothesized that specific types of cognitive difficulties, such as attention and working memory, could contribute both to lower cognitive and linguistic performance [3,24]. The third major finding concerned our effort to predict children's risk status at 3;6 from language and cognitive measures collected at 2;6, for which children's expressive language skills based on MLU appeared to be the strongest cognitive/linguistic predictor of children's future risk status. In fact, for every one-unit decrease in MLU, children's odds for LI at 3;6 increased more than threefold. This finding is consistent with results from longitudinal research on late talkers [45], in which researchers sought to predict children's language ability at 3 years from a range of measures collected at 2 years. Such studies generally show that expressive language skill at 2 years, as compared to receptive language skill or nonverbal cognition, is a particularly strong predictor of late talkers' future language skills [45]. Although use of indirect report instruments, such as the PVB used in this study, clearly have established value for atscale use (e.g., implementation of large-scale screening programs), it is also the case that MLU has a long-standing place in early language assessment because it is a valid and reliable index of growth in the aspect of language most closely affiliated with language impairment. It is not particularly surprising, therefore, that MLU as assessed directly at 2;6 provided the strongest estimate of children's risk for LI at 3;6, as coupled with maternal education, which has surfaced in other studies as a potential contributor to healthy outcomes for children at risk for LI [8,10]. It should also be pointed out that the PVB questionnaire measures the lexicon and, specifically, the number but not frequency, of produced words. Therefore, even if several studies have shown that lexicon and grammar are tightly related both in typical and atypical development [46], we would expect that in terms of predictivity there would be a stronger association among two measures (i.e. MLU) of the same competence (i.e. grammar). As a final comment, we want to reference the post hoc analyses conducted to assess the potential for estimating children's risk for LI from their risk status at 2;6, based on either direct assessment (PRF MLU, using the −1.25 SD cut-off) or indirect assessment (PVB Total Words, using the 10th percentile cut-off). Both measures – MLU and lexical size estimates – and the cut-offs applied are commonly used approaches for identifying children's risk for LI during early childhood [8,14,22]. Our study found that, irrespective of measure used and cutoff applied, identifying children at risk at 2;6 was not a good predictor of their risk status at 3;6 (sensitivity less than chance for both measures). In other words, there is a lack of stability in status from 2;6 to 3;6 for at least some children who have poor language performance at 2;6. We suspect that this is largely due to instability in status for children who hover around the cut-off point, that is, who have more 771 modest language difficulties relative to children on the extremes, as has been discussed previously [47]. In other words, children with severe language difficulties tend to be relatively stable in status over time whereas those with more mild difficulties may periodically appear within the normal range. Of consequence, however, is the implication that the use of stringent cut-points for identifying children who are at risk for LI at toddlerhood and targeting only these children for intensive intervention efforts will fail to address the needs of some children. Some limitations of our study need to be taken into account, both as suggestions for future studies and as cautions for the generalization of the results. First, with regard to the preterm sample, we included in the study preterm children whose parents accepted to take part in this longitudinal study and who participated in both the 2;6 and at the 3;6 assessments. This may present some selection bias, and it is unclear whether these results generalize to preterm children whose parents did not consent to participating in the longitudinal study or who did not attend follow-up assessments. Second, with regard to the fullterm samples at 2;6 and 3;6, which served as comparisons to the preterms at these time points, the data were cross-sectional rather than longitudinal. It is unclear whether results would have differed if a comparison group of full-terms assessed longitudinally had been available, and we propose this as an important design consideration in future work on this topic. Furthermore, although our full-term groups were adequate for comparison with the preterm group and the analyses controlled for maternal level of education, some caution is required in generalizing the results to typical development, because of the smaller size of our full-term sample at 2;6 and a lower representation of mothers with a college level of education in this sample. To sum, this study determined whether very immature preterms exhibit an increased rate of risk for LI during the toddler and preschool years. An important characteristic of this study is that it involved only preterms for whom significant cerebral damage, such as severe cognitive disability and hearing loss, was not apparent; consequently, risk estimates are pertinent for preterms who may be considered at relatively low-risk compared to preterms more severely affected by premature birth. This study provides convincing evidence that approximately one in three preterms exhibits significant lags in language development at 3;6, characterized by limitations in grammatical expression (i.e., short utterance length) and a history of slow lexical development, concomitant with modest weakness in overall cognition. The most obvious implication of these findings is that they beg the importance of early identification of risk for LI among preterms and the initiation of intensive early intervention to improve language outcomes. Conflict of interest statement None declared. Acknowledgements We would like to thank the children and parents for their participation in the research and Giulia Aquilano for help with the medical examination. Funding: This research was supported by research grants from University of Bologna: Basic Oriented Research ex 60% 2004, 2005, 2006; Strategic Project 2007–2009 “Early communicative-linguistic and cognitive abilities: risks linked to preterm birth”. This research was also supported by a national research grant PRIN 2008 “Gestures and language in children with atypical and at risk developmental profiles: relationships among competences, mother–child interaction modalities and proposals of intervention”. A fellowship from the Institute of Advanced Studies at the University of Bologna to the third author made this collaboration possible. 772 A. Sansavini et al. / Early Human Development 86 (2010) 765–772 Ethical approval: The study met ethical guidelines, including adherence to the legal requirements of the study country, and received a formal approval by the Research Ethical Committee of the Department of Psychology at the University of Bologna. Moreover, all parents of the preterm and full-term infants gave informed written consent for participation to the study, data analysis, and data publication. References [1] Bhutta AT, Cleves MA, Casey PH, Cradock MM, Anand KJS. Cognitive and behavioural outcomes of school-aged children who were born preterm. A metaanalysis. JAMA 2002;288:728–37. [2] Sansavini A, Rizzardi M, Alessandroni R, Giovanelli G. The development of Italian low- and very-low-birthweight infants from birth to 5 years: the role of biological and social risks. Inter J Beh Dev 1996;19(3):533–47. [3] Sansavini A, Guarini A, Alessandroni R, Faldella G, Giovanelli G, Salvioli G. Are early grammatical and phonological working memory abilities affected by preterm birth? J Commun Disord 2007;40:239–56. [4] Pritchard VE, Clark CAC, Liberty K, Champion PR, Wilson K, Woodward LJ. Early school-based learning difficulties in children born very preterm. Early Hum Dev 2009;85:215–24. [5] Huddy CL, Johnson A, Hope PL. Educational and behavioural problems in babies of 32-35 weeks gestation. Arch Dis Child Fetal Neonatal Ed 2001;85:F23–8. [6] Cherkes-Julkowski M. Learning disability, attention deficit disorder and language impairment as outcomes of prematurity: a longitudinal study. J Learn Disabil 1998;31(3):294–306. [7] Cole C, Binney G, Casey P, Fiascone J, Hagadorn J, Kim C. Criteria for determining disability in infants and children: low birth weight. Evidence report/technology assessment No. 70 (Prepared by Tufts New England Medical Center Evidencebased Practice Center under Contract No. 290-97-0019). AHRQ Publication No. 03E01. Rockville, MD: Agency for Healthcare Research and Quality; 2002. December. [8] Zubrick SR, Taylor CL, Rice ML, Slegers DW. Late language emergence at 24 months: an epidemiological study of prevalence, predictors, and covariates. J Speech Lang Hear Res 2007;50:1562–92. [9] Beitchman J, Hood J, Rochon J, Peterson M, Mantini T, Majumdar S. Empirical classification of speech/language impairment in children. Identification of speech/ language categories. J Am Acc Child Adol Psych 1989;28:112–7. [10] Catts H, Fey M, Tomblin J, Zhang X. A longitudinal investigation of reading outcomes in children with language impairments. J Speech Lang Hear Res 2002;45:1142–57. [11] Skibbe LE, Grimm K, Stanton-Chapman T, Justice LM, Bowles R. Reading trajectories of children with specific language impairment from preschool through fifth grade. Lang Speech Hear Serv Sch 2008;39 475-386. [12] Fazio BB, Naremore RC, Connell PJ. Tracking children from poverty at risk for specific language impairment: a 3-year longitudinal study. J Speech Lang Hear Res 1996;39:611–24. [13] Redmond S, Rice ML. The socioemotional behaviors of children with SLI: social adaptation or social deviance? J Speech Lang Hear Res 1998;41:688–700. [14] Briscoe J, Gathercole SE, Marlow N. Short-term memory and language outcomes after extreme prematurity at birth. J Speech Lang Hear Res 1998;41:654–66. [15] Jansson-Verkasalo E, Valkama M, Vainionpää L, Pääkkö E, llkko E, Lehtihalmes M. Language development in very low birth weight preterm children: a follow-up study. Folia Phoniatr Logop 2004;56:108–19. [16] Stolt S, Klippi A, Launonen K, Munck P, Lehtonen L, Lapinleimu H, et al. Size and composition of the lexicon in prematurely born very-low-birth-weight and fullterm Finnish children at two years of age. J Child Lang 2007;34:283–310. [17] Sansavini A, Guarini A, Alessandroni R, Faldella G, Giovanelli G, Salvioli GP. Early relations between lexical and grammatical development in very immature Italian preterms. J Child Lang 2006;33:199–216. [18] Gayraud F, Kern S. Influence of preterm birth on early lexical and grammatical acquisition. First Lang 2007;27:159–73. [19] Foster-Cohen S, Edgin JO, Champion PR, Woodward LJ. Early delayed language development in very preterm infants: evidence from the MacArthur–Bates CDI. J Child Lang 2007;34:655–75. [20] Stolt S, Haataja L, Lapinleimu H, Lehtonen L. The early lexical development and its predictive value to language skills at 2 years in very-low-birth-weight children. J Commun Disord 2009;42:107–23. [21] Reynell J. Reynell developmental language scales. Windsor: NFER; 1977. [22] Heilman J, Weismer ES, Evans J, Hollar C. Utility of the Macarthur–Bates Communicative Development Inventory in identifying language abilities of latetalking and typically developing toddlers. Am J Speech Lang Pathol 2005;14: 40–51. [23] Crunelle D, Le Normand MT, Delfosse MJ. Langage orale et écrit chez des enfants prématurés: résultats à 7½ ans. Folia Phoniatr Logop 2003;55:115–27. [24] Guarini A, Sansavini A, Fabbri C, Alessandroni R, Faldella G, Karmiloff-Smith A. Reconsidering the impact of preterm birth on language out come. Early Hum Dev 2009;85:639–45. [25] Guarini A, Sansavini A, Fabbri C, Savini S, Alessandroni R, Faldella G, et al. Longterm effects of preterm birth on language and literacy at eight years. J Child Lang 2010;37:865–85. [26] Singer LT, Siegel AC, Lewis B, Hawkins S, Yamashita T, Baley J. Preschool language outcomes of children with history of bronchopulmonary dysplasia and very low birth weight. J Dev Behav Pediatr 2001;22(1):19–26. [27] Newborg J, Stock JR, Wnek L, et al. Battelle developmental inventory. Allen, Tex: DLM; 1988. [28] Woodward LJ, Moor S, Hood KM, Champion PR, Foster-Cohen S, Inder TE, et al. Very preterm children show impairments across multiple neurodevelopmental domains by age 4 years. Arch Dis Child Fetal Neonatal Ed 2009;94:339–44. [29] Vohr BR, Wright LL, Dusick AM, Mele L, Verter J, Steichen JJ, et al. Neurodevelopmental and functional outcomes of extremely low birth weight infants in the National Institute of Child Health and Human Development Neonatal Research Network, 1993–1994. Pediatrics 2000;105:1216–26. [30] Caselli MC, Casadio P. Il primo vocabolario del bambino. Guida all'uso del questionario MacArthur per la valutazione della comunicazione e del linguaggio nei primi anni di vita. Milano: Franco Angeli; 1995. [31] Caselli MC, Pasqualetti P, Stefanini S. Parole e frasi nel Primo Vocabolario del Bambino. Milano: Franco Angeli; 2007. [32] Fenson L, Dale PS, Reznick JS, Thal DJ, Bates E, Hartung JP, et al. MacArthur communicative development inventories: user's guide and technical manual. San Diego, CA: Singular Publishing; 1993. [33] Fenson L, Marchman VA, Thal DJ, Reznick JS, Bates E. MacArthur–Bates communicative development inventories: user's guide and technical manual. 2nd Edition. Baltimore: Brookes; 2007. [34] Devescovi A, Caselli MC. Una prova di ripetizione di frasi per la valutazione del primo sviluppo grammaticale. Psicologia Clinica dello Sviluppo 2001;5:341–64. [35] Devescovi A, Caselli MC. Sentence repetition as a measure of early grammatical development in Italian. Int J Lang Comm Dis 2007;42(2):187–208. [36] Bozzo MT, Mansueto Zecca G. Scala di intelligenza Standford–Binet, forma L-M (III revisione di Terman–Merrill). Firenze: Organizzazioni Speciali; 1968. [37] Vicari S, Caravale B, Carlesimo GA, Casadei AM, Allemand F. Spatial working memory deficits in children at ages 3–4 who were low birth weight preterms infants. Neuropsychology 2004;18:673–8. [38] Rescorla L, Mirak J, Singh L. Vocabulary acquisition in late talkers: lexical development from 2;0 to 3;0. J Child Lang 2000;27:293–311. [39] Rice ML, Redmond SM, Hoffman L. MLU in children with SLI and younger control children shows concurrent validity, stability, and parallel growth trajectories. J Speech Lang Hear Res 2006;49:793–808. [40] Bishop DVM, Edmundson A. Language-impaired 4-year-olds: distinguishing transient from persistent impairment. J Speech Lang Hear Res 1987;52:156–73. [41] Hack M, Wilson-Costello D, Friedman H, Taylor GH, Schluchter M, Fanaroff AA. Neurodevelopment and predictors of outcomes of children with birth weight of less than 1000 g: 1992–1995. Arch Pediatr Adolesc Med 2000;154(7):725–31. [42] La Paro K, Justice LM, Skibbe L, Pianta R. Relations among maternal, child, and demographic factors and the persistence of preschool language impairment. Am J Speech Lang Pathol 2004;13:291–303. [43] Rescorla L, Roberts J, Dahlsgaard K. Late talkers at 2: outcome at age 3. J Speech Lang Hear Res 1997;40:556–66. [44] Wolke D, Samara M, Bracewell M, Marlow N. Specific language difficulties and school achievement in children born at 25 weeks of gestation or less. J Pediatr 2008;152:256–62. [45] Rescorla L, Merrin L. Communicative intent in toddlers with specific expressive language impairment (SLI-E). Appl Psycholinguistics 1998;19:393–414. [46] Bates E, Goodman JC. On the inseparability of grammar and the lexicon: evidence from acquisition, aphasia, and real-time processing. Lang Cogn Process 1997;12: 507–84. [47] Justice LM, Bowles R, Pence Turnbull K, Skibbe LE. School readiness among children with varying history of language difficulties. Dev Psychol 2009;45:460–76.