Preterm birth is likely to impact significantly on brain development since the central nervous system of the premature baby is not fully prepared to function independently outside the intra-uterine environment [4]. In fact, neuroimaging research has identified anatomical abnormalities as the result of premature birth, such as smaller hippocampus [5], lower gray-to-white-matter ratio [6], and smaller cerebellum [7]. These structural changes in the brain are thought to be related to deficits in cognitive functioning [8].

Moreover, other neonatal medical complications are also common in prematurity such as septicaemia, intraventricular haemorrhage, chronic lung disease, apnea, bradycardia, and so on [9]. Medical complications and treatment interventions, coupled with prolonged hospital stays, can impact negatively on brain development and contribute to accentuate long-term neurobehavioral deficits [10].

Environmental factors can also play an important role in developmental outcomes of prematurity. For example, preterm infants from high SES families show fewer problems later in life when compared to infants from low SES families [11]. Frontal and posterior brain regions implicated in language and attention skills may be particularly affected by environmental factors [8]. In fact, programmes to decrease environmental risk in preterm infants have been shown to improve neurobehavioral functioning involving frontal and occipital areas [12]. Gender differences have also been observed. Boys seem to attain less favourable cognitive outcomes than girls [13], although this finding is not consistent across studies [14].

Language difficulties are prevalent in premature children and include articulation problems and expressive language delays, which can manifest themselves as poor vocabulary and grammar. Difficulties with phonological awareness are also common and predict later poor reading and writing [15,16]. In fact, preterm birth is likely to have long-term consequences, affecting linguistic development beyond preschool [15].

There is some controversy regarding the conceptualization of language development as either solely dependent on language processes or as tied to more domain-general cognitive processes [17]. Several studies have attributed language impairments in premature infants, especially those born extremely preterm, to a general cognitive deficit affecting several areas of functioning [16]. However, not much is known about the potential impact of more general cognitive mechanisms, such as attention, on language development [17,18].

Studies on executive function and attention skills in premature infants have considered the general-cognitive-deficit hypothesis as having insufficient explanatory power, suggesting that attention deficits might be a specific area of weakness in preterm children [19]. A recent meta-analysis has confirmed that attention skills are especially problematic in preterm children [20]. Attention problems seem to occur even in cases of non-extreme low birth weight [21]. Factors such as gestational age, gender, and environmental factors (e.g., mother's SES) strongly contribute to the extent of attention problems [20].

Attention and language become increasingly interdependent throughout development. Initially, the infant's attention system is basically comprised of orienting processes characterized by visual exploration of the environment. Later on, by the end of the first year, a second attentional system, the executive attention system, emerges. It enables the child to engage in goal-oriented activity and inhibition of behaviour. This system is also closely associated with language development by enabling the production of verbal cognitions to control behaviour [22].

Several studies have found high co-morbidity of cognitive deficits early in development in both full- and preterm children [1,23,24]. In particular, there is a frequent co-morbidity of language and attention problems [4,25]. Neurological studies have shown that children with language disorder show differential activation of cerebral regions involved in several attention processes [26].

Children with better attention are expected to be faster in language learning because they tend to be better at following adults' gazes, at engaging in joint attention, and at tracking the referents of other people's verbalizations [17]. Dysregulation of attention and arousal are expected to interfere with the ability to maintain a focused state, affecting children's opportunity to engage in social interactions and decreasing opportunities for language learning.

Not many studies have addressed relations between attention and language in preterm children. According to the general-cognitive-deficit hypothesis, cross-sectional associations between these two skills would be expected both in infancy and early childhood. Moreover, since attention seems to be a particularly weak area for preterm children [19], it is possible that early attention problems have a precursor role in predicting later language ability. In fact, early attention deficits have been identified as important in signalling future cognitive difficulties [4], such as those related to language processing. Language problems in infancy are difficult to diagnose [27] and the identification of early cognitive markers of language impairment can facilitate earlier detection and increase our understanding of underlying mechanisms associated with less favourable outcomes of prematurity.

Initial PLBW-group analyses revealed that attention problems and language ability were quite stable over time when adjusted for reciprocal influence. Moreover, significant albeit modest cross-sectional associations between attention problems and language were observed both at 18 and 36 months, in line with previous studies [22]. Given that attentional and linguistic processes become increasingly interdependent throughout development [22], it is perhaps not surprising that associations between these two skills were modest, especially at 18 months. We found slightly stronger associations between residual change in language ability and attention problems at 36 months, both for PLBW and control children, possibly reflecting a trend toward higher interdependence between the two variables over time. Moreover, as anticipated, attention at 18 months emerged as a significant predictor of language ability at 36 months (adjusting for language ability at 18 months).

Similar relations between parameters reflecting stability in attention problems and language ability were observed in PLBW children and controls, as well as similar patterns of cross-sectional associations. However, the two models differed with regards to lagged paths. For the PLBW children, the path from attention problems at 18 months to language ability at 36 months was larger than that observed for the control group. This could potentially suggest a stronger predictive role of early attention problems on later language in PLBW children than in controls. However, since the latent variables seem to have a different meaning in the two groups (lack of group measurement invariance), one cannot determine whether the difference between lagged coefficients is statistically significant. That is, one cannot determine whether level of attention problems at 18 months provides more information about (adjusted) language ability at 36 months in PLBW children than in controls.

The within-group relation between the two lagged paths was also distinct between the groups. In the control group, both paths (from attention at 18 months to language at 36 months and from language at 18 months to attention at 36 months) were statistically significant, reflecting equivalent reciprocal influences between attention and language (we tested also the model in a randomly selected subsample of approximately the same size as the PLBW group and both cross-lagged paths remained significant). In the PLBW group, the path from attention problems at 18 months to language ability at 36 months was larger (and statistically significant) than the (non-significant) path from language ability at 18 months to attention problems at 36 months. It seems therefore that the general-deficit-hypothesis [16] might lack some explanatory power in PLBW children. Attention problems might deserve special consideration in this group, instead of being regarded as another manifestation of an underlying general cognitive deficit [19].

Associations between attention problems and language ability, both cross-sectional and cross-lagged, were somewhat low when compared to associations between the same constructs over time. The most robust finding of the study is that early attention problems are the best predictors of later attention problems and that early language ability is the best predictor of later language ability. However, when it comes to reciprocal influences (which are the main focus of this paper) it is noteworthy that, in PLBW children, the latent variable attention problems at 18 months was as good a predictor of adjusted language ability at 36 months as it was of contemporary language ability (18 months). This finding seemed to be unique to the PLBW group and points to modest but non-negligible evidence supporting a precursor role of attention problems in prematurity.

With regards to the PLBW model adjusted for covariates, age corrected for prematurity was an important predictor of language ability at 18 months. Even within this group of children born before 38 weeks of gestation, lower gestational age (reflected in younger age corrected for prematurity) predicted poorer language skills over and above the effect of other covariates such as gender and mother's education. In fact, gestational age has been considered as a better indicator of developmental and neurological maturity than birth weight [41]. Gender was an equally important predictor of language ability at 18 months. Premature boys showed significantly poorer language ability than premature girls, similarly to what has been found in other studies [16]. The most important predictor of attention problems at age 18 months was mothers' educational level. Lower levels of maternal education seemed to predict more attention problems in the child. In fact, maternal education has been used as a marker of environmental risk in prematurity and as a proxy for quality of mother-child interactions and IQ [21].

Preliminary subgroup analyses with covariates were also carried out. Although these analyses were conducted in a reduced sample of children born very premature/with very low birth weight, there was a trend for an increased magnitude of the cross-lagged parameter from attention to language, pointing to a stronger precursor role of attention in severe prematurity. Further research is needed using samples of very premature and very low birth weight children. Some of the children in our "very premature" subgroup had actually birth weight above 1500 g.

In fact, our PLBW group can be regarded as having a relatively low medical risk since it was composed mostly of children born "mildly" premature and with relatively high birth weight. Furthermore, the environmental risk associated with mothers' demographic variables was also reduced. For example, the sample included a large percentage of women with higher education. Although this reflects the educational level in Norway for women in this age range (approximately 50% have higher education, according to Statistics Norway 2009), there was a slight overrepresentation of highly educated women in this sample (60%). With regards to representativeness of the overall MoBa sample, there is underrepresentation of women under 25 years, those living alone, mothers with more than two previous births and with previous stillbirths. Smokers are also underrepresented in the cohort [42]. Reduction of these unfavourable environmental factors might have impacted on the results by further decreasing the risk associated with prematurity in the current sample. It is therefore noteworthy that even in a relatively low-risk sample of PLBW children, we still found a precursor role of attention in relation to language. This effect might be greater in samples with higher levels of biological (e.g., VLBW children) or environmental (e.g., mother's low education) risk.

Some limitations of the study should be pointed out. The mother-report nature of the items used demands caution when comparing these findings with those obtained in observational studies. However, it has been shown that parents can offer accurate reports and constitute a valuable source of information [43,44]. Parents are good at reporting behaviours relevant to the developmental assessment of their children, especially those behaviours that can be observed and do not involve recall of past events. Parent report is problematic when parents have cognitive difficulties or low educational level [44,45]. The items used in our latent variables included assessment of present observable behaviours and our sample was composed of a large percentage of highly educated mothers. Furthermore, the parent-report items used in this study were drawn from well validated instruments. The CBCL is widely used and is considered to have good psychometric properties. In particular, the attention problem items of the CBCL have been able to distinguish referred and non-referred children [31]. There is also an extensive body of literature supporting the reliability and validity of language measures based on parental report [46-49], including studies using the ASQ. Its validity has been extensively demonstrated in samples including both normative and medical risk children (e.g., premature), with high overall agreement between the questionnaires and standardized assessments (88%) [45]. The validity of the ASQ communication scale has also been addressed. This scale is able to identify late language emergence, when using the scale as a whole but also at the single-item level [50]. Furthermore, the validity of the ASQ has been investigated in Norwegian samples. Decreased scores found in premature children have been regarded as supporting the construct validity of the Norwegian version of the scales [51]. Further studies are needed to assess the validity of the specific language and attention items included in the MoBa questionnaires, namely with regards to concurrent validity in relation to standardized assessments.

Another limitation concerns the items used in our language latent variables, which were drawn from a measure designed to be used as a screening instrument of language difficulties. Although we focused mainly on a vulnerable group of PLBW children and looked for correlates of attention problems, the language items included (especially at 36 months) might have constrained the variability of language skills observed, especially for the control group (ceiling effect). However, the very large size of the control group resulted in a reasonable spread of answers across all category responses in most analyses. Nevertheless, a more accurate way of framing our results would be to interpret them as pertaining to relations between attention problems and presence/absence of language delay. In fact, the ASQ high negative predictive value has supported its use as a screening tool in premature children [51].

Another issue concerning item selection must be mentioned. One of the indicators included in the latent variable measuring attention at 18 months is described in the CBCL manual as an overlapping item, present in both attention problems and hyperactivity disorder (see At18, item 3, in Appendix 1). One can argue that, conceptually, this indicator does not reflect "pure" attention problems. The item was nevertheless retained for reasons associated with viability of model building. Subsequent factor analyses revealed strong factor loadings for this item (over .70). In fact, this item was found to be one of the best to discriminate between moderate/severe symptoms and mild symptoms in a sample of clinically referred children, although these were older than the children participating in the current study [52].