|
nif:isString
|
-
This study is part of a longitudinal follow up of a cohort of VLBW (<1500g) infants born 1998–1999 in the southeast region of Sweden [11]. This region has five hospitals with obstetric and pediatric departments, including one level-three university hospital with a regional intensive care unit. The study protocol was approved by the Regional Ethical Review Board of Linköping, Sweden (Registration number 2011–3731), and oral and written informed consent was obtained from the participants and one of their parents. The families were informed that they could withdraw from the study at any time. Perinatal data for both groups were collected from medical registers following parental approval, and reported together with demographics in Table 1. All participants had to be fluent in Swedish. Exclusion criteria included reported concomitant neurological or psychiatric illness, and metal implants that could interfere with the fMRI investigation. All children attended regular schools.
Table data removed from full text. Table identifier and caption: 10.1371/journal.pone.0185571.t001 Perinatal and performance data for the included participants and the original cohort. †n) Data missing for n participants.a) Wechsler Intelligence Scale for Children, for version see annotation per column.b) At age 9: Wechsler Intelligence Scale for Children, WISC-III. (3rd ed.) Swedish manual. Stockholm: Psykologiförlaget, 2002.c) at age 13WISC–IV (4th ed.). Swedish manual. Stockholm: Pearson Assessment and Information AB, 2007. VLBW = very low birth weight (<1500 g), NBW = normal birth weight, n = number, SD = standard deviation, g = gram, ROP = retinopathy of prematurity.
During the intake period (1998–1999), 103 VLBW infants were born in the southeast region of Sweden, of which 93 (90%) survived the neonatal period. At the age of both 7 and 9 years, this cohort was enrolled in prior studies focusing on behavior, cortisol levels, cognition, and reading comprehension; 50 children completed the follow-up at both 7 and 9 years of age [11,28]. Our current follow-up study of children age 12–14 –who are now participating as young adolescents–includes 13 of the 50 children studied at ages 7 and 9; 25 declined, 11 did not respond, and one withdrew. The mean age of the 13 included VLBW was 13.5 years (SD = 0.6). Cognitive assessments were obtained of all 13 participants, fMRI data were collected from 11, and brain volume data from 12 participants. All VLBW participants were right-handed, exclusion from the fMRI analysis was due to excessive head movement or technical problems. The relative high rate of VLBW adolescents that were small for gestational age has been observed in the larger cohort tested at age 7 [11]. Since that cohort showed reading impairments, SGA was tested for correlation with reading variables and no significant correlations were found in the study on 7-year olds [11]. Our current study sample size prohibited further investigation of SGA in this study. We confirmed that our VLBW group was representative for the initial cohort tested at age 7, by performing a statistical comparison on all perinatal variables that were available to us in this study. We performed t-tests on birth weight and gestational age, Mann-Whitney U-tests on ordinal factors of surfactant use, septicemia, and maternal education, and Chi-square tests on SGA, extra low birth weight, respiratory distress syndrome, bronchopulmonary dysplasia, retinopathy of prematurity, intracranial bleeding (collapsed grading), and periventricular leukomalacia. None of these tests showed significant differences in relation to the inclusion factor.
The former studies of children age 7 to 9 included a control group of 51 normal birth weight (NBW, ≥ 2500 g) children, which was selected as follows: for each VLBW participant, a child living in the same municipality was selected from the national birth register for comparison. These NBW children, who were matched for sex and number of siblings, were selected on the basis of being born as close in time to the VLBW child as possible and with no diagnosis in the standard maternity protocol. Of this group, 13 NBW agreed to participate in this current follow-up at age 12–14; seven had declined during earlier studies, 21 declined for the present study, nine did not respond and one child had moved abroad at the time of recruitment and was excluded. The mean age of the included group of 13 NBW was 13.0 years (SD = 0.2). Cognitive assessments and MRI data from all 13 NBW adolescents were included in the analyses, and data from 12 NBW participants were included in the brain volume analysis, one participant having been excluded due to technical problems. One NBW participant was left-handed, however since this individual showed no indication of right-lateralized activation on the fMRI tasks, we found no justification to exclude this participant from the study.
Based on previous observed reading problems [11], a language task based on a Swedish test to screen for dyslexia was developed for this study [29]. This test included a word pair task and a recognition task (the latter will be presented elsewhere). The word pair task comprised three language conditions, namely a phonological choice condition, an orthographic choice condition, a semantic judgment condition, and a line orientation baseline condition. The phonological and orthographic conditions were similar to the tasks that previously have been used to characterize reading problems in VLBW [10,11]. The tasks were presented through rear projection on a screen behind the head of the participants. The screen was visible to the participants through a mirror mounted on the head coil. The word pair task consisting of the described four language choice conditions was presented according to a block design (Fig 1). Each condition block started with a specific question presented to the participants for five seconds. Next, a word pair was shown; one of the words was the correct answer to the question, all real words were proper nouns. The participants answered by pressing one out of two buttons on a response box (Lumina LU444-RH, Cedrus Corporation, San Pedro, U.S.A.) with their index finger (for the word presented at the left of the screen) or their middle finger (for the word presented at the right) of their right hand. The mean duration of word pair presentation was five seconds, but varied from three to eight seconds according to a Poisson distribution. There were five word pairs per block, all relating to the initial question. The semantic condition asked for word categories, each semantic block asked about a different category. The orthographic condition investigated spelling with a word pair of a correctly and an incorrectly spelled word. The phonological condition asked "Which word sounds correct?” after which pairs of pseudo-words were presented. One of the pseudo-words was a real word when sounded out, while the other was not. The line orientation condition served as a baseline condition and showed two strings of "x", in one string one "x" was replaced by an "y".
Figure data removed from full text. Figure identifier and caption: 10.1371/journal.pone.0185571.g001 Schematic overview of the word pair task used during the fMRI session.The different blocks for each language choice condition are shown, the example block shows the timing in seconds for the blocks, and the sequence of 5 word pairs after each question. Each new stimulus was separated from the previous by a short 100 ms fixation break, while a five second pause separated the condition blocks from each other. After presenting four condition blocks (semantic, orthographic, phonological, and baseline), there was a 10 second break, followed by a new set of the four condition blocks presented in a different order and with different words (and a different category for the semantic block). In all, four condition blocks were presented four times, resulting in a total of 20 pairs of words for each condition, to make a grand total of 80 presented word pairs. To become familiar with the task, all participants completed an off-line training session prior to the fMRI session. The training session words were different from those presented during the scanning procedure.
All examinations were conducted at the Center for Medical Image Science and Visualization (CMIV, Linköping University), located at Linköping University Hospital, Sweden. Images were acquired using a Philips Ingenia 3.0 Tesla scanner with a standard head coil, which was halfway the study replaced by a head-neck coil due to technical complications, this affected both the VLBW and the NBW group equally. Structural and functional brain images were obtained during a scanning session of approximately 40 minutes. The functional images were compiled using a single-shot gradient-echo echo planar image (EPI) sequence sensitized to T2* blood oxygen level dependent (BOLD) signal changes. Whole brain coverage was obtained with 34 or 35 slices with a 0.3 mm slice gap (10% voxel size). Repetition time = 2 s, time to echo = 35 ms, flip angle = 75°, voxel size = 3x3x3 mm3, acquisition matrix = 72x68 voxels, number of dynamics = 248. The slices were aligned between the floor of the sella turcica and the posterior angle of the fourth ventricle. The structural images were collected with T1-weighted sequences covering the whole brain with the following parameters: repetition time = 7.5 ms, time to echo = 3.5 ms, flip angle = 8°, voxel size = 1.1x1.1x0.6 mm3, acquisition matrix 228x227 voxels, with 301 slices obtained.
Preprocessing and statistical analysis were performed in SPM8 (www.fil.ion.ucl.ac.uk/spm/software/spm8). The DARTEL toolbox was used for preprocessing [30], with default settings for EPI data as outlined in the SPM8 manual (www.fil.ion.ucl.ac.uk/spm/doc/spm8_manual.pdf). We used DARTEL to create a template of extracted grey and white matter from the anatomical images of all participants in this study (both VLBW and control group) with the default number of 6 outer iterations. The template was then normalized within the DARTEL toolbox to a 2x2x2 mm3 Montreal Neurological Institute (MNI) template and smoothed with an 8 mm full width half maximum (FWHM) Gaussian kernel. In SPM, the individual realigned and resliced EPI data from the fMRI tasks were coregistered and then normalized to 2x2x2 mm3 MNI space with help of the normalized DARTEL template and smoothed with an 8 mm FWHM Gaussian kernel.
Components of intellectual capacity was assessed on the same day prior to the fMRI session, using Vocabulary and Block Design subtests from the Wechsler Intelligence Scale of Children (WISC-IV, Wechsler 2003). WISC Vocabulary asks for word definitions, and measures verbal comprehension. WISC Block Design measures visuo-spatial processing by asking to recreate patterns with colored blocks. Furthermore, the parents of each participant were asked to fill in a questionnaire regarding their child’s health status, family constellation and school situation. The participant data, cognitive measures from the WISC tests, and fMRI performance data can be found in S1 Table.
The individual preprocessed images were analyzed for brain activation related to sentence processing by applying contrasts between the language conditions and the line orientation baseline. This resulted in three contrast images per participant for semantic > baseline, orthographic > baseline, and phonological > baseline. Head movement parameters were filtered out by entering those into the analysis as multiple regressors of no interest. To qualify as an exclusion criterion, excessive head movement was defined as > 3 mm in-plane motion; since this interfered with successful normalization. Group activation for all participants per language condition in the whole brain at a significance threshold of 0.001 uncorrected was used for multiple regression (Fig 2). Performance on the fMRI tasks was measured in terms of accuracy and reaction time. Accuracy was defined as the hit rate on the answers, which was calculated by subtracting the number of incorrect from correct answers for each condition and each participant.
Figure data removed from full text. Figure identifier and caption: 10.1371/journal.pone.0185571.g002 Neural activation (warm/orange) and deactivation (cool/blue) during the three fMRI language tasks (semantic, orthographic, and phonological processing) for all participants.The locations of the transversal sections are shown in the sagittal midline section on the right. Locations of transversal section are identical for all tasks. For visualization purposes to show the extent of the activation, these images are thresholded at p = 0.001 uncorrected. L = left hemisphere, R = right hemisphere.
Between-group differences were investigated per language condition with three multiple regression analyses, each with the inclusion of the following covariates: age, sex, accuracy on fMRI tasks (per condition), WISC Vocabulary Raw Scores and WISC Block Design Raw Scores. Age and sex were included to control for deviation from the original matched groups in the first study (at age 7), as this study was not aimed towards investigating age effects or sex differences. Accuracy on fMRI tasks, WISC Vocabulary Raw Scores and WISC Block Design Raw Scores are considered to be potential correlates to language performance, and therefore covariates of interest. All covariates were centered around the overall mean of that covariate. We investigated group and performance effects in 12 pre-defined regions of interest, six in each hemisphere, that we hypothesized to be most likely to show main effects. This assumption was based on the involvement of these regions in reading [22,31] and previous results observed in studies on language functioning in preterm children and adolescents [24,25]. The pre-defined regions were based on the Talairach Daemon labels atlas in the WFU pickatlas [32,33]. The regions were as follows: Inferior Frontal Gyrus, Middle Temporal Gyrus, Superior Temporal Gyrus, Angular Gyrus, Supramarginal Gyrus and Fusiform Gyrus in the left and the right brain hemisphere respectively. F-tests were applied to investigate the main group effect, any confounding main effects of age or sex, and main effects of the covariates of interest; accuracy on fMRI tasks, WISC Vocabulary Raw Scores and Block Design Raw Scores. Post-hoc t-tests were calculated to investigate direction of activation that was significant under the F-tests in our regions of interest, and to investigate interactions between group and the covariates of interest. All between-group F- and t-tests were thresholded per pre-defined region of interest at a peak-level significance threshold of p < 0.05, FWE corrected, with a minimum cluster extent of 10 voxels. Only for the interaction effects non-corrected post-hoc t-tests per group were made with age and the effect of interest as covariate, to verify the direction of the interaction. The remaining statistical analyses were carried out using IBM SPSS Statistics 22.0. To test the hypothesis that the VLBW group had lower performance on the fMRI tasks, in the form of lower accuracy and slower reaction times, we applied a multivariate ANOVA including either of these two variables on all four different conditions (baseline, semantic, orthographic, and phonological) and tested between-group differences. The hit rate data underlying the accuracy on fMRI tasks measurement showed large variance, which necessitated a transformation of the data. A transformation of 1/(1.1—hit rate) showed to be the most optimal in reducing variance. For the 2-independent samples tests comparing WISC Block Design Raw Scores and WISC Vocabulary Raw Scores between groups, we opted to use the non-parametric Mann-Whitney U-test, since the small sample sizes per group (minimum of 11) makes it difficult to meet the assumption of normal distribution, and a significance threshold of p < 0.05 (one-tailed) was applied. Maternal education was tested with the Mann-Whitney U-test as well, since the data was categorical (categories: Elementary education, Gymnasium, Higher education).
|
![[This page as RDF]](http://data.gesis.org/somesci/static/rdf-icon.gif){kind=icon}