Participants
Participants were drawn from a pilot natural history protocol conducted at the NIH Clinical Center in Bethesda, Maryland, USA (protocol number 11-M-0144; NCT01339767). The study was approved by the NIH Institutional Review Board. Consent from the parent or guardian of each child was obtained prior to participation. Inclusion criteria were being between the age of 9 of 21 months, born full-term, no significant medical or motor impairment deemed responsible for the delays (or any known genetic disorder for the typically developing group), and English reported as the primary language spoken in the home. Additionally, at the initial evaluation, eligibility for the language delay (LD) group included receptive and expressive language scores in the very low range (T scores ≤ 30) on the Mullen Scales of Early Learning37 (MSEL), and eligibility for the typically developing (TD) group included scores within1.5 standard deviations (SDs) of the mean on all domains of the MSEL.
A total of 62 (LD, n = 19; TD, n = 43) toddlers enrolled in the natural history study. As described below, the analysis sample for the present study that included analyzable sleep studies was n=55 toddlers (LD, n = 16; TD, n = 39) aged 12 to 22 months at their first visit (mean age 17.4±2.8 months) (Table 1). The sample was majority male (n = 34, 62%) and white (n = 36, 65%). Seven (41%) of the LD participants did receive a research ASD diagnosis (based on clinical judgment of DSM-5 criteria informed by research-reliable administrations of the Autism Diagnostic Interview and Autism Diagnostic Observation Schedule) during their study participation. This small sample size precluded investigation of relationships to an ASD diagnosis.
Table 1. Sample characteristics.
|
Typically Developing (TD)
|
Language Delayed (LD)
|
|
Baseline
|
Follow-up
|
Baseline
|
Follow-up
|
Total N
|
39
|
23
|
17
|
11
|
Male sex, n(%)
|
24 (62%)
|
15 (65%)
|
11 (65%)
|
7 (64%)
|
Age in years, mean (SD)
|
1.43 (0.25)
|
3.10 (0.11)
|
1.65 (0.11)
|
3.19 (0.28)
|
Age range in years
|
1.05 – 1.78
|
2.91 – 3.34
|
1.48 – 1.85
|
2.82 – 3.82
|
Ethnicity, n(%)
|
|
|
|
|
Latino or Hispanic
|
4 (10%)
|
2 (9%)
|
2 (12%)
|
2 (18%)
|
Not Latino or
|
35 (90%)
|
21 (91%)
|
14 (82%)
|
9 (82%)
|
Hispanic
|
|
|
|
|
Unknown
|
0 (0%)
|
0 (0%)
|
1 (6%)
|
0 (0%)
|
Race, n(%)
|
|
|
|
|
Black or African-American
|
3 (8%)
|
2 (9%)
|
6 (35%)
|
4 (36%)
|
Asian
|
2 (5%)
|
1 (4%)
|
1 (6%)
|
0 (0%)
|
White
|
30 (77%)
|
18 (78%)
|
9 (53%)
|
7 (64%)
|
Multiple races
|
4 (10%)
|
2 (9%)
|
1 (6%)
|
0 (0%)
|
Behavioral scores, mean (SD)
|
|
|
|
|
NVDQ
|
115.70 (13.81)
|
113.67 (12.86)
|
88.41 (19.58)
|
85.25 (16.62)
|
VDQ
|
99.84 (10.63)
|
108.60 (12.86)
|
46.77 (16.06)
|
76.64 (23.19)
|
Vineland-II Socialization
|
101.81 (6.05)
|
105.55 (7.19)
|
86.35 (6.13)
|
84.38 (9.81)
|
ASD diagnosis, n(%)
|
|
0 (0%)
|
|
7 (41%)
|
NVDQ = Non-Verbal Developmental Quotient. VDQ = Verbal Developmental Quotient
Procedures
The study design was longitudinal, comprising study visits at 12 and/or 18 months (depending on age of study entry), 24 months, and 36 months. At each study visit, participants received a neurodevelopmental assessment supervised by doctoral level clinicians, including the MSEL and Vineland Adaptive Behavior Scales, 2nd Edition38. Two scores were computed from the MSEL and used in the current analysis:, the Verbal Developmental Quotient (VDQ; using the Receptive and Expressive subscales) and the Nonverbal Developmental Quotient (NVDQ; using the Fine Motor and Visual Reception subscales). DQs are analogs of IQs and are calculated by averaging the age equivalents from each subscale, dividing by the chronological age, and multiplying the result by 100. From the Vineland-II, the Socialization standard score was used to reflect a continuous metric of social skills relative to normative expectations. The Socialization score is norm-referenced with a population mean of 100 and SD of 15.
An optional overnight sleep EEG was performed at the participants’ first visit (baseline; BL) and at the 36-month visit (follow-up; FU); those with valid sleep EEG data were included in the current analysis. A total of 34 (LD, n=11; TD, n=23) participants had both BL and FU sleep studies, while 21 (LD, n=5; TD, n=16) had only BL. Digital EEGs were recorded during awake, drowsy, and sleep states using the 10-20 system of electrode placement (Fig. 1).
Analysis
EEG Coherence. “Clean” ten-minute segments of awake, slow-wave sleep (SWS), and rapid eye movement (REM) sleep were selected for coherence analysis, which was performed masked to participant group using Neuroguide software (Applied Neuroscience Inc., St. Petersburg, FL) using a linked-ear montage.
Select electrode pairs were pulled out for coherence analysis based on findings from Buckley et al.26 which found increased coherence in children with ASD compared to typically developing controls, concentrated in frontal-parietal pairs. Electrode pairs selected for analysis are F3-P3, F4-P4, F4-Pz, P3-Fz, P4-Fz, P3-F7, C3-C4, Cz-Pz, F3-F4, Fz-Cz, and P3-P4 (Fig. 1). Frequency bands selected for coherence analysis are Delta (1-4 Hz), Theta (4-8 Hz), Alpha 1 (8-10 Hz), Alpha 2 (10-12 Hz), Beta 1 (12-15 Hz), Beta 2 (15-18 Hz), and Beta 3 (18-25 Hz).
Coherence values were subjected to Fisher transformation and fixed effects of group (LD versus TD) and age in months (centered at 18 months) were evaluated with the lme439 package for R version 4.2.140 using a linear mixed model with a random subject-level intercept. In separate exploratory analyses, we evaluated the relationship between coherence and behavioral outcomes (MSEL VDQ and NVDQ, and Vineland-II Socialization standard score) by adding the mean-centered behavioral score as a fixed effect. Because these variables are aliased with group membership, we included an interaction between group and behavioral variable, and evaluated the slope of the behavioral variable on coherence within each group to avoid confounding.
Spindle Properties. Both slow spindles (9 Hz and 11 Hz) and fast spindles (13 Hz and 15 Hz) were evaluated separately for density (number of spindles per minute), amplitude, duration, frequency (Hz), and chirp (intra-spindle frequency change). Spindle data were analyzed from all 16 channels using Luna41, a software package developed at Harvard Medical School designed to analyze polysomnogram (PSG) recordings and automate spindle analysis. The Luna software utilized a Morlet wavelet transformation method of spindle detection. Spindle frequency, amplitude, duration and epoch counts were recorded for each spindle and plotted on a graph for visual analysis and verification.
The effect of age on spindle features was evaluated using linear mixed effects models with fixed effects for study visit (BL vs. FU), group, and sex, and a random subject-level intercept. The relationship of spindle density to behavioral scores were evaluated as described above for coherence.
Given the volume of statistical tests performed in this analysis, we used a graphical approach to summarizing results and a p-value threshold of .01 to determine which effects to focus upon for discussion. However, because these were exploratory analyses, we did not adjust p-values for multiplicity. The complete set of uncorrected p-values and test statistics is available as supplementary materials.