Clinical utilisation of the Infant Monitor of vocal Production (IMP) for early identification of communication impairment in young infants at-risk of cerebral palsy: a prospective cohort study

ABSTRACT

Aim

To report prospective longitudinal data of early vocaliszations of infants identified “at-risk” of cerebral palsy (CP) for early identification of communication impairment.

Method

This case-control longitudinal prospective cohort study reports on the assessment of 36 infants, 18 identified as at-risk of CP at the time of enrolment and 18 typically developing (TD) children, at three time points: 6 months, 9 months and 12 months of age, Data were obtained through criterion and norm referenced assessments of vocaliszation behaviors.

Results

Early vocal behaviors of infants identified as at-risk of CP did not differ from their age matched peers at 6 months of age, however, significant group differences emerged at 9 and 12 months when pre-canonical and canonical babble typically emerge. Generalized linear mixed models analysis showed that the rate of development of early language ability and more complex speech-related vocal behaviors was slower for infants at risk of CP when compared to TD infants, with over 75% of infants with CP showing below normal vocal production and impaired language by 12 months of age.

Interpretation

Our data suggest characteristics of infant vocalizations associated with pre-canonical and canonical babbling provide a strong evidence base for predicting communication outcomes in infants at risk of CP.

KEYWORDS:

• Cerebral palsy
• infant vocalizations
• Infant Monitor of vocal Production
• early identification
• communication impairment

Introduction

Communication impairments associated with cerebral palsy (CP) are typically chronic in nature,¹ and further contribute to activity and participation limitations, including restricted ability to meet basic needs, form friendships, and achieve educational success.^2–5 Studies show 50–85% of children with CP experience communication impairments that may arise from a range of difficulties.^4–12 Speech motor impairment or dysarthria is the most frequently characterized co-occurring communication impairment with a reported prevalence range from 36%¹¹ to 90%.¹³ Whilst a number of risk factors have been identified, including the timing of the brain insult, type of CP, level of gross motor function and epilepsy,^{2,6,7,12,14–20} the interaction between these risk factors is not understood and, as yet, no early biomarkers for communication impairment have been identified.¹³

Early identification is critical to accessing early intervention services focused on optimizing the child’s communicative ability and minimizing the negative long-term consequences of impairment.²¹ It is well documented that the first 12 months of life represent a critical window of heightened plasticity to language learning, modulated by the interplay between maturational and experiential influences.^22–25 Yet, there is a paucity of data evaluating the speech and language profiles of children at risk of CP younger than 2 years²⁶ with research limited largely to retrospective file audits²⁷ and registry data⁴ without detailed speech and language measures, preventing benchmarking to peers. Our understanding of communication impairment associated with CP has, therefore, been based primarily on profiles of children ≥2 years of age.^5,8,9,28,29

One reason for the lack of data for <2 years of age is that historically CP has been diagnosed between 12 and 24 months of age.³⁰ However, in 2017, international clinical practice guidelines^31,32 have enabled accurate early identification of CP before 6 months, and as early as 3 months corrected age, using a combination of standardized measures of early motor ability with predictive capabilities, such as the General Movement Assessment,³³ and clinical reasoning. In contrast, similar tools that facilitate early identification of communication impairment in infants at risk of CP have yet to be identified.³⁴

The characteristics of infant vocalizations provide the strongest evidence base for predicting communication outcomes in infants, including motor speech impairment. Babbling is one of the most studied stages of early speech development.³⁵ The screening of infant vocalizations was initially proposed by Oller³⁶ as a tool to “facilitate prevention or amelioration of important communicative disabilities” (p. 240). This is predicated on evidence showing the progression to speech-like vocalizations is a milestone of speech motor development³⁷ and a fundamental component of language learning,³⁸ and the quality of the vocal interactions between an infant and the caregiver is predictive of verbal capacity as well as other social and cognitive behaviors later in development.^37,39

Infant vocalizations follow a trajectory of increasing complexity, from early vegetative sounds (e.g., crying, coughing) to pre-canonical babble, such as quasi-resonant sounds, raspberries, and marginal babble, with canonical babble marking the transition to first words.⁴⁰ Canonical babble is defined as a “consonant-vowel-syllable with rapid transition between the consonant and vowel”⁴¹(p112). It is reported to arise from the rhythmical oscillations of the jaw and requires motor control and coordination.^42,43 Canonical babble has been identified as a developmental phenomenon that is typically established by 10 months and is consistent across multiple languages⁴¹ and social environments.⁴⁴

The predictive value of early infant vocalizations is evident in studies investigating children with language delay,^36,41 childhood apraxia of speech⁴⁵ and hearing impairment.⁴⁶ Differing developmental trajectories in infant vocalizations have also been reported in infants with neurodevelopmental conditions, such as autism spectrum disorder,⁴⁷ Down syndrome,⁴⁸ and Rett syndrome.⁴⁹ Little research, however, has investigated infant vocalizations in children with CP. Levin⁵⁰ reported qualitative and quantitative differences in the babbling of eight children with CP at 12 months of age, including delayed onset of canonical babbling and predominantly monosyllabic vocalizations, suggesting further research into infant vocalizations as a marker of early speech impairment warrants attention.

Although research has identified the value of infant vocalizations as a marker of early communication impairment, methods for analyzing or coding infant vocalizations in research (i.e., phonetic transcription, acoustic analysis, calculation of babble ratios) are generally time-consuming and resource intensive,⁵¹ thus hampering clinical utility.⁵² In contrast, the Infant Monitor of vocal Production (IMP), which takes around 15 minutes to administer via interview with the parent, is a criterion referenced clinical tool to prospectively evaluate and identify “any departure from the typical birth-12 month continuum of prelinguistic vocal behavior”.⁵³ The IMP was initially developed for use with children with hearing loss but has been used more broadly since.⁵³ The sensitivity of the IMP in identifying communication impairment in prelinguistic children at risk of CP and in comparison to standard tests of emerging receptive and expressive language has not yet been investigated.

The purpose of this study was, therefore, to report on the prospective longitudinal assessment of early vocal behaviors of infants who were identified “at risk” of CP, as compared with age matched typically developing (TD) infants, using the IMP. It was expected that the vocal behaviors assessed by the IMP signaling the emergence of canonical babble, which reflect an important milestone in the development of speech motor control, would be sensitive to infants with oro-motor challenges and therefore differentiate the at-risk of CP and TD groups. We also predicted that infants at-risk of CP would show poorer emergence of early language ability using standardized norm-referenced assessments, such as the Receptive Emergent Language Test Version 3 (REEL-3). The REEL-3 is a psychometrically sound tool that provides receptive, expressive, and combined language ability standard scores and percentiles in young children.⁵⁴ Furthermore, given the high prevalence of speech motor impairment in children with CP, and the impact that speech impairment has on language and communication generally, we expected a significant departure (i.e., delay) from the normal trajectory of infant vocalization development, as measured by the IMP, would identify children at risk of CP with communication impairment as assessed by the REEL-3. In summary, the following research questions were posed:

1. Do the vocal behaviors of children at-risk of CP differ to TD developing children at 6, 9 and 12 months of age, as measured on the IMP?

   1. 1.1 At what age are group differences in vocal behaviors observed?

   2. 1.2 Do items of the IMP assessing pre-canonical and canonical babble, in particular, show group differences consistent with early signs of speech motor impairment?

• (2) Is there a difference in the emergent language abilities at 6, 9, and 12 months of age of children with/at risk of CP using an established standardized, norm-referenced assessment (i.e., the REEL-3) when compared to TD infants?

• (3) Finally, does the IMP show high levels of sensitivity and specificity in identifying communication impairment in children at risk of CP using the language ability quotient of the REEL-3?

This research was undertaken not only to address an important clinical need in the provision of data on the early development of communication abilities in infants at risk of CP, but also to contribute to our understanding of the role of prelinguistic vocal development in the identification of emerging communication impairment in infants at risk of CP.

Method

Study Design

A case–control prospective longitudinal cohort study design was utilized. Approval for this study was obtained from the Child and Adolescent Health Services Ethics Committee (study number 2015221).

Setting

Participants were recruited as part of a larger longitudinal study evaluating the development of communication in children aged 6 months to 2 years through the early intervention service of Perth Children’s Hospital (PCH). This is the state-wide tertiary rehabilitation service in Western Australia. In 2015, the service was extended to implement a specialized service focused on early detection and intervention for young babies identified “at risk” of CP, and their families. This tertiary service was designed to provide a multidisciplinary assessment service and to provide interim early intervention services whilst facilitating timely referral to community-based services. In addition, clinical research trials were embedded within the service to provide access to the most up-to-date CP-specific interventions as they evolve. The tertiary service is framed within family-centered practice principles with a focus on supporting transition into community services and adheres to recommendations for early diagnosis.³²

Participants

The participants reported in this study consist of 36 infants: 18 identified as “at risk” of CP at the time of enrollment into the study and 18 TD children, recruited between March 2016 and April 2019.

Infants at risk of CP. Children identified as “at-risk” of CP were recruited through Western Australia’s state-wide tertiary early intervention service for infants at-risk of CP. The inclusion criteria were (a) infants considered at high risk of CP based on abnormal MRI findings or General Movement Assessment or medical history that placed the infant at high risk, as determined by the medical consultant (e.g., neonatal encephalopathy, neonatal stroke, hypoxic-ischemic encephalopathy)^55,56; (b) referral and acceptance into the early intervention service before 6 months of age; and (c) completion of the Receptive-Expressive Emergent Language Test – third edition (REEL-3) and IMP just prior to 6 months of age by the managing speech-language pathologist within the early intervention service.

Infants were excluded if they lived in rural/remote locations that precluded participation in the three assessment time points within ± 7 days of each time-point, or if they were deemed medically unstable.

All participants underwent the newborn hearing screening protocol, and Hearing Australia’s targeted surveillance and pathway to review by 1^st birthday, for those identified at-risk, as per the Operational Directive of the Newborn Hearing Screening WA Health.⁵⁷ Follow-up and monitoring of any neonatal diagnosis of hearing loss were documented by the national government service provider, Hearing Australia.

Typically developing children. Children were recruited from the Perth metropolitan region through advertising or word of mouth. Criteria for inclusion were (a) age-appropriate performance on the 4 month Ages and Stages Questionnaire,⁵⁸ completed by the Child Health Nurse; (b) no reported history of speech, language, or learning problems, and (c) passed their newborn hearing screening.

The functional hearing status (ear health) of all participants (TD and at-risk) was monitored through clinical observation and parent report at each assessment point. No infant with an uncorrected hearing loss <25 dB presented to the study.

Materials

Infant Monitor of vocal Production (IMP). The IMP is a criterion-referenced instrument containing 16 probe questions that target vocal behaviors of increasing complexity, from early reflexive behaviors and capacity to control phonation, to the emergence of canonical babbling and first word representations. Using a semi-structured interview format with the parent as the informed observer, the child’s level of achievement in relation to target vocal behaviors (e.g., combining different sounds together, babbles fluently and rhythmically) is discussed and scored using a 5-point Likert scale (Never = 0, Rarely = 1, Sometimes = 2, Often = 3, Always = 4). The primary outcome measure used in the present study is the question ceiling score (0–16), which is the highest IMP item with a non-zero rating before two zero ratings in succession are obtained (after which no further items are presented) or, typically for children at or close to 12 months of age, all items are presented. The question ceiling reflects the child’s current stage of vocal achievement. The item ratings were also used to generate a secondary outcome measure, the percentage ranking score. This is a quotient score based on cumulative item ratings expressed as a percentage of the ceiling total (i.e., maximum cumulative score possible given the ceiling obtained, i.e., ceiling score multiplied by 4) combined with a variability score, which is an extra 1% added for each different type of CV pattern produced, as reported by the parent. The percentage ranking is a measure of the infant’s robustness or degree of achievement at their current stage of vocal development.

The IMP has established psychometric properties⁵³ showing strong inter-rater reliability and correlation for agreement (0.94). The ceiling score has been shown to be sensitive to age-related changes in vocal behaviors in TD infants up to 12 months of age, as well as variations in progress toward speech in infants with hearing impairment and the impact of interventions for hearing impairment.^53,59 The IMP has also been shown to be sensitive to delays in vocal development in infants with oro-motor deficits.⁵³

Receptive-Expressive Emergent Language Test Version 3 (REEL-3). The REEL-3⁵⁴ is a standardized assessment of emerging language in children from birth to 3 years of age. Information is obtained through parent interview. Raw scores from the receptive and expressive language scales are converted to standard ability scores (M = 100, SD = 15) and percentile ranks, with three scores calculated: receptive language ability, expressive language ability, and combined language ability score. The REEL-3 has been identified as a reference standard for early language assessment,⁶⁰ with established psychometric properties.⁶¹ In addition, Rome-Flanders and Cronk⁶² report longitudinal stability, with predictive validity of later testing results at 15 months and 18 months.

Procedure

Following enrollment in the study, data were collected within the home environment. The first assessment took place on average 7 days prior to each infant turning 6 months of age and continued at three monthly intervals for a total of three testing occasions. For convenience, the first time point is labeled as 6 months of age. These interviews, conducted by an experienced speech-language pathologist trained in the administration of the IMP, were completed by the same parent for all testing occasions and in the presence of the infant. A research assistant, blinded to the aims of the study, prepared the data for analysis.

Analyses

To address Research Questions 1 and 2, the IMP question ceiling and percentage ranking scores, treated as continuous data, and REEL-3 standard scores, were analyzed separately using generalized linear mixed models (GLMM) implemented through the SPSS Version 26 GENLINMIXED procedure. GLMM was used with the robust statistics option and has the advantage of accommodating variables with normal and non-normal distributions and missing cases without excluding participants or imputing data. Group status (at-risk vs. TD) and time point (6, 9 and 12 months), as well as the interaction between group and time point, were tested as fixed effects, and with participants treated as a random effect. Follow-up analyses were undertaken using sequential Bonferroni corrected pairwise contrasts. The effect size reported is partial eta squared (ηp2) from the corresponding General Linear Model analysis in SPSS, and the alpha level used throughout is .05.

A dimension reduction approach using principle component analysis (PCA) with Direct Oblimin factor rotation was undertaken to quantify performance on different aspects or dimensions of vocal development as measured by the items of the IMP. Combining item scores from both groups, the underlying dimensional structure (i.e., factors) with eigen values greater than 1, reflecting shared item variance for the sample data, was determined.⁶³ The suitability of the data for PCA was also examined (details presented below). To address Research Question 1.2, participant regression factor scores were obtained from each extracted factor. Independent samples t-tests of factor scores were used to determine which dimensional components of the IMP are sensitive to group differences. Dimensions were interpreted based on items with only high loadings (> .80) on each extracted factor, given the small sample size for PCA.⁶³

Regarding Research Question 3, in order to determine the sensitivity and specificity of the IMP, the modified t-test for comparing single-cases to control samples⁶⁴ was used to identify children at risk of CP with a significantly poorer IMP question ceiling score than the age-matched TD sample. Following Hustad et al.,⁸ a REEL-3 language ability score equivalent to the 10th percentile was used as the cut point to identify children in the at-risk group with impaired language.

Bias

Participants identified “at risk of CP” were recruited from the state-wide pediatric tertiary service of Western Australia. It is the only state-wide center for the management of motor disorders in young infants and children with CP. At the time of this study, two additional multisite research trials were in process within this clinical population and age group. A database audit identified that just over 50% of participants recruited to this study were recruited in 2018 and this represented approximately 30% of eligible infants, with the study sample biased to children with bilateral CP. The constraints of the recruitment context were also a primary determinant of sample size. However, a sensitivity analysis based on conventional analysis of variance⁶⁵ shows the current sample size of 36, power level of .80 and alpha = .05, biases large between group effect sizes (η = .14) but medium effect sizes (η = .06) or larger for time-related repeated measures factors, including the group-by-time interaction. These effect sizes are deemed clinically significant when the focus is to identify differences associated with impaired communication development.

The managing speech-language pathologists within the early intervention service conducted the initial assessments and facilitated recruitment to the study. The first author who has 30 years clinical experience (RW) completed all subsequent assessment time points, ensuring the continuity of data collection.

All children were followed for the three time points reported in this study, with one datapoint missing for one child (CP0007, at 12 months).

Results

Table 1 provides a summary of the key participant characteristics of the infants identified as at-risk of CP, at the time of enrollment in the study. All 18 participants identified at-risk presented with abnormal MRI findings. Classification using the MRI Classification System (MRICS) developed by the Surveillance of Cerebral Palsy in Europe (SCPE) Working Group⁶⁶ shows 61% presented with predominantly gray matter injury and 33% with white matter predominate injury. One participant presented with maldevelopment. At 24 months of age, 15/18 participants received a diagnosis of CP and 3/18 a diagnosis of developmental delay and not CP.

Table 1. Participant Characteristics

Participant Characteristics	At-risk/with CP	18 (%)	TD 18 (%)
Gender	M	9 (50)	13 (72)
	F	9 (50)	5 (28)
Gestational Age	< 28 weeks	3 (17)	-
	28 – < 32 weeks	2 (11)	-
	32 – < 37 weeks	4 (22)	2 (11)
	38–40 weeks	9 (50)	16 (89)
CP motor type	Dyskinetic	9 (50)	-
	Spastic	6 (33)	-
	n/a	3 (17)	-
Distribution	Unilateral	1(6)	-
	Bilateral	14(78)	-
	n/a	3(17)	-
MRI	Timing of Insult
	Anti	3 (17)	-
	Post	1 (5)	-
	Peri	14 (78)	-
	Description
	Grey	11 (61)	-
	White	6 (33)	-
	Maldevelopment	1 (6)
GMFCS (as at 2 years)	GMFCS 1	3 (17)	-
	GMFCS 2	0 (0)	-
	GMFCS 3	2 (11)	-
	GMFCS 4	2 (11)	-
	GMFCS 5	6 (33)	-
	n/a	3 (17)	-
	missing	2 (11)	-
Hearing Impairment	*WNL	15 (83)	15 (83)
	Aided (sensorineural loss)	2 (11)	-
	Episodes of Otitis Media	1 (6)	3 (17)
Oropharyngeal Dysphagia		10 (55)	-
	Feeding Method		-
	Oral	12 (67)	18(100)
	PEG	6 (33)	-

Note: * based on parent report, with one awaiting outcome of hearing assessment. PEG = percutaneous endoscopic gastrostomy tube

Of the 15 participants diagnosed with CP, nine presented with dyskinetic CP and six with spastic CP. Four of five levels of the Gross Motor Classification System ^GMFCS,67 were represented: I (n = 3), III (n = 2), IV (n = 2), and V (n = 6).

Question 1. Do the vocal behaviors of children at-risk of CP differ to TD developing children at 6, 9 and 12 months of age, as measured on the IMP?

Table 2 provides the IMP raw question ceiling scores on each of the three testing occasions for each of the participants, and Figure 1 shows the mean question ceiling score with 95% CI for the at-risk and TD groups at 6, 9 and 12 months of age. GLMM analysis showed a significant effect of group, F(1, 102) = 22.19, p < .001, ηp² = .38, time, F(2, 102) = 159.73, p < .001, ηp² = .78, and interaction between group and time, F(2, 102) = 27.50, p < .001, ηp² = .40. As shown in Figure 1, there was a significant increase in question ceiling scores across time for both groups reflecting age-related achievement in vocal behaviors; however, the rate of vocal development was lower for the at-risk group. While the two groups did not differ at 6 months, t(102) = 1.00, p = .319, ceiling scores were significantly lower for at-risk infants at both 9, t(102) = 4.18, p < .001, and 12, t(102) = 6.34, p < .001, months of age.

Table 2. IMP question ceiling item at <6, 9 and 12 months of age

	At-risk/CP (months)				TD (months)
ID	Diagnosis	< 6	9	12	ID	< 6	9	12
CP1	Spastic	8	10	10	TD001	8	13	16
CP2	Spastic	11	13	16	TD002	8	13	16
CP3	Dyskinetic	7	7	9	TD003	11	13	15
CP4	Dyskinetic	8	8	8	TD004	10	10	14
CP5	Dyskinetic	8	8	10	TD005	8	14	16
CP6	Dyskinetic	10	10	10	TD006	8	15	16
CP7	Dyskinetic	7	10	10	TD007	11	13	16
CP010	Dyskinetic	7	9	11	TD008	9	13	14
CP011	Dyskinetic	8	7	8	TD009	9	13	16
CP012	Spastic	7	7	7	TD010	9	13	14
CP013	Dyskinetic	7	7	10	TD011	10	10	14
CP014	Spastic	8	9	12	TD012	9	16	16
CP0015	DD	10	11	13	TD013	9	14	16
CP0016	Spastic	13	15	16	TD014	9	13	16
CP0017	Dyskinetic	10	10	10	TD015	9	13	16
CP0018	DD	8	13	13	TD016	7	13	15
CP0019	DD	11	13	13	TD017	11	11	16
CP0020	Spastic	11	13	16	TD018	11	11	16

Figure 1. Mean question ceiling score from the Infant Monitor of vocal Production at 6, 9 and 12 months of age for infants at risk of cerebral palsy (CP) and typically developing (TD) infants (with 95% CI).

Figure 2 shows the mean percentage ranking score with 95% CI for the children at-risk of CP and TD children for each time point. The GLMM showed a significant effect of group, F(1, 102) = 9.47, p = .003, ηp² = .21, and time, F(2, 102) = 4.842, p = .010, ηp² = .13. Infants with CP had a lower percentage rank overall than the TD infants. Although the group difference was smaller at 6 months, and the TD infants increased in their mean percentage ranks with age to a greater extent than the CP infants, the interaction between group and time was not statistically significant, F(2, 102) = 2.50, p = .087, ηp² = .07.

Question 1.2. Do items of the IMP show group differences consistent with early signs of speech motor impairment?

Figure 2. Mean percentage rank score from the Infant Monitor of vocal Production at 6, 9 and 12 months of age for infants at risk of cerebral palsy (CP) and typically developing (TD) infants (with 95% CI).

PCA was conducted on item scores separately for the 9 and 12 months of age data to determine the dimensional structure, that is, patterns of shared item variance, and to see which dimensions were associated with group differences by analyzing corresponding factor scores. At 9 months, after excluding 7 out of 16 items due to either lack of variability (i.e., later achieved items with consistent ratings of 0, items 12, 14, 15 and 16), multicollinearity (r > .80, item 11) or low communalities (< .6, items 6 and 7), a two-factor structure was obtained explaining 69% of the variance (see item loadings from the Pattern Matrix in Table 3). Although the sample size was small, the data were deemed suitable for PCA given Bartlett’s Test of Sphericity was statistically significant, χ2(36) = 173.53, p < .001, the determinant was sufficiently high (.004), and the Kaiser-Meyer-Olkin measure of sampling accuracy was high (.84) suggesting distinct factors in the solution.⁶³ Items loading onto the primary Factor 1, which accounted for over 55% of the total variance, suggest this factor relates to the achievement of pre-canonical babble (e.g., makes closed-mouth “raspberry”-like sounds, uses different sounds together) and canonical babbling (e.g., babbles fluently and rhythmically) and was, therefore, interpreted to be a factor associated with the early development of speech motor control. The two items loading onto Factor 2 suggest a social-receptive dimension associated with phonation control and social interaction with voice. Factor 1 scores were significantly higher for the TD group than the at-risk of CP group, t(27.80) = 5.89, p < .001, suggesting the groups differed in terms of their development of speech motor control with poorer achievement for infants at-risk of CP. There was no difference in Factor 2 scores between the two groups, t(34) = 0.59, p = .56.

Table 3. Factor Loadings for Extracted Dimensions from the Principle Components Analysis of IMP Item Scores at 9 Months of Age (N = 36)

Item	Description	Factor 1	Factor 2
		Speech Motor	Social Receptive
1	A pleasing voice	.639	.322
2	Looks intently at the speaker’s face	.544	.429
3	Makes expressive sounds other than crying	.616	.463
4	Uses voice to respond to an unseen speaker	.084	.813
5	Changes of volume of voice (soft-loud)	−.120	.867
8	Makes closed-mouth “raspberry”-like sounds	.862	−.061
9	Uses voice while mouthing objects	.881	−.139
10	Uses different sounds together	.847	−.083
13	Babbles fluently and rhythmically	.800	.020

Note. Factor 1 explained 55.1%, and Factor 2 14.5%, of total variance among item scores. Loadings in bold are greater than .80 and are taken as significant for dimension interpretation.

The PCA at 12 months replicated in part the two dimensions extracted from the PCA at 9 months, plus a third dimension that was defined by only one high loading item (.93, item 7, makes open-mouth vowel-like sounds/ah/). The solution we report excluded items due to low variability (item 15), multicollinearity (items 11, 14 and 16), and low communality (item 6), as well as item 7. Table 4 shows the factor loadings from the Pattern Matrix for the resulting two-factor solution, which explained 69% of total variance. The data were suitable for PCA given the significant Bartlett’s test of sphericity, χ2(36) = 202.58, p < .001, satisfactory Kaiser-Meyer-Olkin measure of sampling adequacy (.76), and sufficiently large determinant (.001). Factor 2 (social-receptive dimension) appears to be reliable given that the same two items loaded highly on Factor 2 in the 9 month PCA. The primary Factor 1, accounting for 53% of total variance, was similar to the speech motor factor from the 9 month PCA, sharing two high loading items (makes closed-mouth “raspberry”-like sounds, babbles fluently and rhythmically). The third item (copies a familiar sound or word [i.e., moo]) also reflects the emergence of canonical babbling and, therefore, development of speech motor control. The at-risk of CP group was significantly lower on the speech motor dimension scores at 12 months, t(27.842) = 6.90, p < .001, but there was no group difference in social receptive scores at 12 months, t(34) = 1.33, p = .19, findings that replicate the pattern of group differences observed in the 9 month data.

Question 2. Is there a difference in the emergent language abilities of children with/at risk of CP using the REEL-3 when compared to TD infants

Table 4. Factor Loadings for Extracted Dimensions from the Principle Components Analysis of IMP Item Scores at 12 Months of Age (N = 36)

Item	Description	Factor 1	Factor 2
		Speech Motor	Social Receptive
1	A pleasing voice	.527	.521
2	Looks intently at the speaker’s face	.367	.628
3	Makes expressive sounds other than crying	.464	.626
4	Uses voice to respond to an unseen speaker	.032	.830
5	Changes of volume of voice (soft-loud)	−.230	.857
8	Makes closed-mouth “raspberry”-like sounds	.827	.075
9	Uses voice while mouthing objects	.730	.153
10	Uses different sounds together	.758	−.032
12	Copies a familiar sound or word (i.e., moo)	.828	−.161
13	Babbles fluently and rhythmically	.868	−.024

Note. Factor 1 explained 52.9%, and Factor 2 15.9%, of total variance among included item scores. Loadings in bold are greater than .80 and are taken as significant for dimension interpretation.

Table 5 provides the expressive language, receptive language, and language ability standard score for each testing occasion for each participant, and Figure 3 shows the mean REEL-3 combined language ability score at 6, 9 and 12 month time points for infants at risk of CP and TD infants. Although both groups show similar levels of early emergent language ability at 6 months, there is a significant and widening difference between the two groups at 9 and 12 months of age. GLMM analysis showed a significant effect of group, F(1, 101) = 22.45, p < .001, ηp² = .39, and time, F(2, 101) = 15.03, p < .001, ηp² = .30, as well as an interaction between group and time, F(2, 101) = 32.06, p < .001, ηp² = .46. There is a notable decline over time in language ability for at-risk infants when compared to their peers. The pattern seen in Figure 3 was replicated for both the REEL-3 expressive and receptive language scores showing the group differences were not specific just to the expressive or receptive modality.

Table 5. Receptive-Expressive Emergent Language Test Score (REEL-3) for all Participants Across Three Testing Occasions

	6 months				9 months				12 months
ID	EAS	RAS	LAS		EAS	RAS	LAS		EAS	RAS	LAS
At-risk of cerebral palsy
CP0001	97	93	94		88	90	87		79	91	82
CP0002	95	76	83		93	86	87		80	85	79
CP0003	76	75	71		64	60	54		70	68	63
CP0004	70	61	59		65	<55	52		<55	<55	<46
CP0005	65	82	68		60	67	56		<55	<55	<46
CP0006	85	85	82		69	70	63		60	58	51
CP0007	93	90	90		78	85	78		-	-	-
CP0010	98	90	93		83	83	80		62	63	55
CP0011	70	88	75		75	86	77		65	68	60
CP0012	75	90	79		65	60	55		<55	<55	<46
CP0013	85	88	84		64	75	63		63	63	56
CP0014	93	90	90		73	72	67		68	98	80
CP0015	103	105	108		85	102	92		70	92	77
CP0016	105	93	99		103	90	96		99	91	94
CP0017	85	97	89		69	80	69		55	77	59
CP0018	95	93	93		85	90	85		78	75	72
CP0019	103	93	98		90	95	91		81	77	75
CP0020	105	93	99		83	95	87		95	100	97
Typically Developing
TD001	92	90	89		105	105	106		95	98	96
TD002	80	90	88		89	82	83		90	92	89
TD003	93	82	85		89	95	90		81	92	84
TD004	85	87	83		88	75	78		84	85	81
TD005	93	85	87		83	83	80		99	102	101
TD006	103	98	101		104	105	105		110	106	110
TD007	92	83	85		97	95	95		103	95	99
TD008	95	90	91		85	82	80		84	95	87
TD009	98	97	97		110	97	106		115	102	110
TD010	95	87	89		88	95	90		105	92	98
TD011	92	92	89		90	93	90		81	85	80
TD012	95	92	92		95	97	95		100	92	95
TD013	85	88	84		99	100	99		88	91	87
TD014	98	98	98		94	90	90		92	110	101
TD015	98	97	99		88	113	101		98	102	100
TD016	85	85	82		98	92	94		90	75	79
TD017	95	90	91		90	103	96		89	98	92
TD018	98	90	93		93	105	99		89	98	92

Note: EAS = expressive language ability score, RAS = receptive language ability score, LAS = language ability score.

Figure 3. Mean language ability score from the REEL-3 at 6, 9 and 12 months of age for infants at risk of cerebral palsy (CP) and typically developing (TD) infants (with 95% CI).

Expressive and receptive REEL-3 raw scores were examined descriptively as a function of CP motor type diagnosed at 24 months of age, excluding three participants with developmental delay. Figure 4 shows the mean raw score for the children classified with spastic CP (n = 6) increased for both expressive and receptive modalities on each subsequent testing occasion, indicating an increasing trajectory of emergent language skills. Receptive skills showed a trend to develop at a greater rate than expressive. In contrast, the children who received a diagnosis of dyskinesia (n = 9) remained relatively stable across the testing occasions, showing limited improvement in receptive and expressive emerging language skills over time when compared to children diagnosed with spastic CP. No statistical tests were undertaken given the small sample size.

Question 3. How accurate is the IMP in identifying communication impairment in the children at risk of CP relative to a standard assessment of emerging language?

Figure 4. Mean REEL-3 expressive and receptive raw scores at each time point for infants diagnosed at 24 months with spastic (n = 6) and dyskinetic (n = 9) cerebral palsy (error bars are ± 1 SD).

Because there were no group differences in IMP ceiling scores and REEL language ability scores at 6 months of age, the sensitivity and specificity analysis of the IMP was only undertaken for the 9 and 12 month time points. At 9 months, 11 infants at risk of CP (9 dyskinetic and 2 spastic) were below normal limits based on the 10^th percentile cutoff language ability score from the REEL-3. The same 11 infants plus one other (with spastic CP) were identified as below normal in vocal production based on their IMP ceiling scores. In particular, the Crawford et al.’s (2010) single-case t-test was used to compare the ceiling score of each infant at risk of CP with the distribution of TD infants’ ceiling scores. Infants at risk of CP with at t-test p value of < .05 were unlikely to have come from the same population distribution as the comparison sample and were therefore classed as below normal in terms of their performance on the IMP. Using the REEL-3 language ability score as the reference standard, the sensitivity of the IMP in the identification of communication impairment was 100% at 9 months of age and specificity was 86% (see Table 6 for cross-tabulation data).

Table 6. Cross Tabulation Showing the Number of Children At-risk of Cerebral Palsy Below and Within Normal Limits on the REEL-3 Language Ability Standard Score and IMP Total Ceiling Score at 9 and 12 Months of Age

REEL-3 Language Ability Classification
	9 Months		12 Months
IMP Classification	BNL	WNL	BNL	WNL
BNL (p < .05)	11	1	13	1
WNL (p > .05)	0	6	1	2

Note. One at-risk infant was not assessed on the REEL-3 at 12 months of age and was therefore excluded from analysis for that time point. BNL = below normal limits. WNL = within normal limits. REEL-3 is the Receptive-Expressive Emergent Language Test 3rd edition. IMP is the Infant Monitor of vocal Production .

At 12 months, 14 infants at risk of CP were classified as below normal in language ability based on their REEL-3 score (8 dyskinetic, 3 spastic, 3 developmental delay). Based on IMP ceiling scores 14 infants were classified below normal in vocal development (8 dyskinetic, 3 spastic, 3 developmental delay). The sensitivity and specificity of the IMP at 12 months was 93% and 67%, respectively, with just two infants discrepant in their classification (Table 6). It is also notable that the number of infants being identified with communication impairment increased slightly at 12 months of age compared to when they were 9 months of age, consistent with a worsening trajectory of communication development relative to typically developing peers.

Discussion

The purpose of this study was to provide prospective longitudinal data on the development of early vocal behaviors in infants identified “at risk” of CP, as compared with age matched TD infants, to explore early markers of communication and motor speech impairment, in particular. We used a criterion referenced tool of vocal behaviors (IMP) and a standardized measure of language ability (REEL-3).

This research is framed within the context of robust research that (a) promotes early intervention to capitalize on heightened neuroplasticity,²⁴ (b) promotes advocacy for infants to be identified as high risk rather than waiting for a formal diagnosis^32,68 to facilitate earlier access to intervention³²; and (c) espouses the predictive value of infant vocalizations in identifying the risk of communication impairment in infants with CP.³⁶

Children at-risk of CP were recruited to the study, using the international guidelines for early identification.³² Of the 18 children recruited as at-risk of CP, follow-up of diagnosis at 2 years of age revealed 15/18 received a diagnosis of CP, and 3 a diagnosis of developmental delay. Of these participants over 75% presented with impaired vocal behaviors and language ability that was evident at 12 months of age. Our data suggest the infants at risk of CP were not only behind in their progression toward more complex speech-related vocal behaviors, based on the question ceiling score of the IMP, but their level of achievement toward speech was also less robust as indicated by lower parent ratings of frequency of the achieved vocal behaviors. Limited sample size prevented detailed analyses of vocal behaviors and language ability on the basis of classification of CP. Nevertheless, descriptive analyses indicated greater impairment including reduced rate of development in the vocal behaviors of children diagnosed with dyskinetic CP (GMFCS III–IV).

The study results for each of the questions posed, will be discussed in turn.

The Vocal Behaviors of Children At-risk of CP

Our data show marked group differences in the early vocal behaviors of infants identified at-risk of CP compared to their age matched peers, as measured using the IMP. Whilst little difference was noted at 6 months of age between the two groups, significant group differences were noted between 6 and 9 months when pre-canonical and canonical babble typically emerge.⁴⁰ TD infants as a group demonstrated an increasing trajectory, with a high ceiling score indicating emergence of first words by 12 months. In contrast, infants at-risk of CP showed poorer achievement at pre-canonical (e.g., item 8, closed-mouth raspberry sounds; item 9, vocalizing whilst mouthing) and canonical babble (e.g., item 12, copies a familiar sound; item 13, babbles fluently and rhythmically) levels, with the gap between the at-risk and TD groups widening between 9 months and 12 months.

Canonical babble has been flagged as a potential predictor of language delay in infants with developmental disabilities.^36,41 However, of particular interest to this study is the finding of significant group differences that includes pre-canonical babble, and whether these differences in pre-canonical babble could indicate a flag for early motor speech impairment.

Pre-canonical babble can be described as the period where infants experiment with producing sounds, such as blowing raspberries. It also marks the emergence of vocalizations (such as vowels) that are recognizable to the adult form, and sequences of consonant-like and vowel-like elements with a slower transition time between the consonant and vowel than that observed in canonical babble, referred to as marginal babble.⁴⁰ Our analysis identified infants at-risk of CP showed differences on items of the IMP reflective of early vocal experimentation during mouthing and marginal babble.

A body of research supports the facilitatory role of co-occurring mouthing and vocalization in the development of both language perception and production.^22,69–74 For example, Fagan and Iverson⁷¹ identified changes in articulatory postures arising from the co-occurrence of vocalizations with mouthing, as based on 40 TD infants, aged 6–9 months. They reported that the babbling of infants that occurred concurrently with mouthing contained a greater variety of supraglottal consonants (e.g.,/b/,/d/,/g/) than non-mouthing vocalizations. The longitudinal research paradigm of Majorano et al.,⁷² additionally, identified not only an attentional bias to sounds sampled in the infant’s own vocalizations but also a positive relationship between the preverbal articulatory practice and the number of first words produced. Finally, Choi et al.^22,70 illustrated the capacity to influence an infant’s articulatory configurations and speech perception by temporarily impairing the speech movements of infants using different types of teething toys.

Collectively, the above research findings suggest a bi-directional relationship between the auditory signal produced both internally through the infant’s vocalizations and proprioceptive feedback from the articulators, thus supporting the mutually influential benefits of co-occurring vocalizations with mouthing.^22,75 We, therefore, propose the absence of or impairment in self-directed articulatory experiences within critical periods of language learning and development as identified by specific items on the IMP (items 8, 9, 10), may place an infant at risk of laying down appropriate representations for later speech and language development.²² These findings suggest the potential role of the IMP as a screening tool for the early identification and longitudinal tracking of impaired communication in young infants at-risk of CP. Further research with a larger sample size, validated against direct and objective observation of infant vocalizations (e.g., using acoustic analysis), is required to investigate this more fully in children identified as at-risk of communication impairment associated with CP.

Expressive and Receptive Language Abilities

Analysis of the data pertaining to receptive and expressive language ability supported the concurrent validity of the IMP. Seventy-five percent of participants identified at risk of CP presented with language ability scores at less than the 10th percentile, as assessed using the REEL-3 at 9 and/or 12 months of age. Consistent with the IMP, whilst both the TD and at risk of CP groups show similar levels of communication development at 6 months, a significant difference between the two groups was seen at 9 months of age and this difference became larger at 12 months of age, indicating increasing communication impairment over time, relative to peers, for the at-risk group. Whilst our sample size was small, our findings also indicate the rate of development and level of language impairment differed by type of motor disorder. We found, using the raw scores from the REEL-3, children with dyskinetic CP presented with relatively flat developmental trajectories and greater severity of impairment, when compared to children with spastic CP. This is consistent with previously reported literature.^76,77

It is important to acknowledge that the rate at which children <2 years reach key language milestones is nonlinear and highly variable.⁷⁸ It is, therefore, possible that the children we identified as language impaired at 12 months of age may go on to typical language development, as this has been observed in the typically developing population.^79,80 However, the language scores of the at-risk children tracked in this study indicate their rate of development was much slower than the TD peers, and, as a result, the level of impairment was observed to increase with age. Additionally, Rome-Flanders and Cronk⁶² provide data demonstrating longitudinal stability and reliability of the REEL over time, as based on longitudinal data collected at six time points (6, 9, 12, 15, 18 and 24 months of age). The REEL was administered over five time-points until 18 months of age, and the Reynell Developmental Language Scales administered at 24 months of age. They determined results obtained on the REEL at 15 months to be predictive of language function at 24 months of age.

Existing literature reporting on the communicative functioning of children >2 years of age has established early language abilities are highly predictive of later language abilities.^2,5,81 We, therefore, consider the language ability scores obtained by the infants at-risk of CP in this study, as based on longitudinal data, to be predictive. As such, these data offer an important contribution to the current body of research describing communication development in infants identified at-risk of CP. To our knowledge, this is the first prospective study of communication development in high-risk infants <2 years of age and supports monitoring language development “using brief multiple assessments in early life, rather than a single screening, or in-depth assessment”^82(p342).

Accuracy of the IMP in Identifying Communication Impairment in the Children at Risk of CP

Whilst researchers have long advocated for clinicians to monitor the timing and development of infant vocalizations for early detection and planning of early intervention for communication impairment, this has not yet translated into clinical practice. Barriers to implementation have been attributed to feasibility and clinically valid measures.^83,84

Preliminary evaluations of the clinical utility of the IMP were undertaken using the modified t-test for comparing single-cases to control samples⁸⁵ as benchmarked against the reference standard of the REEL-3. Of the children who presented with language impairment (based on the REEL), 100% and 93% at 9 months and 12 months of age, respectively, also presented with atypical patterns of vocal development. This indicates a high probability of a child with atypical vocal behaviors being positively identified with communication impairment. There was, however, some discrepancy between the REEL-3 and IMP. The IMP produced some false positives with specificity estimates of 85% and 67% at 9 months and 12 months, respectively.

Current literature indicates estimates of sensitivity and specificity of at least 70% are required for screening tools.^86,87 Given this, the risk of falsely identifying impairment using the IMP may be considered unacceptably high. Of course, the associated discrepancy between the IMP and REEL-3 was due only to one to two children being misclassified and a larger sample size would be needed before drawing strong conclusions about the classification accuracy of the IMP. However, if the lower specificity of the IMP is reliable, two possible opposing explanations for this finding are considered. Firstly, the literature has identified decreasing accuracy of specificity with age to be indicative of children resolving their difficulties with time.⁸⁵ This explanation, however, seems unlikely given the severity of the language impairment reported in this study and the tendency for the number of at-risk children to show an impairment increased with age. The second and preferred explanation pertains to the underlying differences in the constructs of the two assessment tools. The REEL-3 is an omnibus tool based on a linguistic model of language production and perception, where vocalization skills are only part of what is assessed in infants at or younger than 12 months of age; whereas the IMP is designed to elicit information pertaining to prelinguistic vocalizations, with specific items sensitive to differences in speech motor control. Whilst further research to replicate the findings is needed, the potential of the IMP tool to flag atypical trajectories of vocal development in infants at risk of CP, in particular, would represent a substantial shift in our capacity to identify early motor speech impairment.

As previously reported, impaired motor speech control (e.g., dysarthria) affects functional communication in >50% of children with CP,^10,13,88 yet, there are currently no criteria for assessing dysarthria in children <2 years of age.⁸⁹ It may be that the IMP is in fact a more sensitive estimate of motor speech impairment and, therefore, a clinically relevant assessment tool for screening early infant vocalizations that enables the tracking of an infant’s progress toward integrated speech production skills over time.

Limitations and Future Research

There are a number of limitations to consider with this study. The sample size was relatively small; thus, it was not possible to explore the relationship between a number of risk factors for communication impairment that have been identified in the literature to date.^5,12,90 These risk factors include children born full term with gray matter injury, increased motor impairment (GMFCS level IV/V), and co-morbid epilepsy. This should be a focus for future studies with a larger sample size. In addition, measures of cognitive function were not reliably available for all participants, and therefore not reported. Furthermore, the small sample size limits the generalizability of the factor structure of the IMP beyond the present sample. Our approach was to use principle components analysis for dimension reduction in order to analyze group differences among related items of the IMP, thereby avoiding testing for group differences separately for each item. Although the factor structure at 9 months appears reliable in the present sample through replication at 12 months, a larger sample of infants is needed, with typically developing children as well as children at risk of communication impairment from a range of etiologies, to determine the underlying dimensional structure of the IMP per se. Similarly, replication with a larger sample size is needed before relying on the reported levels of sensitivity and specificity of the IMP for identifying communication impairment for clinical purposes. Our findings do suggest, however, a larger study for that purpose is warranted.

Our sample was also biased toward children with bilateral spastic CP and dyskinetic CP. One could argue, therefore, that our finding of impaired infant vocalizations and language ability in >75% of our participants at 12 months is not unexpected.

It is well documented that many standardized assessments require behavioral responses that are beyond the ability of children with moderate to severe motor impairment,⁹¹ and there is a paucity of tools available for children <2 years of age. Hence, it is acknowledged that the REEL-3 was not designed for use in children with severe motor limitations. However, the parent interview format of the REEL-3 enabled the investigators to minimize the likelihood of underestimation of language comprehension by personalizing items that required a motor response.

Conclusion

This is the first study to report the longitudinal development of early infant vocalizations in children at high risk or diagnosed with CP. Whilst it is now possible to reliably diagnose CP before 6 months of age,³² the earliest reported classification of communication and/or motor speech impairment is 2 years of age, and current clinical practice guidelines prioritize dysphagia.¹ This has contributed to a dearth of information available to clinicians on which to base evidence-based practice decisions for children presenting with communication impairment <2 years of age.

Whilst the sample size is small, our data suggest that the IMP is a valid tool for tracking the vocalizations of infants with/at risk of CP. The first 12 months of age represent a critical period of learning. Research clearly identifies the characteristics of infant vocalizations associated with pre-canonical and canonical babbling as providing the strongest evidence base for predicting communication outcomes in infants at risk of CP, with a body of research reporting “cascading and mutually influential critical periods in language development across the first months of life”²² (p213), where experiential attunement at one stage influences later development.

Data from this study support the view that referral for a communication assessment of children identified at risk of CP should not be delayed. To ensure early interventions are targeted, efficient, and cost-effective, methods are needed that identify children with high developmental risks at an early age.³² The present study supports further research into markers of impairment using infant vocalizations with tools such as the IMP in achieving that goal.

Disclosure

The authors report there are no competing interests to declare

Acknowledgments

This first author of this project was supported by: a Western Australian Health Translation Network Early Career Fellowship and the Australian Government’s Medical Research Future Fund (MRFF) as part of the Rapid Applied Research Translation program (2019); and Telethon Child and Adolescent Health Allied Health Fellowship (2015). She was also the WA Post-doctoral lead in the Australasian Cerebral Palsy: Clinical Trials Network - Center of Research Excellence (Aus-CP:CTN –CRE 2017 – 2019); and would like to acknowledge the mentorship of Professor Roslyn Boyd and Professor Iona Novak.

This research was embedded within the Early Intervention (EI) Service PCH. Our thanks go to the EI service staff and families. Without their dedication and support, this research would not have been possible.

Additional information

Funding

This work was supported by the Channel 7 Telethon Trust [Telethon Child and Adolescent Health Allied Health]; Western Australian Health Translation Network Early Career Fellowship and the Australian Government’s Medical Research Future Fund (MRFF) [Rapid Applied Research Translation program (2019)].