Hammersmith Infant Neurological Examination Subscores are Predictive of Cerebral Palsy

Abstract

Introduction:

The Hammersmith Infant Neurological Examination (HINE) is a standardized assessment that identifies early signs of cerebral palsy (CP). In practice, the clinician performs this assessment in its entirety, yielding a global score. This study aimed to investigate the individual HINE subscores and “asymmetries” as predictive indicators of CP.

Methods:

In this retrospective nested case control study, a pediatric neurologist performed the HINE on a cohort of 3–4-month-old former NICU infants. The infants’ neurodevelopmental outcomes were determined by chart review when they were 2–3 years of age. We performed univariate and multivariable logistic regression analyses to yield the accuracy of the global HINE score, HINE subscores, and “asymmetries” in classifying infants with and without CP.

Results:

Of the 108 infants on whom HINE was performed, 50 were either discharged due to normal developmental progress or were lost to follow-up. Of the remaining 58 subjects, 17 had CP and 41 did not. Receiver operator characteristic (ROC) curves of univariate models yielded the following area under the curve (AUC) scores: global HINE score (AUC = 0.75), “reflexes and reactions” (AUC = 0.80), “cranial nerve function” (AUC = 0.76), “asymmetries” (AUC = 0.75), and “movements” (AUC = 0.71). The ROC for our multivariable model (AUC = 0.91) surpassed the global HINE score’s predictive value for CP.

Conclusions:

The weighted combination of HINE subscores and “asymmetries” outperforms the global HINE score in predicting CP. These findings suggest the need for revisiting HINE, but further validation with a larger dataset is required.

Introduction:

Cerebral palsy (CP) is characterized as a group of permanent disorders that affect the development of movement and posture, causing activity limitation. It stems from non-progressive brain injuries that occur during early brain development, [1,2] and it affects nearly one to four per thousand children in the US. [3] It often remains undetected until the child starts missing developmental milestones, which typically occurs months to years after the initial insult. This delay in identification then leads to delays in evaluation and treatment. Therefore, alternative methods to evaluate the central nervous system in infants and young children is crucial. Pediatric neurologists and developmental pediatricians commonly rely on primitive reflexes and postural reactions as one of the earliest, simplest, and fundamental tools to assess the integrity of the immature nervous system. [4,5]

Infants with CP often exhibit either prolonged presence of primitive reflexes or absent postural reactions. Various studies have highlighted the clinical importance of measuring reflexes to aid in the early diagnosis and differential diagnosis of CP. Furthermore, infants who have five or more absent or abnormal postural reactions are highly likely to have CP or developmental delay (DD). [4,5] One tool that incorporates the assessment of neonatal reflexes and postural reactions is the Hammersmith Infant Neurological Examination (HINE).

Recently the HINE has emerged as one of the early clinical examination tools for the early diagnosis of CP. [6] It is a standardized neurological exam for infants between 2 and 24 months of age. It takes 10–15 minutes for a skilled clinician to perform. [7–9] It consists of 26 items that are categorized as 5 subscores that assess the “cranial nerve function”, “posture”, “movements”, “tone”, and “reflexes and reactions” of these children. Each item is scored individually on a scale of 0 to 3, and these item scores are summed to obtain category subscores. The category subscores are then summed to obtain a global HINE score ranging from 0 to 78. Any observed asymmetries between sides are also quantified. [8–10]

The primary objective of the HINE is to identify early signs of neuromotor disorders in infants, including CP. For example, in an infant who is 3 months corrected age, a global HINE score of 67 is considered optimal, and a score below 56 has a 96% predictive validity for CP. [11,12] The HINE requires the clinician to guide the infant through a series of maneuvers in the supine, prone, and vertical positions to evaluate their developmental responses.

Our study aimed to investigate the HINE subscores and enhance physicians’ ability to identify and prioritize the most predictive indicators of CP. We hypothesized that the “reflexes and reactions” subscore would demonstrate a high predictive value for CP due to the frequent observation of persistent primitive reflexes and absence of postural reactions in infants diagnosed with CP. [4,5] Additionally, prior research suggests that higher asymmetry scores during HINE assessments may indicate hemiplegic CP. [9] We evaluated the efficacy of “reflexes and reactions” and “asymmetry” as predictors of CP using univariate and multivariable models. By investigating individual subscores and the “asymmetry” score individually and in combination, our study aimed to optimize the HINE’s ability to predict CP at 3–4 months of age. This information could assist physicians in early detection and intervention strategies for infants at risk of CP.

Method:

As part of a larger study, we conducted a retrospective nested case control study involving infants admitted to the level-IV neonatal intensive care unit (NICU) infants at Arkansas Children’s Hospital (ACH). We identified a cohort of infants who had a HINE assessment at 3–4 months of corrected age during an inpatient neurologic consultation from 5/2019–10/2020. Reasons for neurologic consultation included the following: <37-week gestational age, seizures, positional plagiocephaly, hypoxic ischemic encephalopathy (HIE), down syndrome, hydrocephalus, and abnormal head imaging on brain MRI (magnetic resonance imaging) or head ultrasound, with findings that include brain bleeds (grade 2–4 intraventricular hemorrhage (IVH), and hemorrhages) and white matter changes such as periventricular leukomalacia (PVL).

Starting in 12/2021, we employed a chart review approach to determine the developmental outcomes of our subject cohort when they were 2–3 years of age. Among this cohort, a subgroup of individuals continued to receive follow-up care within the NICU follow-up services, while the remaining subjects were either discharged from the follow-up services due to age-appropriate developmental progress or lost to follow-up. For the children still receiving clinical care, their medical records were examined to identify those who had a diagnosis of CP and those without CP at the age of 2–3 years. Children for whom the neurodevelopmental outcome at 2–3 years was not available were excluded from the statistical analyses. A pediatric neurologist or pediatric physiatrist gave the diagnosis of CP based on clinical history, comprehensive neurological examination, neuroimaging, and clinical judgment. In the setting of motor delays, the clinician thoroughly assessed posture, tone, and movement patterns, and evaluated for pathologic reflexes to confirm diagnosis of CP. This study was approved by the University of Arkansas for Medical Sciences (UAMS) institutional review board (IRB).

Statistical analyses:

In this retrospective study, we defined cases as subjects who had been diagnosed with CP and designated all other subjects as controls. The control group included children without CP as determined by a pediatric physiatrist or pediatric neurologist. Some of these children had typical development, and others had DD with no neurologic signs of CP.

Demographic characteristics were reported as frequency and percentage (%) for categorical variables and mean and standard deviation for continuous variables. Comparisons of demographic characteristics between the two groups were made with Fisher’s exact test for categorical variables and the Mann-Whitney U test for continuous variables, due to non-normal data distribution. To evaluate the relationships between our “No CP” and “CP” group global HINE score and individual HINE subscores, the Mann-Whitney U test was used, again due to non-normal data distribution. We created univariate and multivariable logistic regression models, addressing collinearity amongst variables. We created receiver operator characteristic (ROC) curves to yield the accuracy of HINE and its subscores in classifying infants with and without CP. We used leave-one-out cross-validation (LOOCV) to internally validate our model as a means to assess the generalizability and predictive performance of our model within the study population. Statistical analysis was conducted using RStudio (Rstudio Team, 2020). [13]

Results:

As part of a larger study, 108 infants had a HINE evaluation in the NICU follow-up program at 3–4 months of corrected age. Of the original 108 infants, 58 were re-evaluated after 2 years of age to determine their neurodevelopmental outcome. The remaining 50 were either discharged from the NICU follow-up services due to normal developmental progress or were lost to follow-up. Among the children still receiving care, 17 were diagnosed with CP, 29 had developmental delay (DD), and 12 demonstrated age-appropriate neurodevelopment.

In summary, 17 children were determined to have CP, and the remaining 41 did not have CP and were designated as controls. These were labeled as CP and No CP, respectively.

Of the 17 subjects with CP, 12 had spastic quadriplegic CP, 2 had hemiplegic CP, 1 had spastic diplegic CP, 1 had ataxic CP, and 1 had other CP that was still unspecified at the time of data collection. The diagnosis of CP was given based on clinical judgment, medical history, neuroimaging results, neurological and neuromotor assessment results. The demographic characteristics and the comparison between the groups are depicted in Table 1.

Table 1: Demographic characteristics for all subjects.

In regards to CP risk factors, abnormal head ultrasound and IVH were more prevalent in children with CP. HIE was less prevalent in children with CP; however, the majority of our cohort underwent head cooling protocols. All other CP risk factors and demographic information were similar between groups.

Subject demographics	CP (n=17)	No CP (n=41)	p-value
Gender
Female, n (%)	6 (35.3%)	15 (36.6%)	1
Race
White, n (%)	11 (64.7%)	22 (53.7%)	0.57
Black or African American, n (%)	5 (29.4%)	12 (29.3%)	1
Other, n (%)	1 (5.9%)	7 (17.1%)	0.71
Unknown, n (%)	0	1 (2.4%)	1
Insurance
Medicaid, n (%)	14 (82.4%)	33 (80.5%)	1
Private, n (%)	3 (17.7%)	7 (17.1%)	1
None, n (%)	0	1 (2.4%)	1
Medical History (known risk factors for CP)
Birthweight (Kg), mean (SD)	1.98 (1.1)	2.6 (1.2)	0.09
Premature (<37 weeks GA), n (%)	13 (76.5%)	20 (48.8%)	0.08
Abnormal Brain MRI, n (%)	13 (76.5%)	28 (68.3%)	0.25
Abnormal Head ultrasound, n (%)	16 (94.1%)	18 (43.9%)	0.0018
Seizures, n (%)	4 (23.5%)	4 (8.3%)	0.22
Brain Bleeds
Intraventricular Hemorrhage (IVH), n (%)	10 (58.8%)	12 (29.3%)	0.043
Hemorrhages (Intracranial and Cerebellar), n (%)	0	4 (9.8%)	0.58
Periventricular Leukomalacia (PVL), n (%)	6 (35.3%)	5 (12.2%)	0.064
Hypoxic Ischemic Encephalopathy (HIE), n (%)	2 (11.8%)	20 (48.8%)	0.0088
Anoxic brain injury, n (%)	1 (5.9%)	0	0.29
Positional Plagiocephaly, n (%)	6 (35.3%)	13 (31.7%)	1
Down Syndrome, n (%)	0	2 (4.9%)	1
Hydrocephalus, n (%)	5 (29.4%)	4 (9.8%)	0.11

Abbreviations: n: number; SD: Standard Deviation; GA: Gestational age

The distribution of global HINE scores, individual HINE subscores, and “asymmetry” scores in both the “No CP” and “CP” groups is visually represented in Fig. 1 and Fig. 2. The mean, standard deviation, median, and interquartile range (IQR) of the scores for each group are summarized in Table 2. Notably, there were significant differences (p < 0.05) between the “CP” and “No CP” groups’ global HINE score and individual HINE subscores, except for the “posture” and “tone” subscores, as depicted in Table 2. Using equally weighted subscores, the global HINE score yielded an AUC = 0.75 (Fig. 3A). To improve the predictive performance of HINE, a full multivariable model of HINE subscores was developed, incorporating “asymmetries” as one of the variables. In this model, higher weightage was assigned to subscores that exhibited greater predictability for CP. The multivariable HINE model demonstrated a notable improvement in predicting CP with an AUC of 0.91, outperforming the global HINE score. To further ensure the validity and robustness of our multivariable HINE model, a multicollinearity analysis was done. The “tone” subscore was found to be moderately collinear with other HINE subscores, prompting its removal from our multivariable model. Additionally, given the absence of significant differences in “tone” and “posture” scores between the CP and No CP groups, “posture” was also removed from the multivariable model. This left us with a reduced model that combined three HINE subscores (“cranial nerve function”, “reflexes and reactions”, and “movements”) and “asymmetries” as the most predictive subscores for CP at 3 months of age.

Figure 1: The distribution of global HINE scores between the No CP and CP groups.

*Single column fitting image

Figure 2. The distribution of individual HINE subscores and asymmetry scores between No CP and CP groups.

*2 column fitting image

Table 2: Comparison of Hammersmith Infant Neurological Examination (HINE) scores, individual subscores, and asymmetries between the No CP and CP groups.

Using the Mann-Whitney U test, the two groups were significantly different in all variables except “tone” and “posture” (p < 0.05).

Variable	No CP		CP		p-value
	Mean ± SD	Median (IQR)	Mean ± SD	Median (IQR)
Global HINE	61.0 ± 6.7	63 (57–65)	54.3 ± 7.6	53 (48–60)	0.003
Individual HINE subscores
Cranial nerve function	13.7 ± 1.6	15 (13–15)	12.2 ± 1.4	12 (12–13)	0.001
Posture	11.4 ± 2.7	11 (10–13)	9.9 ± 3.2	10 (8–12)	0.084
Movements	5.6 ± 0.9	6 (6–6)	4.9 ± 1.2	5 (4–6)	0.002
Tone	21.7 ± 2.0	22 (21–23)	20.6 ± 2.2	21 (19–23)	0.094
Reflexes and Reactions	8.5 ± 1.7	8 (7–10)	6.6 ± 1.8	6.5 (5.67–7)	<0.001
Asymmetries	0.1 ± 0.5	0 (0–0)	1.1 ± 1.2	1 (0–2)	<0.001

Abbreviations: SD: Standard Deviation; IQR: Interquartile Range; CP: Cerebral Palsy

Figure 3: Receiver Operator Characteristic (ROC) curves relating the following measures and CP diagnosis:

A) HINE Models: Global HINE score univariate model and multivariable HINE models.

B) Univariate Models: Subscores and Asymmetries.

A) HINE Models: This figure depicts the comparison of different HINE models in predicting CP in 3-month-old infants. The global HINE score univariate model, where all the subscores of HINE are added together with equal weightage, had an AUC of 0.75. A multivariable HINE subscore and asymmetries model, which assigned higher weightage to HINE subscores with greater predictability for CP, achieved a significantly improved AUC of 0.91. Furthermore, a reduced multivariable model that incorporated “cranial nerve function”, “reflexes and reactions”, “asymmetries”, and “movements” subscores also yielded an AUC of 0.91.

B) Univariate Models: AUCs for all HINE subscores and asymmetries are shown in the legend in descending order.

*2 column fitting image

AUCs for all individual subscores and “asymmetries” are shown in the legend in descending order Fig 3B. The reduced model included “reflexes and reactions” (AUC=0.80), “cranial nerve function” (AUC=0.76), “asymmetries” (AUC=0.75), and “movements” (AUC=0.71). The reduced model yielded an AUC of 0.91, surpassing the predictive ability of the original unweighted global HINE score (AUC=0.75) (Fig. 3A).

Internal validation with LOOCV quadratic discriminant analysis was performed. The global HINE score univariate model, full subscore multivariable model, and reduced subscore multivariable model each correctly classified 74% of the subjects. Assessment of individual HINE subscore univariate models correctly classified subjects with accuracies ranging from 64% to 78%, with “tone” and “cranial nerves” having the lowest correct classification rate of 64% and 69%. Our multivariable model that incorporated individual HINE subscores and “asymmetries” was more predictive of CP than the global HINE score alone.

Discussion:

The HINE was developed to assess different aspects of neurological findings in infancy and was later found to be predictive of CP. [6,8,14] For this study, we had hypothesized that even though the global HINE score is predictive of CP; certain HINE subscores, when used individually or in combination, would outperform the global HINE score when predicting CP. Our hypothesis that the “reflexes and reactions” subsection of HINE is highly predictive of CP was corroborated by this study.

Our study also demonstrated that the incorporation of asymmetries in the full multivariable HINE model significantly improved its predictive performance. This enhancement is particularly important in instances where infants with CP, especially those with hemiplegic and diplegic CP, surpass the predefined global HINE cutoff score.[6,9,15] In our own cohort, the only child with diplegic CP and one of two children with hemiplegic CP each had global HINE scores above 56. Incorporating asymmetry scores in addition to the global HINE score can help us categorize the topographic characteristics of CP in infants. This aligns with similar findings reported by Hay, et. al. in infants aged 10 months and older. [9]

Furthermore, the reduced multivariable model, consisting of “reflexes and reactions”, “cranial nerve function”, “movements”, and “asymmetries,” displayed enhanced predictive capabilities for CP in 3–4-month-old infants compared to the original unweighted global HINE score (Fig. 3A). These findings highlight the potential benefits of prioritizing the more predictive subscores and including “asymmetries” to help diagnose CP within this age group.

Interestingly, “tone” and “posture” subscores were found to be the least predictive of CP in our study. This observation is intriguing, since “tone,” “posture,” and “movements” are integral components of Prechtl’s general movement assessment (GMA), a tool recognized for its strong predictability for CP at 3–4 months of age. [6,11,16] Furthermore, any “asymmetries” noted during the whole HINE assessment were found to be more predictive than “tone”, “posture”, and “movements” subscores in isolation. This finding might suggest that asymmetries in “posture” and “tone” subscores may be valuable information when predicting CP.

The low predictability of “tone,” “posture,” and “movements” subscores could also potentially be attributed to the relatively small sample size in our study, highlighting the need for further investigation with a larger sample size. Additionally, the utilization of both HINE and GMA together is noted to be more effective in early prognostication of the neurodevelopmental outcome of high-risk infants than relying on either assessment independently. [6,9,11] Incorporation of the GMA in 3–4-month-old infants in our algorithm will likely help to strengthen the ability to predict CP than the use of HINE alone.

In addition to the small sample size, our focus solely on the 3-month HINE scores and subscores may have further limited the predictive potential of the “tone” and “posture” subscores for CP. A previous study highlighted the 2-year locomotion outcome prediction of specific tasks within the “movement”, “tone”, and “posture” subsections of the HINE at the age of 3 months. Tasks within the same HINE subsections were not as predictive at this early age, yet they became more predictive after 9 months of age.[17] Therefore, to gain a more comprehensive understanding, further longitudinal investigations are warranted to explore how these subscores evolve over time and how they ultimately contribute to predicting CP in infants. Additionally, the “tone” subscore from the HINE could be compared to the Modified Ashworth Scores [18] after a CP diagnosis has been established.

Other limitations in our study that must be acknowledged are the retrospective study design and single clinic cohort. This design led to the loss of almost 50% of our sample to loss-to follow-up, rendering our study vulnerable to selection bias towards sicker children who require ongoing medical management at our institution. Future studies employing a prospective multi-center study design will enable the recruitment and retention of a higher number of infants, leading to more robust associations.

Despite these limitations, our study contributes new information to the existing literature on early CP diagnosis. By highlighting more predictive HINE subscores, our findings offer practical implications for clinical practice. Moreover, our discovery of the importance of incorporating asymmetry scores in the global HINE score contributes to refining the predictive capability of this assessment tool.

In conclusion, we recommend that clinicians consider the value of “reflexes and reactions”, “cranial nerves”, “movements”, and “asymmetries” when determining CP risk in high-risk infants.

Highlights:

• Hammersmith Infant Neurological Examination (HINE) subscores predict cerebral palsy (CP).

• “Reflexes and reactions” subscore outperforms other subscores in CP prediction.

• Incorporating “Asymmetries” improves the HINE’s prediction of CP.

• Our multivariable HINE subscore model outperforms HINE global score in predicting CP.

Abbreviations:

CP

Cerebral Palsy

HINE

Hammersmith Infant Neurological Examination

ROC

Receiver Operator Characteristic

NICU

Neonatal Intensive Care Unit

PT

Physical Therapy

OT

Occupational Therapy

ST

Speech Therapy

NDD

Neurodevelopmental Disabilities

ACH

Arkansas Children’s Hospital

HIE

Hypoxic Ischemic Encephalopathy

IVH

Intraventricular Hemorrhage

PVL

Periventricular Leukomalacia

MRI

Magnetic Resonance Imaging

UAMS

University of Arkansas for Medical Sciences

IRB

Institutional Review Board

DD

Developmental Delays

GDD

Global Developmental Delay

LOOCV

Leave-one-out cross-validation

IQR

Interquartile Range

GA

Gestational age