Abstract
Objective
To evaluate the predictive validity of best practice early detection tools for cerebral palsy (CP) in a high-risk cohort.
Study design
Prospective longitudinal cohort study.
Setting
Neonatal intensive care unit of a regional tertiary hospital in Bangladesh.
Participants
Neonates with risk factors for CP admitted to Mymensingh Medical College Hospital Neonatal Intensive Care Unit between November 2019 and March 2020.
Outcome measures
General Movements Assessment (GMA) at writhing and fidgety periods; Hammersmith Infant Neurological Examination (HINE) and Peabody Developmental Motor Scales Second Edition (PDMS-2) conducted in person at 3, 12 and 24 months. The Developmental Assessment of Young Children (DAYC-2), Ages and Stages Questionnaire (ASQ-3) and Developmental Milestones Chart (DMC) were administered remotely at 6, 9, 12, 18 and 24 months. Due to the impact of COVID-19, a proportion of the cohort was not able to have GMA fidgety videos completed and the first HINE assessment was delayed.
Results
A total of 227 infants were enrolled. Of the surviving infants assessed at 24 months, 36 (29%) had a confirmed diagnosis of CP. The most accurate combination of tools for early detection was GMA and HINE at 3 months (sensitivity 0.91; specificity 1.00). The PDMS-2 Total Motor Quotient, with an optimised cut-off of 59, showed high accuracy at 24 months (sensitivity 0.94; specificity 0.99). Among the tools administered remotely, the DAYC-2 PD, DMC (Gross and Fine Motor domains) and ASQ-3 (Gross and Fine Motor domains) demonstrated strong predictive validity—both individually and in combination—at 9, 12, 18 and 24 months, supporting their use as practical alternatives when in-person assessments are not feasible.
Conclusions
Despite pandemic-related disruptions, an accurate diagnosis was possible as early as 3 months of age using the best practice tools. Our findings support the practicability of scalable early detection models integrating in-person and remote assessments to improve access to timely diagnosis.
Keywords: Cerebral Palsy, Children, Low and Middle Income Countries
WHAT IS ALREADY KNOWN ON THIS TOPIC
While early accurate diagnosis of cerebral palsy (CP) is increasingly becoming standard practice, evidence supporting best practices for early detection in low and middle-income countries (LMICs) remains limited.
WHAT THIS STUDY ADDS
This study provides robust data on the predictive validity of multiple tools for early detection of CP, using both in-person and remote assessments—filling a critical gap in the global literature and offering pragmatic alternatives for diverse settings.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
As the largest prospective longitudinal cohort study of infants in an LMIC focusing on the predictive validity of tools for early detection of CP, these findings offer critical insights that could inform clinical practice, support wider implementation of these tools and guide future research.
Introduction
Cerebral palsy (CP) is an umbrella term encompassing a group of disorders affecting movement and posture, caused by a non-progressive disturbance in the developing brain.1 It is the most common physical disability in childhood. While the search for a cure continues, prevention and early identification of infants at risk of CP are critical priorities. In recent years, a growing body of evidence highlighted the vital role of clinical tools in early detection. Among these, the Prechtl General Movements Assessment (GMA)2,4 and the Hammersmith Infant Neurological Examination (HINE) are notable.5,7 When applied, these tools—both supported by strong predictive validity—enable accurate diagnosis as early as 3 months of age.8
Despite the publication of international guidelines on the early and accurate diagnosis of CP in 2017,8 delayed diagnosis remains common, often occurring well beyond the critical window of optimal neuroplasticity. While some settings have seen improvements in early accurate diagnosis, many low and middle-income countries (LMICs) continue to experience significant delays. In these contexts, the use of validated early detection tools is limited, and parental reports of delayed motor milestones are frequently the first indication of concern.
The Bangladesh CP Register (BCPR)—the first population-based CP register in an LMIC—has driven a meaningful shift in the landscape. Between 2015 and 2019, the median age at diagnosis fell from 5.5 years to 3 years.9 This reflects the sustained efforts of the BCPR team, including ongoing surveillance, active community engagement and the scaling of community-based services.10 11 While these gains are commendable, the window for earlier identification of children with CP—and with it, the potential to optimise neurodevelopmental outcomes through timely interventions—remain. Validated early detection tools with strong predictive validity can enable timely access to intervention. This is especially important as a growing number of early intervention programmes are being implemented in Bangladesh and other LMICs, many of which show the greatest gains when delivered early in life.12 13
In this study, we aimed to evaluate the predictive validity of best practice early detection tools for CP, administered both in-person and remotely, in a high-risk cohort of infants admitted to a neonatal intensive care unit (NICU) in Bangladesh.
Methods
Study design and setting
We conducted a prospective longitudinal cohort study. Eligible neonates with risk factors for developing CP admitted to Mymensingh Medical College Hospital (MMCH) NICU were consecutively enrolled and recruited from November 2019 to March 2020. Data collection, including 2-year outcomes, was completed in March 2022. MMCH is a postgraduate government medical college and tertiary referral hospital located in Mymensingh, a major city in north-central Bangladesh.
Study participants
Eligibility criteria included (1) <37 weeks gestational age (preterm birth);14 (2) clinical presentation or history suggestive of neonatal encephalopathy; (3) neonatal sepsis and/or (4) severe jaundice or kernicterus. Neonates were identified by NICU staff, and medical charts were screened by a study physician using a study-specific eligibility tool (online supplemental table 1). An eligibility tool was developed using the best available evidence on risk factors for CP,8 15 considering available standard clinical tests in MMCH in consultation with a neonatologist, paediatrician and physiotherapists on the study team.
Study variables
CP was defined and diagnosed using the international consensus definition at the time of study design16 and is considered the reference standard.16 Diagnosis was confirmed at 12 and 24 months of age by clinicians. Diagnosis fidelity was achieved by physician and clinician training, clinical case discussions with authors and a clinical overview of all assessments by a trained clinician to ensure each clinical picture aligned with CP definition. All ages are defined and reported as corrected age (CA) unless specified. ‘Tool(s)’ denotes best practice early detection tools to detect CP and are considered index tests for this study. ‘Remote assessment tools’ are best practice early detection tools conducted via telephone.
Data sources
Originally, in a larger study, five in-person assessment timepoints were planned: enrolment (<4 weeks), 3, 6, 12 and 24 months with in-person tools outlined in figure 1. COVID-19 safety restrictions imposed during the study period required design adaptation and introduction of remote assessments for a proportion of infants at the 3-month assessment, and for all infants at 6, 12 and 24 months (figure 1). Furthermore, 9-month and 18-month timepoints were added, making a total of seven timepoints. Allowing for safe study conduct, only enrolment, 12-month and 24-month timepoint assessments were completed in-person. All tools and time points are outlined in figure 1, of which a subset was used for analysis in this study.
Figure 1. Flow diagram of study participants and assessment timepoints. Darker text denotes tools analysed in the current study; lighter text denotes tools collected in the larger study but not analysed in the current study. *Tool conducted remotely via telephone. †Tool conducted both in-person and remotely via telephone. No star * or † is in-person only. Abbreviations: AIMS, Alberta Infant Motor Scale; ASAS, Australia Spasticity Assessment Scale; ASQ-3, Ages & Stages Questionnaires, Third Edition; CPDF, Cerebral Palsy Description Form; DAYC-2, Developmental Assessment of Young Children, Second Edition; DMC, Developmental Milestone Chart; GMA, General Movements Assessment; GMFCS, Gross Motor Function Classification Scale; GMFM-66, Gross Motor Function Measure-66; HINE, Hammersmith Infant Neurological Examination; MiniMACS, Mini Manual Ability Classification System; MTS, Modified-Tardieu Scale; PDMS-2; Peabody Developmental Motor Scales, Second Edition. DAYC-2 denotes Physical Domain (PD) and Cognitive Domain (CD) were collected; however, only DAYC-2 PD was analysed in the current study.
Measures included in analysis for this study
GMA, a video-based assessment of infant spontaneous movements.17 Recently, predictive validity of GMA for detecting CP at writhing age has been shown as 0.90 (95% CI 0.70 to 0.99) sensitivity (Se) and 0.51 (95% CI 0.43 to 0.58) specificity (Sp) at writhing age, and 0.95 (95% CI 0.74 to 1.00) Se and 0.98 (95% CI 0.95 to 1.00) Sp at fidgety age in Sri Lanka,18 a lower middle-income country. These results align with predictive values reported in upper middle-income countries (0.83 to 1.00 Se and 0.96 to 1.00 Sp)19 and high-income countries (HICs). GMA videos were taken during the GMA writhing period while neonates were still admitted in NICU. Furthermore, prior to COVID-19 restrictions, GMA was also conducted during the fidgety period, primarily during the 3-month clinic-based timepoint. Infants preferably had fidgety GMA at 13 weeks CA because gestational age can be more challenging to determine accurately in some resource-limited environments.20 Therefore, conducting the GMA at 13 weeks CA maximises the likelihood that infants are assessed during the fidgety period. To determine the most accurate gestational age for scheduling optimal GMA video recordings, the estimated date of delivery (EDD) based on fetal ultrasound was used when available. If the ultrasound-based EDD could not be accurately determined, the date of the last menstrual period (LMP) was used. In cases where the LMP was also unavailable, the infant’s date of birth was used to estimate the EDD. GMA were double-blind scored, with at least one score by an advanced General Movement Trust certified rater or trainer. Abnormal GMA at writhing age was classified as poor repertoire, cramped-synchronised or chaotic general movements. Abnormal GMA at fidgety age was classified absent or abnormal fidgety movements. Normal GMA was considered normal writhing or fidgety general movements at respective ages.21 HINE is a standardised and scorable neurological examination used up to 24 months22 demonstrating high predictive validity for CP in LMIC setting (Se 0.90, Sp 0.80;18 Se 0.83, Sp 0.88)23 comparable to those reported in HICs (Se 0.88, Sp 0.87).24 It is recommended for diagnosis in the international clinical guideline.8 Clinicians were trained in HINE by a certified trainer. Neuroimaging: hospital records were screened for cranial ultrasound (CUS) and/or magnetic resonance imaging (MRI) completed as routine clinical care, with the aim to collect reports and images for analysis. Peabody Developmental Motor Scales Second Edition (PDMS-2) is a norm-referenced assessment of gross and fine motor development for children from birth to 6 years.25 It is considered to have moderate predictive validity for detecting CP in HIC contexts; however, values are not reported.8 It is commonly used in clinical practice to monitor motor development in children with CP. The Alberta Infant Motor Scale (AIMS) is a widely used, performance-based, norm-referenced assessment of gross motor development from birth to 18 months26 with moderate predictive validity for detecting CP or severe global developmental delay in HIC contexts from 4 months (Se 0.77, Sp 0.81).26,8 Bangladeshi norms are not yet available for PDMS-2 or AIMS. The Developmental Milestone Chart (DMC)27 is a simple developmental milestone tool in which age specific milestones are reported as present or absent. It is an acceptable, practical and implementable assessment for children aged 1 month to 8 years in rural Bangladesh.28 Although it does not have predictive validity for CP reported, it was included to understand if its addition would be a simple method to increase accuracy of CP detection when used in combination with other tools.
Addition of tools due to COVID-19 restrictions
The study was amended to include two tools to ensure continuation of safe assessment, which could meet the study objectives for the larger study. Assessment tools required the following attributes: (1) able to be undertaken over telephone assessment (video-based telehealth assessment was deemed unfeasible due to limited participant access to smartphone technology); (2) also able to be completed in an in-person capacity on lifting of COVID-19 restrictions and (3) assessed motor and cognitive development. Additional tools that met these criteria were: (1) Physical domain (PD) and cognitive domain (CD) of the Developmental Assessment of Young Children Second Edition (DAYC-2),29 a parent-reported scored checklist of infant development skills and (2) Ages and Stages Questionnaire-3 (ASQ-3).30 In the current study, only the DAYC-2 PD and motor domains of the ASQ-3 were analysed.
Finally, perinatal and sociodemographic data were collected at enrolment, 12 and 24 months (online supplemental table 2, perinatal and sociodemographic form). ‘Documented’ denotes infant characteristics considered positive only when supported by verifiable medical evidence.
Statistical methods
Demographic and clinical characteristics of the cohort at baseline were investigated using descriptive statistics. The predictive validity of all tools for detecting CP at 24 months of age was investigated using logistic regression models. Univariate analyses were performed investigating the value of all individual tools used up to (and including) the 24-month timepoint for predicting CP at 24 months of age. Continuous assessment scores were dichotomised at previously published best-practice cut-off values; if not available, then cut-off values giving the best predictive performance in the study cohort were identified by computing the area under the receiver operating characteristic curve from logistic regression models.
To establish the value of combination of tools (at single timepoint and across multiple time points) for predicting CP at 24 months, logistic regression models were built combining multiple assessments to define predictive rules. Due to the very high number of possible predictive combinations across assessment tools and multiple timepoints, we employed a hierarchical approach to identify the tools with the strongest predictive validity. First, any individual tool with the highest predictive validity at any single timepoint was prioritised. Then, we evaluated whether combining any additional tools from the same timepoint improved predictive validity on either Se or Sp values. Finally, we examined whether combining tools from multiple timepoints enhanced the predictive validity of assessment tools completed at any single timepoint. Recommended tools were determined using the most accurate assessment at each time point. The predictive validity of indvidual and combination of tools have been grouped into in-person and remote assessments to offer pragmatic alternatives for diverse settings. The Se and Sp with 95% CIs for recommended tools or combinations of tools are reported.
Reliability of remote assessments was assessed by computing the correlation between remote and in-person tool results obtained at 12 months and at 24 months, using Pearson’s correlation coefficient and intracluster correlation for continuous assessment measures, and Kendall’s τ correlation and percentage agreement for discrete assessment measures. Reliability was considered poor (<0.5), moderate (0.5–0.75), good (0.75–0.9) and excellent (>0.9).31 All statistical analysis was carried out in R V.4.1.0. The study was reported using Standards for Reporting Diagnostic accuracy studies.32
Results
In all, 227 neonates were recruited. Cumulatively, 19% (n=43/227), 26% (n=59/227) and 27% (n=61/227) of infants died before 3, 12 and 24 months of age, respectively. A further n=30/227 (13%) withdrew or were lost to follow-up by 24 months of age (figure 1). Infant, clinical and parental characteristics of the cohort at baseline are given in tables1 2. Of the surviving children assessed in-person at 24 m, n=36 (29%) infants received a CP diagnosis.
Table 1. Cohort characteristics at enrolment (n=227).
| Infant characteristics (n=infants with reportable data) | |
|---|---|
| Gender (n=227) | |
| Male | 150 (66%) |
| Female | 77 (34%) |
| Gestational age at birth (n=204) | |
| Preterm (<37 weeks) | 21 (10%) |
| Term (≥37 weeks) | 183 (90%) |
| Birth weight45 (n=224) | |
| Low birth weight (<2500 g) | 59 (26%) |
| Very low birth weight (<1500 g) | 4 (2%) |
| Extremely low birth weight (<1000 g) | 0 (0%) |
| Birth weight (g) (n=224), mean (SD; min, max) | |
| Preterm (<37 weeks) | 2255 (555; 1100, 3200) |
| Term (≥ 37 weeks) | 2686 (429; 1000, 3600) |
| Multiple birth (n=227) | 4 (2%) |
| Consanguinity (n=226) | 38 (17%) |
| Documented infant seizure (n=225) | 159 (71%) |
| Perinatal characteristics | |
|---|---|
| Maternal age at birth (n=226) (years), mean (SD; min, max) | 22.4 (4.2; 17, 38) |
| Mode of delivery (n=212) | |
| Normal vaginal delivery | 157 (74%) |
| Elective caesarean section | 11 (5%) |
| Emergency caesarean section | 43 (20% |
| Birth attended by (n=227) | |
| Dai/untrained birth attendant | 63 (28%) |
| Trained birth attendant | 48 (21%) |
| Medical doctor | 109 (48%) |
| Midwife | 7 (3%) |
| Documented birth complications (n=226) * | 178 (79%) |
| Meconium aspiration | 22 (10%) |
| Pre-eclampsia | 6 (3%) |
| Prolonged labour | 153 (67%) |
| Premature rupture of membranes | 91 (40%) |
| Other (placental abruption, haemorrhage, GBS) | 4 (2%) |
| Required resuscitation at birth (n=227) | 211 (93%) |
Multiple complications reported for some.
BDT, Bangladeshi Taka; GBS, Group B streptococcus; max, maximum; min, minimum.
Table 2. Maternal history and parental sociodemograogic characteristics (n=227).
| Maternal characteristics | |
|---|---|
| Parity (n=225) | |
| 1 | 172 (76%) |
| 2–3 | 49 (22%) |
| 4–5 | 5 (2%) |
| Gravidity (n=226) | |
| 1 | 157 (69%) |
| 2–3 | 55 (24%) |
| 4+ | 15 (7%) |
| Birth complications during previous pregnanc/ie,s, if any (n=70)* | 35 (50%) |
| Preterm birth | 4 (6%) |
| Miscarriage | 29 (41%) |
| Stillbirth | 8 (11%) |
| Gestational diabetes | 1 (1%) |
| Parental sociodemographic characteristics | |
|---|---|
| Maternal education (n=227) | |
| No formal education | 16 (7%) |
| Primary | 60 (26%) |
| Some secondary | 79 (35%) |
| Secondary | 49 (22%) |
| Undergraduate degree | 16 (7%) |
| Post-graduate degree | 7 (3%) |
| Paternal education (n=227) | |
| No formal education | 46 (20%) |
| Primary | 50 (22%) |
| Some secondary | 66 (29%) |
| Secondary | 42 (19%) |
| Undergraduate degree | 16 (7%) |
| Post-graduate degree | 7 (3%) |
| Monthly household income (BDT ৳), per capita (n=222), mean (SD) | 2077 (2104) |
| Monthly household income total (n=222), mean (SD) | 12 095 (8772) |
| Household income by urban poverty line | |
| Below upper urban poverty line | 196 (86%) |
| Above upper urban poverty line | 22 (11%) |
| Unknown | 5 (3%) |
Multiple complications reported for some.
BDT, Bangladeshi Taka; max, maximum; min, minimum.
Individual tools
The predictive validity of each individual tool (in-person and remote) at each timepoint is reported in online supplemental table 3. The tools with the highest predictive validity to detect CP are reported in figure 2.
Figure 2. Tools with the highest predictive validity for the detection of cerebral palsy at 24 months, completed individually or in combination. Abbreviations: 3M, 3 month; 12M, 12 month; 18M, 18 month; 24M, 24 month; ASQ-3, Ages & Stages Questionnaires, Third Edition; DAYC-2 PD, Developmental Assessment of Young Children, Second Edition (Physical Domain); DMC, Developmental Milestone Chart; FM, Fine Motor; GMA, General Movements Assessment; GM, Gross Motor; HINE, Hammersmith Infant Neurological Examination; PDMS-2 TMQ; Peabody Developmental Motor Scales, Second Edition Total Motor Quotient; Se, sensitivity; Sp, specificity.
In-person assessments
Overall, individual tools at single time points showed good predictive validity. At recruitment: all writhing GMA videos were taken within 7.3 weeks CA (mean 0 weeks, SD 2.3). Of n=227 writhing videos, n=70 (31%) could not be scored due to infant state (sleeping or distressed and a proportion of babies were hypokinetic with no general movement observed that could be scored). Of the scoreable videos, n=23 (10%) were normal, n=9 (4%) cramped-synchronised, none were chaotic and the majority (n=125, 55%) scored poor repertoire. Abnormal GMA at writhing age had high Se (1.00, 95% CI 0.85 to 1.00) but low Sp (0.18, 95% CI 0.10 to 0.29) for predicting CP. The predictive validity was improved when using only cramped-synchronised versus normal, but not when using only poor repertoire versus normal (online supplemental table 3). No infant had CUS or MRI completed as part of routine clinical at the NICU, therefore no analysis for neuroimaging was completed. 3 Months: to allow safe conduct of study in COVID-19 restrictions, only n=40 infants had in-person assessment at 3 months. The average age of fidgety GMA was 13 (SD 0.3) weeks. Of the individual assessment tools completed in-person at 3 months, an absent/abnormal fidgety GMA or PDMS Total Motor Quotient (TMQ) ≤79 and a HINE ≤52 had the highest predictive validity (online supplemental table 3). The published HINE cut-off of ≤56 had equal Se (0.91, 95% CI 0.59 to 1.00) but lower Sp (0.60, 95% CI 0.36 to 0.81) in this cohort compared with our cut-off of ≤52. 12 Months: from all in-person assessments conducted at 12 months, a DAYC-2 PD standard score ≤82 had the best predictive validity. This was marginally better than the published HINE cut-off of ≤6533 (Se 0.94, 95% CI 0.81 to 0.99; Sp 0.97, 95% CI 0.91 to 1.00), and similar to the most predictive HINE cut-off of ≤66 (Se 0.97, 95% CI 0.85 to 1.00; Sp 0.97, 95% CI 0.91 to 1.00) in our study cohort. 24 Months: a HINE ≤67 and PDMS-2 TMQ ≤59 had equally good predictive validity for identifying those with a CP. Both HINE and PDMS-2 TMQ had marginally better predictive validity than DAYC-2 PD standard score ≤84, which had slightly lower Sp (0.94, 95% CI 0.87 to 0.98).
Remote assessments
In general, remote assessment tools used after 12 months of age predicted CP more accurately than those before 12 months (online supplemental table 3). However, the remote tool with highest predictive validity before 12 months was the DAYC-2 PD standard score ≤82 at 9 months. 12 Months: a change of more than 15 points on the DAYC-2 PD standard score between 6 and 12 months of age had good predictive validity (Se 0.79, 95% CI 0.62 to 0.91; Sp 0.99, 95% CI 0.91 to 1.00). A 12-month ASQ-3 Gross Motor Total Score ≤15 and a Fine Motor Total Score ≤30 both had good predictive validity. 18 Months: the best individual tools were a DAYC-2 PD Fine Motor score ≤86, and the absence of DMC motor milestones. 24 Months: at 24 months, the best individual remote assessment tools were a DAYC-2 PD Fine Motor ≤87, and absent motor milestones on the DMC. These assessments had comparable predictive validity to in-person assessments including HINE and PDMS TMQ at 24 months.
Combination of tools
Single timepoint
The combined validity of tools at any single time point is reported in online supplemental table 4. Tools combined from earlier timepoints which increased predictive validity at each single timepoint are reported in figure 2. 3 months: a combination of any two out of absent/abnormal fidgety GMA, HINE ≤52 and PDMS TMQ ≤84 had better predictive validity compared to each assessment individually; the combination with the best predictive validity was absent/abnormal fidgety GMA and a 3-month HINE ≤52, figure 2 and online supplemental table 4. Predictive validity was not improved by adding any PDMS-2 TMQ scores. When investigating previously published predictive rules for the DAYC-234 (change of >15, >20 or >30 points between two assessments completed 6 months apart), we found that apart from a change of >15 between 6 and 12 months of age, these had poorer predictive validity in this cohort compared to other tools (online supplemental table 3).
Multiple timepoints
The predictive validity of tools across multiple time points is reported in online supplemental table 5. Combination of tools with the highest predictive validity across timepoints are reported in figure 2. Prior to 18 months: no combination of tools was more predictive than individual tools or a combination of tools at any single time point. 18 months: combining 18-month DMC GM with 12-month HINE completed in-person had similar predictive validity to combinations of remote only assessments (figure 2, online supplemental table 5). 24 months: similarly, 24-month DMC GM combined with either in-person assessments or remote assessments only had equally good predictive validity (figure 2, online supplemental table 5).
Reliability of assessments
The correlation between the ASQ-3 and DAYC-2 assessments completed remotely versus in-person at 12 and 24 months is shown in table 3. There was good to excellent reliability for all ASQ-3 domains at 12 months, but poor reliability at 24 months. 12 Months: reliability between remote and in-person scores was excellent for DAYC-2 PD Fine Motor subdomain, and moderate for DAYC-2 PD Gross Motor subdomain. 24 Months: reliability was good for DAYC-2 PD Gross and Fine Motor subdomains. However, the reliability between remote and in-person DAYC-2 PD was poor (table 3).
Table 3. Reliability of remote versus in-person assessments at 12 and 24 months.
| 12 months | 24 months | |||
|---|---|---|---|---|
| DAYC-2 |
Pearson’s
r (95% CI) |
ICC (95% CI) |
Pearson’s
r (95% CI) |
ICC (95% CI) |
| Physical Domain* | 0.85 (0.80 to 0.89) | 0.85 (0.80 to 0.89) | 0.48 (0.33 to 0.60) | 0.47 (0.32 to 0.60) |
| Gross Motor* | 0.74 (0.63 to 0.82) | 0.88 (0.84 to 0.90) | 0.73 (0.62 to 0.81) | 0.77 (0.70 to 0.83) |
| Fine Motor* | 0.93 (0.89 to 0.95) | 0.93 (0.90 to 0.95) | 0.79 (0.71 to 0.85) | 0.81 (0.74 to 0.86) |
| Cognitive Domain | 0.93 (0.90 to 0.95) | 0.87 (0.82 to 0.91) | 0.88 (0.83 to 0.92) | 0.90 (0.87 to 0.92) |
| ASQ-3 |
Kendall’s τ (95% CI) |
Percent agreement (± 5 points) |
Kendall’s τ (95% CI) |
Percent agreement (± 5 points) |
| Gross Motor* | 0.87 (0.82 to 0.91) | 100/134 (75%) | 0.32 (0.16 to 0.47) | 51/124 (41%) |
| Fine Motor* | 0.90 (0.86 to 0.93) | 95/134 (71%) | 0.39 (0.23 to 0.53) | 51/124 (41%) |
| Communication | 0.92 (0.89 to 0.94) | 113/134 (84%) | 0.38 (0.22 to 0.52) | 53/124 (43%) |
| Problem-solving | 0.92 (0.89 to 0.94) | 103/134 (77%) | 0.37 (0.21 to 0.52) | 55/124 (44%) |
| Personal Social | 0.89 (0.85 to 0.92) | 106/134 (79%) | 0.29 (0.12 to 0.44) | 57/124 (46%) |
*Only the motor domains of DAYC-2 and ASQ-3 were analysed to evaluate the predictive validity of tools for early detection of CP in the current study.
ASQ-3, Ages & Stages Questionnaires, Third Edition; CI, Confidence Interval; DAYC-2, Developmental Assessment of Young Children, Second Edition; ICC, intracluster correlation.
Discussion
This prospective cohort study evaluated the accuracy of best-practice early detection tools for CP in a high-risk infant population in Bangladesh, using both in-person and remote assessments between 3 and 24 months of age. The predictive validity of these tools, used independently and in combination, reveals promising insights and opportunities to scale up these practices in LMICs. Our findings confirm that accurate early detection of CP is achievable and support the adaptability of recommended early detection pathways in LMICs.
Our findings align with previous research, which highlight the caveats of applying tools standardised using data from HICs in LMIC settings.35 In our study, the most predictive cut-off for PDMS-2 (TMQ ≤59) was lower than the normative mean, and the DAYC-2 showed limited utility using published cut-offs. These differences suggest that developmental trajectories in this population may not align with standard expectations, or that caregiver-reported changes vary due to cultural and contextual factors—highlighting the need to adapt these tools for use in diverse settings.
Our study also highlights critical implementation challenges. Although neuroimaging is a key component of international guidelines for early detection of CP, it was unavailable for all infants in this cohort. This reflects a broader pattern in LMICs, where structural and economic barriers often limit access to neuroimaging.36 37 In the absence of imaging, functional and clinical assessments such as GMA, HINE and caregiver-reported developmental milestones assume even greater importance. Although contextual challenges in video collection, infant state regulation and clinician training impacted GMA in our study, the preliminary findings from the fidgety age are promising, although based on limited data. Similar environmental and training constraints have been reported by studies from other LMICs, including India and Sri Lanka.38 The HINE was practical to administer in our study setting and provided meaningful diagnostic insights in alignment with previous work in both high-resource and LMIC settings demonstrating moderate-to-high Se and Sp when administered by trained clinicians.39 This reinforces its potential as a cornerstone of early detection even in resource-limited environments.
Our findings underscore the value of building detection pathways that do not rely on neuroimaging, a direction also suggested by recent evidence from other LMICs.19 Despite the disruptions of the COVID-19 pandemic, remote assessments maintained continuity and diagnostic accuracy, illustrating the adaptability of some of these tools. Remote methods—particularly for caregiver-reported tools —may play a critical role in scaling early detection efforts in LMICs, provided training and infrastructure support are in place, though it should not replace earlier neurological assessment when feasible. This is consistent with global calls for pragmatic, context-sensitive strategies that leverage existing systems to detect neurodevelopmental disability early.40
Integration of early screening and detection tools into existing health and community programmes could meaningfully reduce the age of CP diagnosis in Bangladesh, currently reported at over 3 years.10 Early diagnosis enables early intervention, which improves long-term motor and cognitive outcomes.41 The present study provides empirical evidence that such outcomes are within reach in LMICs, even under real-world conditions.
Strengths and limitations
This study represents the first prospective developmental cohort in an NICU population in Bangladesh targeting infants with CP. Best practice tools were used, and where able, adapted for the local context. The use of both in-person and remote assessments allowed continuity of assesments during COVID-19 disruptions, which represents a strength of the study under the circumstances.
However, several limitations must be acknowledged. Some assessment tools lack back-translation, and their psychometric properties are not established in LMICs. Developmental tools such as PDMS-2 and DAYC-2, while globally recognised, may not fully reflect culturally embedded developmental expectations. Some conventional risk factors such as low birth weight or gestational age <32 weeks could not be comprehensively analysed due to missing data or reliance on recall. Gestational age is known to have lower accuracy in low literacy settings.42 Future research in LMICs would benefit from prospective newborn examinations such as The New Ballard scoring or Dubowitz scoring to improve risk modelling around gestational age.42
Finally, the rate of infants who died and had CP showed signs of NE that was higher than expected, even for a high-risk population. In total, 96.5% of infants who died and 100% of infants with CP were classified as NE grades 1–3. This may be explained by health facilities being the final location for infants to receive care following complications at birth.43 Further barrier to attending health facilities can be additional costs.44 While there has been substantial improvement in access to care for pregnant women,44 equitable access for all high-risk infants to NICU care requires further support. However, there may also have been an overestimation of NE rates in the cohort due to multiple comorbidities in the infants included in the study.
Conclusions
Infants at risk of CP were accurately diagnosed as early as 3 months of age using recommended tools. Our findings support the practicability of scalable early detection models that integrate in-person and remote assessments to improve access to timely diagnosis in LMICs.
Supplementary material
Acknowledgements
Data collection was supported by Dr Md Jahangir Alam. We acknowledge all infants and their families for their contribution to this research.
Footnotes
Funding: This project was funded by Cerebral Palsy Alliance Research Foundation (PG01318). The funder did not influence the results/outcomes of the study despite author affiliations with the funder.
Provenance and peer review: Part of a topic collection; not commissioned; internally peer-reviewed.
Patient consent for publication: Consent obtained from parent(s)/guardian(s)
Ethics approval: Ethical approval was obtained from Asian Institute of Disability and Development (southasia-hrec-2019-5-02) and Bangladesh Medical Research Council (BMRC/NREC/2016-2019/72). Participants gave informed consent to participate in the study before taking part.
Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Data availability statement
Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.
References
- 1.Blair E, Cans C. In: Cerebral palsy. Panteliadis C, editor. Cham: Springer International Publishing; 2018. The definition of cerebral palsy; pp. 13–7. [Google Scholar]
- 2.Prechtl HF, Einspieler C, Cioni G, et al. An early marker for neurological deficits after perinatal brain lesions. Lancet. 1997;349:1361–3. doi: 10.1016/S0140-6736(96)10182-3. [DOI] [PubMed] [Google Scholar]
- 3.Burger M, Louw QA. The predictive validity of general movements – A systematic review. Eur J Paediatr Neurol. 2009;13:408–20. doi: 10.1016/j.ejpn.2008.09.004. [DOI] [PubMed] [Google Scholar]
- 4.Kwong AKL, Fitzgerald TL, Doyle LW, et al. Predictive validity of spontaneous early infant movement for later cerebral palsy: a systematic review. Develop Med Child Neuro. 2018;60:480–9. doi: 10.1111/dmcn.13697. [DOI] [PubMed] [Google Scholar]
- 5.Pietruszewski L, Moore-Clingenpeel M, Moellering GCJ, et al. Predictive value of the test of infant motor performance and the Hammersmith infant neurological examination for cerebral palsy in infants. Early Hum Dev. 2022;174:105665. doi: 10.1016/j.earlhumdev.2022.105665. [DOI] [PubMed] [Google Scholar]
- 6.Morgan C, Romeo DM, Chorna O, et al. The Pooled Diagnostic Accuracy of Neuroimaging, General Movements, and Neurological Examination for Diagnosing Cerebral Palsy Early in High-Risk Infants: A Case Control Study. J Clin Med. 2019;8:1879. doi: 10.3390/jcm8111879. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Romeo DM, Ricci D, Brogna C, et al. Use of the Hammersmith Infant Neurological Examination in infants with cerebral palsy: a critical review of the literature. Dev Med Child Neurol. 2016;58:240–5. doi: 10.1111/dmcn.12876. [DOI] [PubMed] [Google Scholar]
- 8.Novak I, Morgan C, Adde L, et al. Early, Accurate Diagnosis and Early Intervention in Cerebral Palsy: Advances in Diagnosis and Treatment. JAMA Pediatr. 2017;171:897–907. doi: 10.1001/jamapediatrics.2017.1689. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Khandaker G, Muhit M, Karim T, et al. Epidemiology of cerebral palsy in Bangladesh: a population-based surveillance study. Dev Med Child Neurol. 2019;61:601–9. doi: 10.1111/dmcn.14013. [DOI] [PubMed] [Google Scholar]
- 10.Al Imam MH, Jahan I, Das MC, et al. Population-based surveillance of children with cerebral palsy enables early diagnosis and intervention. Dev Med Child Neurol. 2021;63:883–4. doi: 10.1111/dmcn.14861. [DOI] [PubMed] [Google Scholar]
- 11.Karim T, Das MC, Muhit M, et al. Improving epilepsy control among children with cerebral palsy in rural Bangladesh: a prospective cohort-based study. BMJ Open. 2022;12:e052578. doi: 10.1136/bmjopen-2021-052578. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Al Imam MH, Jahan I, Das MC, et al. SUpporting People in extreme POverty with Rehabilitation and Therapy (SUPPORT CP): A trial among families of children with cerebral palsy in Bangladesh. Dev Med Child Neurol. 2023;65:773–82. doi: 10.1111/dmcn.15445. [DOI] [PubMed] [Google Scholar]
- 13.Karim T, Muhit M, Jahan I, et al. Outcome of Community-Based Early Intervention and Rehabilitation for Children with Cerebral Palsy in Rural Bangladesh: A Quasi-Experimental Study. Brain Sci. 2021;11:1189. doi: 10.3390/brainsci11091189. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.World Health ORganisation . World Health Organisation; 2023. Preterm birth.https://www.who.int/news-room/fact-sheets/detail/preterm-birth Available. [Google Scholar]
- 15.Mcintyre S, Taitz D, Keogh J, et al. A systematic review of risk factors for cerebral palsy in children born at term in developed countries. Develop Med Child Neuro. 2013;55:499–508. doi: 10.1111/dmcn.12017. [DOI] [PubMed] [Google Scholar]
- 16.Rosenbaum P, Paneth N, Leviton A, et al. A report: the definition and classification of cerebral palsy April 2006. Dev Med Child Neurol Suppl. 2007;109:8–14. doi: 10.1111/j.1469-8749.2007.00001.x. [DOI] [PubMed] [Google Scholar]
- 17.Prechtl HFR. Qualitative changes of spontaneous movements in fetus and preterm infant are a marker of neurological dysfunction. Early Hum Dev. 1990;23:151–8. doi: 10.1016/0378-3782(90)90011-7. [DOI] [PubMed] [Google Scholar]
- 18.Hewawitharana G, Darshana Ila N, Madhushani Ui A, et al. Prediction of cerebral palsy and cognitive delay among high-risk children in a developing nation: A successful early detection programme. Dev Med Child Neurol. 2025;67:892–900. doi: 10.1111/dmcn.16197. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.King AR, Al Imam MH, McIntyre S, et al. Early Diagnosis of Cerebral Palsy in Low- and Middle-Income Countries. Brain Sci. 2022;12:539. doi: 10.3390/brainsci12050539. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Tikmani SS, Mårtensson T, Khalid S, et al. Assessing the diagnostic accuracy of postnatal clinical scoring methods and foot length measurement for estimating gestational age and birthweight of newborns in low- and middle-income countries: a systematic review and meta-analysis. BMJ Paediatr Open. 2024;8:e002717. doi: 10.1136/bmjpo-2024-002717. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Einspieler C, Prechtl HFR, Bos AF, et al. In: Prechtl’s Method on the Qualitative Assessment of General Movements in Preterm, Term and Young Infants. Einspieler C, Prechtl HFR, editors. London: Mac Keith Press; 2004. [Google Scholar]
- 22.Haataja L, Mercuri E, Regev R, et al. Optimality score for the neurologic examination of the infant at 12 and 18 months of age. J Pediatr. 1999;135:153–61. doi: 10.1016/S0022-3476(99)70016-8. [DOI] [PubMed] [Google Scholar]
- 23.Apaydın U, Erol E, Yıldız A, et al. The use of neuroimaging, Prechtl’s general movement assessment and the Hammersmith infant neurological examination in determining the prognosis in 2-year-old infants with hypoxic ischemic encephalopathy who were treated with hypothermia. Early Hum Dev. 2021;163:105487. doi: 10.1016/j.earlhumdev.2021.105487. [DOI] [PubMed] [Google Scholar]
- 24.Bosanquet M, Copeland L, Ware R, et al. A systematic review of tests to predict cerebral palsy in young children. Develop Med Child Neuro. 2013;55:418–26. doi: 10.1111/dmcn.12140. [DOI] [PubMed] [Google Scholar]
- 25.Folio M, Fewell R. Peabody developmental motor scales. 2nd. Austin, Texas: Pro-ed; 2000. edn. [Google Scholar]
- 26.Darrah J, Piper M, Watt MJ. Assessment of gross motor skills of at‐risk infants: predictive validity of the Alberta Infant Motor Scale. Develop Med Child Neuro. 1998;40:485–91. doi: 10.1111/j.1469-8749.1998.tb15399.x. [DOI] [PubMed] [Google Scholar]
- 27.Scherzer AL. Experience in Cambodia With the Use of a Culturally Relevant Developmental Milestone Chart for Children in Low‐ and Middle‐Income Countries. Policy Practice Intel Disabi. 2009;6:287–92. doi: 10.1111/j.1741-1130.2009.00234.x. [DOI] [Google Scholar]
- 28.Karim T, Scherzer A, Muhit M, et al. Use of a Developmental Milestone Chart (DMC) in Rural Bangladesh to Educate Health Workers and Stimulate Referral for Early Diagnosis and Intervention. J Trop Pediatr (Lond) 2019;65 doi: 10.1093/tropej/fmy073. [DOI] [PubMed] [Google Scholar]
- 29.Voress JK, Maddox T. Examiner’s manual austin. 2nd. Texas: Pro-Ed; 2013. Developmental assessment of young children. edn. [Google Scholar]
- 30.Squires J, Bricker D. Ages & stages questionniares, third edition (ASQ-3TM) 3rd. Brookes Publishing Co; 2009. edn. [Google Scholar]
- 31.Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016;15:155–63. doi: 10.1016/j.jcm.2016.02.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Bossuyt PM, Reitsma JB, Bruns DE, et al. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ. 2015;351:h5527. doi: 10.1136/bmj.h5527. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Romeo DMM, Cioni M, Palermo F, et al. Neurological assessment in infants discharged from a neonatal intensive care unit. Eur J Paediatr Neurol. 2013;17:192–8. doi: 10.1016/j.ejpn.2012.09.006. [DOI] [PubMed] [Google Scholar]
- 34.Maitre NL, Slaughter JC, Aschner JL. Early prediction of cerebral palsy after neonatal intensive care using motor development trajectories in infancy. Early Hum Dev. 2013;89:781–6. doi: 10.1016/j.earlhumdev.2013.06.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Semrud-Clikeman M, Romero RAA, Prado EL, et al. [Formula: see text]Selecting measures for the neurodevelopmental assessment of children in low- and middle-income countries. Child Neuropsychol. 2017;23:761–802. doi: 10.1080/09297049.2016.1216536. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Murali S, Ding H, Adedeji F, et al. Bringing MRI to low- and middle-income countries: Directions, challenges and potential solutions. NMR Biomed. 2024;37:e4992. doi: 10.1002/nbm.4992. [DOI] [PubMed] [Google Scholar]
- 37.Jalloul M, Miranda-Schaeubinger M, Noor AM, et al. MRI scarcity in low- and middle-income countries. NMR Biomed. 2023;36:e5022. doi: 10.1002/nbm.5022. [DOI] [PubMed] [Google Scholar]
- 38.Benfer KA, Ghosh AK, Chowdhury S, et al. Early detection of infants with neurodevelopmental concerns indicative of cerebral palsy in a lower middle‐income country (India) Develop Med Child Neuro. 2025 doi: 10.1111/dmcn.16351. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Hidalgo-Robles Á, Merino-Andrés J, Cisse MRS, et al. The Pathway Is Clear but the Road Remains Unpaved: A Scoping Review of Implementation of Tools for Early Detection of Cerebral Palsy. Children (Basel) 2025;12:941. doi: 10.3390/children12070941. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Kim F, Maitre N, Cerebral Palsy Foundation A Call for Early Detection of Cerebral Palsy. Neoreviews. 2024;25:e1–11. doi: 10.1542/neo.25-1-e1. [DOI] [PubMed] [Google Scholar]
- 41.Morgan C, Fetters L, Adde L, et al. Early Intervention for Children Aged 0 to 2 Years With or at High Risk of Cerebral Palsy. JAMA Pediatr. 2021;175:846–58. doi: 10.1001/jamapediatrics.2021.0878. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Ravi AS. Born Too Soon: The Global Action Report on Preterm Birth. Aust Med J. 2012:598. [Google Scholar]
- 43.Abdullah ASM, Dalal K, Yasmin M, et al. Perceptions and practices on newborn care and managing complications at rural communities in Bangladesh: a qualitative study. BMC Pediatr. 2021;21:168. doi: 10.1186/s12887-021-02633-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Nandi A, Charters TJ, Quamruzzaman A, et al. Health care services use, stillbirth, and neonatal and infant survival following implementation of the Maternal Health Voucher Scheme in Bangladesh: A difference-in-differences analysis of Bangladesh Demographic and Health Survey data, 2000 to 2016. PLoS Med. 2022;19:e1004022. doi: 10.1371/journal.pmed.1004022. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Organization UNCsFaWH . Low birthweight: country, regional and global estimates. New York: UNICEF; 2004. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.