Associations between deprivation and postneonatal cerebral palsy over 35 years: a population-based longitudinal study

Abstract

Introduction

Cerebral palsy (CP) is a non-progressive condition involving movement and muscle tone difficulties due to injury to the developing brain. Most cases arise around birth, but a smaller proportion (postneonatal CP) experience brain injuries between 29 days and 2 years of age. Causes of postneonatal CP are often modifiable and may be more prevalent in socio-economically deprived areas, yet few studies have investigated this issue. We addressed this gap using a population-wide CP register in Northern Ireland.

Methods

We conducted a retrospective study linking births between 1981 and 2016 to area-level deprivation data. Bayesian regression models assessed associations between deprivation, postneonatal CP and related outcomes.

Results

Postneonatal CP declined steadily across the study period (IRR=0.89; 89% uncertainty interval (UI) 0.85 to 0.93), especially among individuals from the most deprived backgrounds. While the odds of infection-related postneonatal CP decreased overall, these did not decrease among CP individuals from deprived backgrounds, who also displayed higher odds of postneonatal CP due to head injuries (OR=2.72, 89% UI 1.09 to 5.30). Postneonatal CP was associated with higher odds of severe motor impairment (OR=1.47, 89% UI 1.07 to 2.00), intellectual delay (OR=2.54, 89% UI 1.83 to 3.55) and seizures (OR=2.55, 89% UI 1.84 to 3.56).

Discussion

Although prevalence has declined, socioeconomic inequalities remain in the causes and consequences of postneonatal CP. These findings highlight the need for targeted public health interventions and further research to address persistent disparities and prevent modifiable causes of CP.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Cerebral palsy (CP) caused by brain injuries in the postneonatal period (29 days–2 years of age) is associated with higher risk of severe impairments. Its causes are often modifiable (eg, infections) and are the target of childhood public health campaigns. The prevalence of postneonatal CP has declined in high-income countries.

WHAT THIS STUDY ADDS

• From the 1980s, the prevalence of postneonatal CP has declined markedly among CP individuals from most deprived backgrounds. However, the results suggest persistent disparities in the causes of postneonatal CP, particularly because individuals from more deprived backgrounds have relatively increased likelihood of postneonatal CP caused by infections and by head injuries.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This study is based on a small number of cases with postneonatal CP; therefore, it needs to be extended. However, the results suggest the need for targeted surveillance systems (eg, health visitor reviews) to prevent children from more deprived backgrounds being exposed to key risk factors for postneonatal CP, which may, in turn, lead to more severe impairments. The results may also indicate the need to increase the reach of childhood public health campaigns (eg, infant vaccination or safe infant environments) in socio-economically deprived communities.

Introduction

Cerebral palsy (CP) is a non-progressive condition whereby injuries to the developing brain cause permanent difficulties in controlling movement and muscle tone.¹ Among individuals with CP, a key distinction exists between congenital CP cases, who experienced the brain injury before or during the neonatal period (up to 28 days postbirth), and postneonatal cases, whose brain injury occurred between 29 days and 2 years postbirth. The causes of brain injuries among postneonatal cases may be preventable, spanning events such as infections, cerebrovascular accidents (CVAs) and head injuries. Similar events have been the target of preventive public health initiatives since the 1980s,² for example, infant immunisation campaigns; promotion of safe infant sleep practices³; improved safety measures for infant and toddler car seats.⁴ In recognition of the links between similar initiatives and risk factors for postneonatal CP, Arnaud and colleagues⁵ proposed using the prevalence of postneonatal CP as a prominent public health indicator.

Evidence indicates that the prevalence of postneonatal CP has declined in high-income countries^{6 7}: a study collating data from European registers reported a prevalence rate per 10 000 live births equal to 0.82 (95% CI 0.73 to 0.92) among births between 2006 and 2012.⁸ Since the causes of postneonatal CP are modifiable, some of its risk factors may be more prevalent among families living in more deprived settings; however, limited research exists on this issue. We capitalised on the Northern Ireland CP Register (NICPR), a population-based register covering a distinct administrative polity within the UK, to investigate the prevalence of postneonatal CP in Northern Ireland (NI) considering births from 1981 to 2016: crucially, we analysed differences in the probability of postneonatal CP for individuals from areas with different levels of socioeconomic deprivation.

The events that lead to brain injuries in the postneonatal period can be complex. For example, a viral infection may cause cardiovascular problems, which then cause brain injuries. This complexity had thwarted investigation⁹ until recent work established a reliable categorisation of these causal events.¹⁰ Capitalising on this advancement, our third aim was to investigate prevalence of postneonatal CP causes over time and their associations with socio-economic deprivation.

The events that cause postneonatal CP are often traumatic: previous studies showed associations between postneonatal CP and more severe impairments.^{7 11–13} However, investigating the links between postneonatal CP and impairments necessitates controlling for the biasing effects of socioeconomic deprivation on both the exposure (postneonatal CP) and the outcome (impairment). The NICPR dataset allowed us to estimate these associations while controlling for socioeconomic deprivation, alongside other covariates. To summarise, we aimed to investigate:

1. Prevalence of postneonatal CP on all live births over time.

2. Associations between deprivation and probability of congenital or postneonatal CP over time.

3. Postneonatal CP causes over time and whether these varied conditionally on deprivation levels.

4. Associations of postneonatal CP with severe impairments, while controlling for biasing effects of deprivation.

Methods

Study design and setting

The NICPR is a confidential record of children with CP in NI born from 1981. Cases of CP are notified by multiple informants (eg, paediatricians and physiotherapists): following notification, a health professional known to the child provides information on the child’s diagnosis and characteristics of CP using a standardised assessment form, which is checked for consistency by a paediatrician attached to the NICPR. Diagnosis ascertainment typically occurs after the child’s fourth birthday. Further information about CP cases is collected through linkage with administrative data. In this manuscript, we focus on the multiple deprivation measure (MDM) of the area where individuals resided around birth (or their earliest recorded address, if born outside NI). The NICPR is the last remaining CP register commissioned by the National Health Service in the UK, and it contributed to the UK National Confidential Enquiry into Patient Outcome and Death,¹⁴ which suggested a comprehensive ascertainment rate. CP registers typically use the proportion of persons with multiple notifications as an indicator of ascertainment (see Results section).

Study population

At the time we started these analyses, the NICPR included N=2058 individuals who had a confirmed diagnosis of CP and were born between 1981 and 2016. Among these, n=124 (6.03%) were categorised as postneonatal CP (see Variables section). In table 1 we report study population characteristics.

Table 1. Characteristics of congenital and postneonatal CP participants.

	CP onset
	Congenital		Postneonatal		Total CP
	N	%	N	%	N	%
Sex at birth*
Female	838	43.33	50	40.32	888	43.15
Male	1096	56.67	74	59.68	1170	56.85
Area deprivation quintiles†
Most deprived	441	22.80	37	29.84	478	23.23
Q2	387	20.01	19	15.32	406	19.73
Q3	396	20.48	23	18.55	419	20.36
Q4	380	19.65	31	25.00	411	19.97
Least deprived	276	14.27	12	9.68	288	13.99
Missing information	54	2.79	2	1.61	56	2.72
Gestational age‡
Term (≥37 weeks)	1016	52.53	80	64.52	1096	53.26
MPT (32–36 weeks)	308	15.93	13	10.48	321	15.60
VPT (28–31 weeks)	321	16.60	<5	<3.00	323	15.69
EPT (<28 weeks)	171	8.84	<5	<3.00	174	8.45
Missing information	118	6.10	26	20.97	144	7.00
CP subtype§
Spastic bilateral	953	49.28	50	40.32	1003	48.74
Spastic unilateral	764	39.50	57	45.97	821	39.89
Other	175	9.05	14	11.29	189	9.18
Missing information	42	2.17	3	2.42	45	2.19
GMFCS rating¶
Mobile (level I–III)	1337	69.13	74	59.68	1411	68.56
Nonmobile (level IV–V)	565	29.21	46	37.10	611	29.69
Missing information	32	1.65	4	3.23	36	1.75
Intellect delay**
No	1006	52.02	35	28.23	1041	50.58
Yes	755	39.04	75	60.48	830	40.33
Missing information	173	8.95	14	11.29	187	9.09
Ever seizures††
No	989	51.14	36	29.03	1025	49.80
Yes	715	36.97	76	61.29	791	38.44
Missing information	230	11.89	12	9.68	242	11.76
Total	1934	100.00	124	100.00	2058	100.00

Significant tests at p<0.05 are reported in bold (in the footnote).

^*χ2(1) = 0.43, p=0.512.

^†χ2(4) = 7.33, p=0.120.

^‡χ(3) = 29.07, p<0.001.

^§χ2(2) = 3.78, p=0.151.

^¶χ(1) = 3.98, p=0.046.

^**χ(1) = 26.87, p<0.001.

^††χ(1) = 28.67, p<0.001.

CP, cerebral palsy; EPT, extremely preterm; GMFCS, Gross Motor Function Classification System; MPT, moderately preterm; VPT, very preterm.

Variables

Outcomes

Clinicians reported the causal events that led to a diagnosis of postneonatal CP using free text. These events were categorised independently by two of the authors using the system developed by Pudig and colleagues.¹⁰ Disagreements (see Results section) were reviewed and reconciled before analyses. Causes of postneonatal CP include six mutually exclusive and exhaustive categories: infections; head injuries; surgery-related brain injuries; CVAs; hypoxic brain damaging events of other origin; all other types of events (eg, status epilepticus), see online supplemental material for more details. Notifying clinicians assessed children’s severity of motor impairment using the Gross Motor Function Classification System (GMFCS)¹⁵: in the analyses, we compared those in level IV or V who are not independently mobile, against others. Notifying clinicians reported whether an intellectual delay was present or absent, whereby delay could be moderate (IQ <70) or severe (IQ <50). Clinicians also indicated whether the child had ever shown epileptic seizures, excluding febrile or neonatal seizures. Children’s assessments were reviewed and completed when the child was over 4 years of age.

Area deprivation

We used the earliest recorded postcode of each CP case to obtain the MDM released by the Northern Ireland Statistical Research Agency in 2017. The smallest ecological unit for which the MDM is available is the super output area (SOA), corresponding to areas with approximately 2000 inhabitants. The MDM includes seven domains of deprivation, ranging from income deprivation to crime and disorder, and includes a ‘health deprivation and disability’ domain. To avoid CP cases contributing to lower area deprivation through the latter domain, we calculated an average ranking of each SOA deprivation excluding the health and impairment domains, then dichotomised the rankings obtained to compare cases in areas in the most deprived quintile versus cases in all the other quintiles.

Other covariates

We categorised infants’ gestational age (GA) as: extremely or very preterm (GA <32 weeks); moderately preterm (GA from 32 to <37 weeks); term (GA ≥37 weeks). Individuals’ sex at birth was categorised as male or female.

Plan of analyses

We applied Bayesian models: an introduction to this approach and more details about the models are in the online supplemental material. Bayesian models are eminently appropriate for investigating rare events in a population: they provide valid inference irrespective of how few observations are collected. Furthermore, Bayesian models provide intuitive estimates of population parameters: researchers conventionally report 89% uncertainty intervals (UIs), which indicate the range of values a parameter will display with 89% probability. Data preparation was conducted using Stata V.17,¹⁶ while we ran models using the ‘rethinking’ package¹⁷ for R statistical software, V.4.4.1.¹⁸

Aim 1

We investigated the prevalence of postneonatal CP over the number of live births in NI across time. For this purpose, we only considered postneonatal CP cases born in NI. We used negative binomial regression models where the exposure corresponded to the log of the number of live births in the period of interest. Postneonatal CP cases are rare and small changes from 1 year to another may produce large variations in summary statistics: to provide stable estimates, we considered totals over groups of triennials (1981–1983; etc). We modelled different trajectories of postneonatal CP over time, comparing models using the Widely Applicable Information Criterion (WAIC).¹⁷

Aim 2

We investigated associations between area deprivation and postneonatal CP over all CP cases using logistic regression models whereby postneonatal CP was regressed on triennials and each case’s area deprivation: we tested models with different time polynomials (from linear to cubic) and then models that allowed the postneonatal CP rate of change to vary depending on area deprivation, comparing models using the WAIC.¹⁷

Aim 3

We investigated associations between deprivation and changes in the postneonatal CP causal categories over time through multinomial logistic regressions, including interaction models whereby the probability of causal categories across time varied depending on deprivation levels. The causal category was missing for n=20 participants (see table 1): thus we created 50 complete datasets using multiple imputation: Bayesian models were fitted on each of these imputed datasets, and their posterior draws combined following recommendations by Zhou and Reiter,¹⁹ see online supplemental material. Finally, we tested associations between area deprivation and each of the two most common causes of postneonatal CP (infections and head injuries) using logistic regression.

Aim 4

We investigated the associations between postneonatal CP and severity of impairments using logistic regressions whereby different types of impairment were regressed on individuals’ area deprivation and the covariates described above. To account for missing data in the covariates and outcomes, we generated 20 imputed datasets using chained equations in Stata V.17. We then used the procedure described in Aim 3, whereby regression models fitted on each imputed dataset provided posterior draws, which we combined to obtain posterior parameter distributions.

Results

Approximately 79% (n=1636) of the cases had been multiply notified, suggesting comprehensive ascertainment. The agreement between authors in categorising causes of postneonatal CP was excellent (84.11 %) and significantly reliable: κ=0.80, Z=21.67, p<0.001.

Aim 1: prevalence of postneonatal CP on all live births over time

In table 2, we report the absolute and relative frequency of postneonatal CP cases by birth years. Postneonatal CP prevalence varied between 3.09 and 0.27 cases per 10 000 live births: across the whole study period there were 1.19 postneonatal CP cases per 10 000 live births (89% UI 1.02 to 1.39). Results indicated a reliable decline in the prevalence of postneonatal CP: the risk ratio for postneonatal CP decreased steadily by 11% every triennial, IRR=0.89 (89% UI 0.85 to 0.93), see online supplemental material S4 for more details.

Table 2. Absolute frequency of postneonatal CP cases by year of birth and relative frequency of postneonatal CP cases by number of live births in NI and by cumulative number of CP cases.

Year of birth	Live births in NI	CP cases per 1000 live births	Postneonatal CP born in NI (n)	Postneonatal CP per 10 000 live births	Postneonatal CP registered in NICPR (n)	All CP cases registered in NICPR (n)	Postneonatal CP by total CP (%)
1981/1983	81 064	2.27	10	1.23	10	194	5.15
1984/1986	82 879	2.17	13	1.57	13	192	6.77
1987/1989	80 998	2.77	25	3.09	25	240	10.42
1990/1992	77 633	2.10	12	1.55	13	188	6.91
1993/1995	72 513	2.17	6	0.83	7	173	4.05
1996/1998	72 137	2.36	6	0.83	8	191	4.19
1999/2001	66 431	2.18	6	0.90	10	173	5.78
2002/2004	65 351	2.30	9	1.38	14	172	8.14
2005/2007	70 051	2.10	6	0.86	10	168	5.95
2008/2010	75 856	1.74	8	1.05	8	145	5.52
2011/2013	74 849	1.66	<5	0.40	<5	140	2.86
2014/2016	73 318	1.02	<5	0.27	<5	82	2.44
Total	893 080	2.07	106	1.19	124	2058	6.03%

The rate of postneonatal CP per 10 000 live births is calculated considering postneonatal CP cases registered in the NICPR and born in NI; the percentage of postneonatal CP cases on the cumulative number of CP cases is calculated considering postneonatal CP cases born and residing in NI at the time of registration into the NICPR.

CP, cerebral palsy; NI, Northern Ireland; NICPR, Northern Ireland Cerebral Palsy Register.

Aim 2: associations between socioeconomic deprivation and probability of congenital or postneonatal CP over time

Model comparisons indicated the trajectories of postneonatal CP differed across individuals’ area deprivation (see online supplemental material S5 for details). Those from the most deprived areas displayed increased odds of postneonatal CP at the start of the study (OR=2.35, 89% UI 1.31 to 3.75), but successively these odds decreased linearly among these individuals (OR=0.87, 89% UI 0.80 to 0.95), whereas the odds of postneonatal CP did not decrease reliably among individuals from less deprived areas (OR=0.96, 89% UI 0.91 to 1.01). Observed and predicted trends are represented in figure 1 (left panel).

Figure 1. Observed and predicted percentage of postneonatal CP by deprivation over time (left panel) and postneonatal CP caused by infection by deprivation over time (right panel). Both models included a linear rate of change moderated by participants’ area deprivation. The round markers represent the observed proportions by deprivation and by time. The straight lines represent the fitted linear predictions. The shaded areas represent the 89% UIs for the fitted lines. Estimates are based on 36 000 samples extracted from the posterior distribution of parameters. CP, cerebral palsy; PNCP, postneonatal cerebral palsy.

Aim 3: changes in postneonatal CP causes over time and their associations with area deprivation

Across the study period, infections were the most frequent primary cause, see table 3: comparisons reported in the table indicated infections were reliably more prevalent compared with other causes. Analyses detailed in online supplemental material S6 suggested the trends in prevalence of postneonatal CP caused by infections differed depending on individuals’ area deprivation: being from less deprived areas was associated with a linear decrease in the odds of postneonatal CP caused by infections (OR=0.84, 89% UI 0.75 to 0.94); conversely, these odds did not change reliably across the study period for those from the most deprived areas (OR=1.10; 89% UI 0.93 to 1.29), see figure 1 (right panel). Analyses also indicated that compared with those from less deprived areas, individuals from the most deprived areas displayed increased odds of postneonatal CP caused by head injuries, on average, across the study period (OR=2.72, 89% UI 1.09 to 5.30).

Table 3. Absolute and relative frequencies postneonatal CP causes (1981–2016).

	n	%	89% UIs		Relative risk of infections vs alternative causes
			LB	UB	RRR	89% UI LB	89% UI UB
Infections	44	35.48	28.19	41.65	–
Head injuries	16	12.90	8.69	17.98	2.83	1.73	4.36
Surgical brain injuries	9	7.26	4.53	11.73	4.92	2.71	8.19
CVAs	11	8.87	5.64	13.57	4.09	2.32	6.61
Hypoxic events	13	10.48	6.83	15.30	3.48	2.05	5.48
Other causes	11	8.87	5.65	13.56	4.08	2.33	6.60
Information not provided	20	16.13	10.88	19.63	2.41	1.58	3.51
Total	124	100.00

The RRRs reported indicate the RRR of postneonatal CP caused by infections rather than postneonatal CP caused by the cause in each row, whereby the cause in each row is taken as the reference category. Therefore, RRRs indicate the relative risk of postneonatal CP caused by infections in comparison to the category indicated in the table.

CP, cerebral palsy; CVAs, cardiovascular accidents; LB, lower bound; RRR, relative risk ratio; UB, Upper Bound; UIs, uncertainty intervals.

Aim 4: association of postneonatal CP with severe impairments, while controlling for biasing effects of area deprivation

Descriptive statistics in table 1 suggested postneonatal CP is associated with increased risk of severe motor impairment (GMFCS level IV–V), intellectual delay and a history of seizures. Bayesian logistic regressions confirmed these findings when we controlled for area deprivation and other confounders, see online supplemental material S7. Compared with congenital CP, postneonatal CP was associated with higher adjusted odds of severe motor impairment (OR=1.47, 89% UI 1.07 to 2.00), intellectual delay (OR=2.54, 89% UI 1.83 to 3.55) and seizures (OR=2.55, 89% UI 1.84 to 3.56).

Discussion

We investigated how area deprivation related to postneonatal CP and its primary causes over three decades (births between 1981 and 2016). The prevalence of postneonatal CP in NI decreased reliably during the study period; furthermore, while CP cases from the most deprived backgrounds had reliably higher odds of postneonatal CP at the start of the study in the 1980s, these odds declined steadily during the study period, thus suggesting decreasing disparity between the most and less deprived groups. However, analyses also indicated persisting disparities in causal categories of postneonatal CP. In particular, while the odds of postneonatal CP caused by infections declined reliably among CP cases from less deprived backgrounds, these did not change reliably over time among cases from the most deprived areas. Furthermore, the odds of postneonatal CP caused by head injuries were consistently higher among cases from the most deprived areas, on average, across the study period. The results confirmed that, when controlling for area deprivation, as well as other covariates, postneonatal CP was reliably associated with increased odds of severe motor impairment (GMFCS level IV–V), intellectual delay and a history of seizures.

Our analyses capitalised on the systematic surveillance of a population within a defined administrative region. The results indicating a reliable steady decline in the prevalence of postneonatal CP over time are consistent with previous studies from high-income countries,^{6 8} even if our register covers a smaller population. This decline (see table 1 and online supplemental material S4) has been linked to the cumulative effect of a series of public health campaigns. For example, during the study period, initiatives have promoted infant immunisation, parental education on the risks of infant sudden death syndrome and improved safety of infant car seats (see online supplemental material S9 for a timeline of key initiatives).

The novelty of our study lies in its ability to estimate the role that area-level socioeconomic deprivation plays in affecting the prevalence of postneonatal CP and its causes over time. Using data linkage, we were able to estimate the relative deprivation of the areas where CP cases were born. We also controlled for the contribution that CP cases could make to the area deprivation measures by excluding health and disability indicators from the area deprivation estimate. The results of Bayesian logistic regressions indicated that at the start of the study period in the 1980s, being from the most deprived areas was associated with higher odds of postneonatal CP. However, there was a marked linear decrease of these odds: differences in the odds of postneonatal CP had reduced from the second half of the 1990s (see figure 1, left panel). These findings suggest public health campaigns may have contributed to reducing disparities in this key health indicator, exerting a normalising effect. The results also confirm postneonatal CP is reliably associated with increased odds of more severe motor impairments, intellectual difficulties and seizures, even when controlling for area deprivation. The reported reduction in the odds of postneonatal CP among cases from the most deprived areas may thus herald reduced disparities in key health outcomes. However, we cannot draw firm conclusions concerning the links between childhood public health initiatives and postneonatal CP: further studies with larger datasets and rigorous causal designs (eg, regression discontinuity in natural experiments²⁰) are necessary.

Our findings also provide nuanced results concerning disparities associated with deprivation: we reported persisting disparities in causal categories of postneonatal CP. Across the study period, infections were the most common cause of postneonatal CP (a finding consistent with other studies^{8 11 12}). The odds of postneonatal CP caused by infections declined reliably among cases from less deprived areas, but the estimated trend did not indicate reliable changes among cases from the most deprived areas; however, there was large variation among these participants (see figure 1, right panel). Considering that postneonatal CP caused by infections was associated with increased adjusted odds of intellectual delay and history of seizures (see online supplemental material S8), the absence of a significant decline of this causal category among people in the most deprived areas is a concern. Nonetheless, we urge caution in interpreting these results because of large variability and the small number of participants: future studies pooling data from population registers are necessary. Since most (72%) of the postneonatal CP caused by infections were linked to infection causes (eg, meningitis caused by pathogens), we can speculate that our results may be linked to disparities in the uptake of infant vaccinations among families from deprived socioeconomic conditions.²¹ However, our data did not allow testing if postneonatal CP cases caused by infections had been vaccinated: future studies will be necessary to test this potential link.

Previous studies have consistently reported associations between postneonatal CP and more severe impairments, but few were able to control for the biasing effect of socioeconomic deprivation. Our results confirm that when adjustments are carried out, postneonatal CP is still reliably associated with increased odds of more severe impairments. The traumatic nature of the causes of postneonatal CP concurs in causing more severe impairments.⁸

Our findings are limited by the small number of cases with postneonatal CP. We used Bayesian approaches, which provide valid inference regardless of the size of the observations. We also accounted for missing data using multiple imputation. However, in similar sparse data, there is a risk of ‘overfitting’, that is, fitting models that explain the observed data but have limited potential to generalise to new data. We attempted to minimise this risk by using stabilising priors in all our models (see online supplemental material 1). Our analyses are also limited by considering an area-level variable (deprivation) as a proxy for individual cases’ background. Future studies shall consider area-level as well as individual-level sources of variability in outcomes. Our indicator of area socioeconomic deprivation provided a ‘snapshot’ of deprivation in 2017, that is, beyond the time frame of the birth years of interest.

Despite its limitations, our study is based on data from an established register over an extended period covering 35 birth years. Key findings are consistent with trends observed across high-income countries, but they also suggest persistent disparities in causal events leading to postneonatal CP, which might contribute to disparities in outcomes and individuals’ health and well-being. The findings have thus relevance for prevention and surveillance, suggesting benefits from promoting targeted preventive postnatal health programmes among families exposed to more deprived conditions.

The funder had no role in the study design, execution, analyses, interpretation of data or decision to submit the results.

Data availability statement

No data are available.