You are on page 1of 25

An update on the prevalence of cerebral palsy: a

systematic review and meta-analysis

Aims
The aim of this study was to provide a comprehensive update
on (1) the overall prevalence of cerebral palsy (CP); (2) the
prevalence of CP in relation to birthweight; and (3) the
prevalence of CP in relation to gestational age.

Method
A systematic review and meta-analysis was conducted and
reported, based on the PRISMA (Preferred Reporting Items for
Systematic Reviews and Meta-analyses) statement. Populationbased studies on the prevalence of CP in children born in 1985
or after were selected. Statistical analysis was carried out using
computer package R, version 2.14.

Results
A total of 49 studies were selected for this review. The pooled
overall prevalence of CP was 2.11 per 1000 live births (95%
confidence interval [CI] 1.982.25). The prevalence of CP
stratified by gestational age group showed the highest pooled
prevalence to be in children weighing 1000 to 1499g at birth
(59.18 per 1000 live births; 95% CI 53.0666.01), although
there was no significant difference on pairwise meta-regression
with children weighing less than 1000g. The prevalence of CP
expressed by gestational age was highest in children born
before 28 weeks' gestation (111.80 per 1000 live births; 95% CI
69.53179.78; p<0.0327).

Interpretation
The overall prevalence of CP has remained constant in recent
years despite increased survival of at-risk preterm infants.

1. Maryam Oskoui1,*,
2. Franzina Coutinho2,
3. Jonathan Dykeman3,
4. Nathalie Jett3and
5. Tamara Pringsheim4
Article first published online: 24 JAN 2013

Abstract
Aims
The aim of this study was to provide a comprehensive update
on (1) the overall prevalence of cerebral palsy (CP); (2) the
prevalence of CP in relation to birthweight; and (3) the
prevalence of CP in relation to gestational age.

Method
A systematic review and meta-analysis was conducted and
reported, based on the PRISMA (Preferred Reporting Items for
Systematic Reviews and Meta-analyses) statement. Populationbased studies on the prevalence of CP in children born in 1985
or after were selected. Statistical analysis was carried out using
computer package R, version 2.14.

Results
A total of 49 studies were selected for this review. The pooled
overall prevalence of CP was 2.11 per 1000 live births (95%
confidence interval [CI] 1.982.25). The prevalence of CP
stratified by gestational age group showed the highest pooled
prevalence to be in children weighing 1000 to 1499g at birth
(59.18 per 1000 live births; 95% CI 53.0666.01), although
there was no significant difference on pairwise meta-regression
with children weighing less than 1000g. The prevalence of CP
expressed by gestational age was highest in children born
before 28 weeks' gestation (111.80 per 1000 live births; 95% CI
69.53179.78; p<0.0327).

Interpretation
The overall prevalence of CP has remained constant in recent
years despite increased survival of at-risk preterm infants.
What this paper adds
This is the first worldwide CP prevalence estimate provided
by meta-analysis.
The overall prevalence of CP worldwide is 2.11 per 1000 live
births.
The prevalence of CP has remained constant in recent
years, despite improved survival of at-risk preterm infants.

Cerebral palsy (CP) encompasses a heterogeneous group of


early-onset, non-progressive, neuromotor disorders that affect
the developing fetal or infant brain.[1] CP is one of the most
common causes of childhood physical disability, with an
estimated lifetime cost, for persons born in the United States in
2000, of 11.5 billion dollars.[2] Accurate prevalence estimates

are needed to guide operational health policies and to aid with


appropriate resource allocation.
In the past decade, two limited reviews on the prevalence of CP
have been published.[3, 4] The first, published in 2007, looked
at both the overall prevalence of CP and the prevalence
reported by birthweight in preterm infants. The overall median
prevalence estimate reported was 2.4 per 1000 live births.
[3] For preterm infants, a median prevalence of 11.2 per 1000
live births was reported among children weighing between
1500g and 2499g at birth, and 63.5 per 1000 live births among
children weighing less than 1500g. This review did not follow
the PRISMA (Preferred Reporting Items for Systematic Reviews
and Meta-analyses) statement and had several methodological
limitations. Only English-language articles were included,
published between 1 January 1990 and 31 January 2005,
searching a single bibliographic database (PubMed). A metaanalysis was not performed; rather, the overall prevalence
estimates of selected studies were combined equally to provide
a median estimate. Studies reporting prevalence estimates with
different denominators (live births, neonatal survivors, children
at a certain age) were combined, and studies including
participants born before 1985 were included.
A second systematic review, published in 2008, examined the
prevalence of CP only in relation to gestational age and
demonstrated a significant decrease in the prevalence of CP
with increasing gestational age.[4] The prevalence ranged from
146 (95% confidence interval [CI] 125170) per 1000 children
born at 22 to 27 weeks of gestation, steadily declining thereafter
to an estimated 62 (95% CI 4978) per 1000 children born at 28
to 31 weeks, 7 (95% CI 69) per 1000 at 32 to 36 weeks'
gestation, and 1.13 (95% CI 0.930.14) per 1000 in term-born
infants. This systematic review used rigorous methodology,

searching four bibliographic databases (MEDLINE, CINAHL,


Cochrane library, and Science Direct) with a clearly stated
search strategy, including studies published from 1985 to 2006
and excluding those with participants born prior to 1985. A
meta-analysis was performed, when possible, to provide pooled
estimates of prevalence; however, the studies used various
denominators, such as live births and neonatal survivors
assessed at various ages, limiting the possibility of such
pooling.
A rigorous systematic review of the overall prevalence of CP is
lacking in the literature. Furthermore, more published data have
become available in recent years on CP prevalence, both
overall and in at-risk populations. The goal of this study was to
provide a comprehensive update on the prevalence of CP and
to systematically review (1) the overall prevalence of CP; (2) the
prevalence of CP in relation to birthweight; and (3) the
prevalence of CP in relation to gestational age.

Method
Search strategy
Two bibliographic databases (MEDLINE and EMBASE) were
searched on 8 February 2011 to identify all potential citations
related to the prevalence of CP published between 1985 and
February 2011. Only studies presenting data on children born
during 1985 or later were included, owing to considerable
changes in perinatal care, and also because a more uniform
definition of CP and routine use of imaging began in 1985.
[5] The search strategy was developed with the help of a health
sciences librarian and adapted for MEDLINE and EMBASE
(Appendix SI, supporting information published online).

References from MEDLINE and EMBASE were combined and


downloaded into a reference manager (Endnote X2). Abstracts
of all references were screened independently by two reviewers
(MO and FC) to select population-based studies on the
prevalence of CP reporting cohorts of all live-born children, all
preterm survivors, or expressed by gestational age or
birthweight. Studies published in both French or English were
included. When cohorts ranging before and after 1985 were
reported, we included studies that allowed for a subgroup
analysis of cohorts born after 1985. Studies reporting on a
single subtype of CP (such as spastic diplegia only), conference
proceedings, interventional studies, review articles, and case
control studies not including full cohorts as cases were
excluded. When multiple articles reporting data from the same
study population were encountered, the most comprehensive
study was selected. References of the two available systematic
reviews were manually searched to identify further potential
references. Two reviewers assessed study quality using a
qualitative tool adapted from previously published scales
(Appendix SII, supporting information published online). Studies
with a quality score below 5 (range 14) were excluded (Table
SI, supporting information published online); for articles
included in the study, the mean score was 6. Disagreement
pertaining to inclusion of articles was resolved by consensus
between the two reviewers.

Data extraction
Two reviewers independently abstracted the following
information on the methods and results from each article
meeting all eligibility criteria: reference, year of publication,
geographical location of the study, birth cohorts included, data
sources, assessment of diagnosis and diagnostic criteria used,
age at diagnosis, as well as the overall number of cases,

population surveyed and denominator used (live births,


neonatal survivors, children at a specified age). Data sources
were also classified as population-based data (e.g. data from
patient registries) or administrative data (e.g. data from
physician billing diagnostic codes or hospital discharge
summaries). Subgroup information for number of cases and
numbers surveyed based on gestational age and birthweight
was also collected.

Data analyses
It was decided a priori that a random effects model would be
used to calculate the pooled prevalence of CP and 95% CIs for
all predetermined groups. To assess for significant betweenstudy heterogeneity, the Cochran's Q statistic was calculated
and I2 was used to quantify between-study heterogeneity;
however, both measures of heterogeneity were used only
descriptively and not in decision-making regarding model
selection.[6] The following prevalence estimates were
calculated: (1) the overall prevalence of CP in children at certain
ages and in all live births; (2) the prevalence of CP in relation to
four birthweight categories for live births (<1000g, 10001499g,
15002499g, and >2500g) and using three birthweight
categories for neonatal survivors (<1500g, 15002499g, and
>2500g); and (3) the prevalence of CP in extremely preterm
newborn infants in relation to five gestational weeks (23wks,
24wks, 25wks, 26wks, and 27wks) and the prevalence of CP in
relation to four gestational age categories (<28wks, 2831wks,
3236wks, and >36wks).
Pairwise meta-regression analysis was performed to assess
whether CP prevalence was significantly different between
groups. Furthermore, to investigate whether the prevalence of
CP had changed over time, meta-regressions were performed

including terms for the start and end years of data collection for
each study. These were chosen as surrogates for year-by-year
prevalence estimates, which were often not provided by the
studies. To address the decision to combine studies that
included postneonatal cases with those that did not, a stratified
analysis was completed to see whether these groups of studies
were different from each other. A cumulative meta-analysis was
conducted on the overall prevalence of CP per 1000 live births
to assess the impact of studies published since the last review
on the overall estimate. Publication bias was assessed using
the Egger regression test, Begg and Mazumdar correlation test,
and a visual inspection of funnel plots. A sensitivity analysis was
also carried out to assess the effect of the methodological
quality on the results.
For all tests, p<0.05 was deemed to be significant. Statistical
analyses were carried out in R version 2.14.[7] The meta
package (version 2.10, 2012 was used to produce the pooled
estimates and forest plots. In accordance with the random
effects model, the studies were weighted according to the
inverse of their variance along with the between-study
heterogeneity (tau-squared) arrived at with a DerSimonian
Laird estimator. The metafor package (installed in R) was used
to conduct the meta-regression using restricted maximum
likelihood estimation.[8]

Results
Selected studies
A total of 1521 references from MEDLINE and EMBASE were
initially identified. A manual search of the references cited in two
available systematic reviews of CP prevalence yielded an

additional 16 abstracts. Duplicates were removed (323) and the


remaining 1214 abstracts were screened, with 91 being
selected for full-text review. Of these, 35 were excluded: two did
not provide data on CP prevalence, one was a letter to the
editor, seven did not provide data for sub-analysis of infants
born in or after 1985, 13 had a quality score of 4 or less, and 12
were in languages other than English or French. The 13 studies
excluded based on quality assessment are described in Table
SI. Of the remaining 56 studies, seven were excluded because
they presented data from the same study population, leaving a
total of 49 studies included for the meta-analysis. The metaanalysis included 29 studies on overall CP prevalence and 20
on CP prevalence in preterm survivors (Table 1). The selection
process details are outlined in Figure 1.
Location

Birth
Data/source
cohor
methodology
t

Ag
ea

Rice et al.[9]

Australia

1993
1998

Administrative
database

Robertson
et al.[10]

Canada

1985
1988

Administrative
database

Smith et al.[11] Canada

1991
1995

Administrative
database

Liu et al.[12]

China

1990
1997

Health care records

07

Ravn et al.[13]

Denmark

1987

Administrative

15

Reference
(a)

Reference

Location

Birth
Data/source
cohor
methodology
t

Ag
ea

(a)
1998

database

Iceland

1990
2003

Administrative
database and other
sources

48

Mongan et al.
[15]

Ireland

1990
1999

Administrative
database and other
sources

Dolk et al. [16]

Northern
Ireland, UK

1987
1997

Multiple sources:
agencies, parents,
MD

>5

Wichers et al.
[17]

Netherland
s

1986
1988

Administrative
database and other
sources

An
y

Andersen et al.
Norway
[18]

1996
1998

Administrative
database

Lie et al.[19]

Norway

1986
1995

Administrative
database

15

Hagberg et al.
[20]

Sweden

1991
1994

Administrative
database

>4

Hagberg et al.

Sweden

1987

Administrative

>4

Sigurdardottir
et al.[14]

Reference

Location

Birth
Data/source
cohor
methodology
t

Ag
ea

(a)
[21]

1990

database

Himmelmann
et al.[22]

1995
1998

Hospital clinical
chart review and
other

48

48

Sweden

Himmelmann
et al.[23]

Sweden

1999
2002

Hospital clinical
chart review and
rehabilitation
reports

Hjern and
ThorngrenJerneck[24]

Sweden

1987
1993

Administrative
database

>1

Nordmark et al.
Sweden
[25]

1990
1993

Hospital/chart
review

NA

Westbom et al.
[26]

Sweden

1990
1997

Administrative
database and other
sources

47

Turkey

1990
2004

Multiple sources,
confirmed by two
MDs

>1

Ozturk et al.
[27]

Location

Birth
Data/source
cohor
methodology
t

Ag
ea

UK

1989
1993

Administrative
database and other

Sinha et al.[29] UK

1985
1987

Administrative
database

Surman et al.
[30]

UK

1987
1995

Administrative
database and other
sources

>1

Pharoah et al.
[31]

Scotland,
UK

1985
1999

Hospital clinical
chart review and
other

USA

1999

Administrative
database and other
sources

Bhasin et al.
[33]

USA

1996,
2000

Hospital clinical
chart review and
other

Petrini et al.
[34]

USA

2000
2004

Administrative
database

05

1986
1991

Administrative
database and other

Reference
(a)
Colver et al.
[28]

Arneson et al.
[32]

Winter et al.
[35]

USA

>3

Reference

Location

Birth
Data/source
cohor
methodology
t

Ag
ea

(a)
sources
Yeargin-Allsopp
USA
et al.[36]

1994

Multiple sources

Table 1. Selected studies. (a) Selected studies on overall


studies on cerebral palsy prevalence in infants born prete
Reference
1.

Location

Birth
cohort

Age Number
a
with CP

Numbe
surveye

Age: age at ascertainment (years). The study by Serdaroglu et al.[37] was exclude
questionnaire for children born in 1996. Out of 41 861 children surveyed, 186 we
aged 216y. MD, physicians; NA, not available.
2.

GA, gestational age; ELBW, extremely low birthweight; BW, birthweight.


(b)
Doyle et al.
[38]

Australia

2005

16

163

Doyle et al.
[39]

Australia

1985
1987

12

95

1997

16

144

Roberts et al. Australia

Location

Birth
cohort

Gray et al.
[41]

Australia

1989
1996

30

293

Sutton et al.
[42]

Australia

1992
1993

42

244

Weber et al.
[43]

Austria

1999
2001

70

248

Robertson
et al.[44]

Canada

1986
2003

65

1515

Vincer et al.
[45]

Canada

1993
2003

66

672

Lefebvre
et al.[46]

Canada

1987
1992

1.5

37

217

Jacobs et al.
[47]

Canada

1990
1994

1.5
2

42

274

Salokorpi
et al.[48]

Finland

1991
1994

27

142

Tommiska
et al.[49]

Finland

1996
1997

1.5

23

208

Beaino et al.
[50]

France

1997

159

1812

Reference

Age Number
a
with CP

Numbe
surveye

[40]

Reference

Location

Birth
cohort

Marret et al.
[51]

France

1997

76

1461

Burguet
et al.[52]

France

1990
1992

22

167

Valleur et al.
[53]

France

1992
1997

17

149

Were and
Bwibo[54]

Kenya

Not
provided

14

120

Leversen
et al.[55]

Norway

1999
2000

26

373

Marlow et al.
[56]

UK and
Ireland

1995

32

241

Vohr et al.
[57]

USA

1993
1998

1.5

606

3785

Figure 1. Flow chart of study selection.

Age Number
a
with CP

Numbe
surveye

Overall prevalence of cerebral palsy


The study by Sinha et al.[29] was excluded from analysis
because it focused on a small ethnic community with a high rate
of consanguinity that did not represent a true population-based
estimate in the study region. Pooling 19 studies that reported
estimates using live births as the denominator, the overall
prevalence of CP was 2.11 per 1000 live births (95% CI 1.98
2.25; Fig. 2a). The cumulative meta-analysis demonstrates that
the overall prevalence of CP has remained stable with the
addition of new studies over the past 10 years (Fig. 2b). There
was no significant difference in overall prevalence per 1000 live
births between studies using population-based data (1.93; 95%
CI 1.692.21) and studies using administrative data (2.19; 95%
CI 2.022.37). A meta-regression sorting the studies by start
year of each cohort and end year of cohort was also conducted,
and showed no significant difference by either start (p=0.34) or
end year (p = 0.54). When assessing the importance of
including postneonatal cases, the overall estimates of the two
groups of studies were almost identical. Studies including these
cases had a prevalence of 2.10 (1.932.30) and those
excluding the cases had a prevalence of 2.11 (2.032.19). The
overall prevalence appeared higher using studies reporting
children at a specific age (range 18mo8y) as the denominator
(Fig. S1, supporting information published online), with a pooled
prevalence of 2.91 per 1000 children (95% CI 2.163.93).

Figure 2. Overall prevalence of cerebral palsy per 1000 live


births.

Prevalence of cerebral palsy by birthweight


The pooled prevalence of CP in children per 1000 live births
was calculated by birthweight category (Fig. 3). The prevalence
was highest in children weighing 1000 to 1499g (59.18 per 1000
live births; 95% CI 43.3873.95), and lowest in children
weighing over 2500g (1.33 per 1000 live births; 95% CI 1.19
1.49). The prevalence among children weighing under 1000g
was not significantly different from the prevalence among those
weighing 1000 to 1499g (p=0.8159), but in both cases was
significantly higher than among children weighing 1500 to
2499g (p<0.001). The prevalence among children weighing over
2500g was significantly lower than in all other birthweight
groups (p<0.001).

Figure 3. Prevalence of cerebral palsy (CP) in relation to


birthweight. (a) Birthweight less than 1000g; (b) birthweight
between 1000g and 1499g; (c) birthweight between 1500g and
2500g; (d) birthweight over 2500g.
Three studies reported the prevalence of CP in relation to
birthweight per 1000 neonatal survivors: 60.04 (95% CI 45.21
81.23) in children under 1500g, 8.33 (95% CI 5.3313.03) in
children weighing between 1500g and 2499g, and 1.16 (95% CI
1.041.29) in children weighing over 2500g. The prevalence of
CP is significantly lower among children weighing between
1500g and 2499g than in those weighing under 1500g
(p<0.001). Among children with a birthweight over 2500g the

prevalence of CP was significantly lower than in all other groups


(p<0.001).

Prevalence of cerebral palsy in relation to


gestational age
The pooled prevalence per 1000 live births was calculated for
each gestational age group (Fig. 4). The prevalence was
highest, at 111.80 (95% CI 69.53179.78), among children born
before 28 weeks of gestation and lowest, at 1.35 (95% CI 1.15
1.59), for children born after 36 weeks. Pairwise metaregression showed that the prevalence of CP was significantly
lower among children born between 28 and 31 weeks' gestation
than in children born before 28 weeks' gestation (p=0.0327).
Prevalence was significantly lower among children born
between 32 and 36 weeks' gestation than among those born
more preterm (p<0.001), and was significantly lower among
children born after 36 weeks' gestation than in all other groups
(p<0.001). There were fewer studies reporting the prevalence of
CP in neonatal survivors in relation to gestational age group.
The prevalence was 111.80 (95% CI 69.53179.78) per 1000
neonatal survivors born before 28 weeks and 144.72 (95% CI
63.71328.71) per 1000 neonatal survivors born between
28 weeks and 31 weeks; the number of studies was too small to
calculate a pooled prevalence in older gestational age groups.

Figure 4. Prevalence of cerebral palsy (CP) in relation to


gestational age per 1000 live births. (a) Gestational age less than
28 weeks; (b) gestational age between 28 weeks and 31 weeks;

(c) gestational age between 32 weeks and 36 weeks; (d)


gestational age over 36 weeks.

Prevalence of cerebral palsy per gestational


week in extremely preterm infants
The pooled prevalence per 1000 neonatal survivors at a
predetermined age was calculated per gestational week from 23
to 27 weeks (see Fig. S2, supporting material published online).
The pooled prevalence was highest in children born at
23 weeks' gestation (261.42; 95% CI 140.52486.34) and
lowest in children born at 27 weeks' gestation (102.70; 95% CI
66.44158.77). Pairwise meta-regression showed that the
prevalence among children born at 27 weeks' gestation was
significantly lower than that among children born at 23 weeks'
gestation (p=0.0063).

Evaluation of potential bias


For the overall prevalence estimate of CP per 1000 live births,
no significant publication bias was detected on visual inspection
of funnel plots of the standard error by the logit of the
prevalence rate and supported by statistical testing. For the
Begg and Mazumdar rank correlation test, Kendall's tau-b is
0.15584 (p-value 0.15502). Egger's regression test showed an
intercept of 2.49981 (95% CI 0.92154 to 5.92115),
with t=1.52411, df=20 (p-value 0.07157). For the prevalence
estimate of CP in preterm survivors, no significant publication
bias was detected on visual inspection of funnel plots of the
standard error by the logit of the prevalence rate and supported
by statistical testing. For the Begg and Mazumdar rank
correlation test, Kendall's tau-b is 0.19883 (p-value 0.11712).
Egger's test showed an intercept of 1.02719 (95% CI
4.01953 to 1.96515), with t=0.72424, df=17 (p-value 0.23938).

In our selection process, 12 studies were excluded based on


language: three were written in Spanish, five in Chinese, one in
Polish, two in Japanese, and one in German. Of these, two
presented data on live births, five presented data on children at
a particular age, of which three reported on the same
population, and four presented data on preterm survivors.

Discussion
Main findings
This study, through a systematic review and meta-analysis of
the available literature, provides an updated estimate of the
overall prevalence of CP, as well as its relationship with
gestational age and birthweight. The most commonly reported
denominators were live births, neonatal survivors, and survivors
at a predetermined age. Overall prevalence estimates were
similar using live births or neonatal survivors; however, in the
preterm subgroup the estimates are expected to differ as
perinatal mortality is higher among preterm infants have in termborn newborn infants. There were fewer studies reporting the
prevalence of CP per neonatal survivors in relation to
gestational age group. The pooled estimate for children born at
gestational age 28 to 31 weeks is unexpectedly higher than
children born at a gestational age of less than 28 weeks;
however, there were fewer studies reporting the former group
and the pooled estimate shows a larger confidence interval, and
this difference was not significant in a pairwise meta-regression
analysis. Most studies reporting estimates on preterm survivors
were based on population-based registries following groups of
infants born preterm. For overall prevalence, sources of case
ascertainment varied considerably between studies. Most

European studies used population-based patient registries with


multiple sources of case ascertainment and uniformly applied
diagnostic inclusion criteria. In North America, several states
reported data from population-based registries; however, both
Canadian studies used administrative databases with ICD-9
codes as diagnostic criteria. Prospective population-based
patient registries are favoured as more validated data sources,
but they are associated with higher operational costs than large
pre-existing administrative databases. It is encouraging,
however, that estimates from population-based studies and
administrative data did not differ significantly in our analysis.
Although prevalence estimates were comparable across
geographic sites and cultural groups, one study that was
excluded from analysis reported prevalence estimates that were
much higher.[29]This study focused on a British ethnic
community (Bradford District Health Authority) with a high rate
of consanguinity, highlighting the potential genetic aetiology in
this population.
The overall prevalence of CP estimate in our review is 2.11 per
1000 live births, consistent with the prevalence reported by an
earlier review without meta-analysis.[3] This prevalence was
also shown to have remained constant in recent years with the
addition of new published studies. A number of factors may
contribute to an increased prevalence of CP, such as a
improved survival in preterm infants and higher numbers of
multiple births, which often result in preterm births. There are
also a number of factors that may contribute to a decreased
prevalence of CP, such as the use of antenatal corticosteroids,
cooling for term-born asphyxiated infants, and the use of
magnesium sulphate. Contributing genetic aetiologies may
either be increasing or decreasing. The analysis of CP based
on birthweight showed, as expected, that the prevalence
decreases significantly among children born with a birthweight

above 1500g. This prevalence estimate was similar irrespective


of the denominator used in the calculation, that is per 1000 live
births or per 1000 neonatal survivors. One of the reasons for
this similarity in prevalence estimates could be the small
number of high-quality studies included that report prevalence
of CP using neonatal survivors as the denominator and
inclusion of term-born low-birthweight infants, among whom
mortality may be lower. As reported in an earlier meta-analysis,
the prevalence of CP shows an inverse relationship to
gestational age.[4] The prevalence estimate for preterm births
varied significantly based on the denominator, though a decline
in CP prevalence was observed with increased gestational age.
[3, 4] When reported as a proportion of neonatal survivors, the
prevalence estimates were, as expected, larger than the
corresponding estimates reported as a proportion of live births,
although these proportions revealed a similar declining trend
with increasing gestational age.

Strengths and limitations


This review is based on rigorous methods and quality
assessment of the selected studies. Efforts were made to report
the meta-analysis based on the PRISMA statement.
[58] Additionally, analysis using the random effects model
aimed to account for heterogeneity between studies. In
comparison with previous reviews, our meta-analysis analysed
a greater number of studies, used only data from study cohorts
born in or after 1985 to account for changes in clinical practice,
and computed pooled prevalence estimates based on
comparable common denominators. All efforts were made to
adjust for potential bias.
Though many studies used administrative data, the diagnosis of
CP based on ICD-9 coding has not, to our knowledge, been

validated in these databases. Administrative databases are


susceptible to limitations such as inaccuracy in diagnostic code
entry, resulting in possible misclassification bias, multiple
entries from varying sources, a lack of universal case
definitions, and financial incentives in coding in some countries.
[59] North American studies are also restricted to a few
provinces or states, with questionable generalizability of the
available data to the rest of the country, where demographic
and social variables may differ.
There was significant heterogeneity across studies, and an
attempt was made to adjust for this by using a random effects
model and by pooling only studies with similar denominators.
There were several factors contributing to potential selection
bias in our review. First, there is a potential ascertainment bias
by individual studies using different definitions and age at
diagnosis of CP without reference to functional severity.
Mortality in the first years of life would lead to exclusion of more
severe cases, while a late age of diagnosis or ascertainment
only through rehabilitation facilities may exclude milder cases.
This may have had an impact on the numerator, though the
extent of impact cannot be established. Furthermore, estimating
the impact on health care costs and planning of services would
need to take into consideration the severity of dysfunction in
affected children, which is not provided in most studies. A
second potential source of selection bias is referral filter bias, as
some studies reported cases from academic institutions, which
may impose a bias in access to care. In this meta-analysis, data
from 8 of the 48 studies were hospital based. These hospitalbased studies were included as the institutes were considered
to be centres of excellence and representative of the population
they served.

A third potential source of selection bias is publication bias.


Although efforts were made to make the search strategy as
comprehensive as possible, the search did not include
unpublished material or data from the grey literature. However,
based on inspection of funnel plots and supported by statistical
testing, we did not find significant evidence of publication bias
for either the overall prevalence estimate or the estimate in
premature survivors.
The fourth potential source of selection bias in this review is
language bias, as only studies published in English and French
were selected. The impact of excluding studies published in
languages other than English has been shown to have
generally little effect on summary treatment effect estimates.
[60] In our selection process, 12 studies were excluded based
on language. A language bias cannot be excluded.

Conclusions
This meta-analysis provides updated prevalence estimates of
CP across different populations, showing a constant overall
estimate per live births in recent years despite the addition of
new data and increased survival of at-risk preterm infants.
Although the prevalence of disease is unchanged, future
studies are needed to evaluate a change over time in the
phenotypic spectrum of CP. Studies on the prevalence of CP in
adolescence and adulthood would also be important future
goals as the majority of children with CP are expected to
survive into adulthood. Further studies are also needed in North
America and Canada, as the available data are restricted to a
few regions, principally regions in Canada. Direct comparison of
case ascertainment from patient registries and administrative

databases is needed to establish the validity of the diagnosis of


CP in the latter, as administrative databases are increasingly
used for surveillance because of their larger sampling frame,
lower operational cost, and availability of longitudinal data.
Accurate and updated prevalence estimates are imperative in
estimating the burden of illness and planning appropriate health
resource allocation for this vulnerable patient population.

Acknowledgements
This study was funded by an operational grant from the Public
Health Agency of Canada (PHAC) in conjunction with the
Neurological Health charities of Canada (NHCC).The PHAC
and the NHCC did not participate in the design and conduct of
the study; in the collection, analysis, and interpretation of the
data; or in the preparation, review, or approval of the
manuscript.

Disclosures
N Jett is an Alberta Innovates Health Solutions Population
Health Investigator and holds a Canada Research Chair Tier 2
in Neurological Health Services Research. T Pringsheim
receives salary/research support from Alberta Health and
Wellness, the Canadian Institute of Health Research, and the
Public Health Agency of Canada.