- Top of page
- What this paper adds
Aim This meta-analysis provides a systematic review of studies into intellectual and attentional functioning of paediatric brain tumour survivors (PBTS) as assessed by two widely used measures: the Wechsler Intelligence Scale for Children (3rd edition; WISC-III) and the Conners’ Continuous Performance Test (CPT).
Method Studies were located that reported on performance of PBTS (age range 6–16y). Meta-analytic effect sizes were calculated for Full-scale IQ, Performance IQ, and Verbal IQ as measured by the WISC-III, and mean hit reaction time, errors of omission, and errors of commission as measured by the CPT. Exploratory analyses investigated the possible impacts of treatment mode, tumour location, age at diagnosis, and time since diagnosis on intelligence.
Results Twenty-nine studies were included: 22 reported on the WISC-III in 710 PBTS and seven on CPT results in 372 PBTS. PBTS performed below average (ps<0.001) on Full-scale IQ (Cohen’s d=−0.79), Performance IQ (d=−0.90), and Verbal IQ (d=−0.54). PBTS committed more errors of omission than the norm (d=0.82, p<0.001); no differences were found for mean hit reaction time and errors of commission. Cranial radiotherapy, chemotherapy, and longer time since diagnosis were associated with lower WISC-III scores (ps<0.05).
Interpretation PBTS have seriously impaired intellectual functioning and attentiveness. Being treated with cranial radiotherapy and/or chemotherapy as well as longer time since diagnosis leads to worse intellectual functioning.
In the USA the incidence of cancer in children aged 0 to 14 years is almost three per 100 000.1 Approximately 17 to 22.5% of children with cancer have a brain tumour.1,2 Advances in medicine have led to an increasing number of children surviving cancer. The 5-year survival rate of children diagnosed with a brain tumour under the age of 15 increased from 57% for patients diagnosed from 1975 to 1977 to 74% for patients diagnosed from 1996 to 2004.3
With more children becoming long-term survivors, the need has grown to understand fully the nature and magnitude of the late effects of the tumour and treatment. Compared with survivors of other malignancies, survivors of brain tumours in childhood bear the greatest risk of neurocognitive impairment.4 Numerous studies have shown that 40 to 100% of paediatric brain tumour survivors (PBTS) show some form of neurocognitive deficit.5 Frequently reported impairments in PBTS are declining levels of general intelligence and attention deficits. Deficits in these areas can have a deleterious effect on academic achievement and psychosocial functioning.6–8
Besides the burden of the tumour itself, the treatment can contribute to neurocognitive impairments. Radiotherapy is especially considered to have an impact on neurocognitive functioning.9,10 Chemotherapy, however, has also been found to be associated with poor outcomes in PBTS.11 In addition to the treatment, tumour location can affect the neurocognitive outcome of PBTS, with infratentorial tumours being associated with worse outcomes than supratentorial tumours.12 Furthermore, age at diagnosis is known to have an impact on neurocognitive outcome.13 The young brain is especially vulnerable to the adverse effects of treatment because of the rapid cell proliferation, dendritic and axonal outgrowth, as well as myelination, which take place during infancy, childhood, and adolescence. Therefore, radiotherapy is postponed or omitted in most protocols if the child is under the age of 3 years.11 In addition, time since treatment is an important determinant of neurocognitive deficits, as the deficits often increase over time, owing to a slower rate of acquiring new skills and knowledge compared with healthy peers.13,14
The current paper reports the results of a quantitative meta-analysis, investigating the magnitude and consistency of neurocognitive deficits in PBTS. Analysis of the literature determined general intelligence and attention as two frequently studied areas of neurocognitive functioning. General intelligence provides insight into the generic cognitive functioning of the patient and is measured most often using the Wechsler Intelligence Scale for Children (3rd edition; WISC-III).15 Attention is required to some extent for nearly all components of neurocognitive functioning and is therefore a crucial area to study thoroughly. The Conners’ Continuous Performance Test (CPT, CPT II) is the most widely used measure for attention.16,17 Besides intelligence and attention, processing speed and working memory are often studied in PBTS. These areas are important in understanding the neurocognitive functioning of a patient; they are, however, beyond the scope of this meta-analysis, which focuses on the two key areas: intelligence and attention. The CPT comprises a measure for processing speed; therefore, this area is reported as well. Additionally, exploratory analyses investigated the possible impact of cranial radiotherapy, chemotherapy, tumour location, age at diagnosis, and time since diagnosis on general intelligence.
- Top of page
- What this paper adds
Figure 1 shows the selection of studies in a flowchart. Twenty-nine studies met inclusion criteria. Twenty-two studies reported scores on the WISC-III for a total of 710 PBTS.9,12,23–26,29,34,36–49 Seven studies reported CPT results for a total of 372 PBTS.32,33,50–54 When two or more studies reported on the same participants, we included the most recently published study to prevent erroneously inflated homogeneity of meta-analytic results. Grill et al.44 and Kieffer-Renaux et al.47 report partly on the same participants. The most recent publication by Grill et al. reports on PIQ and VIQ, but does not report on FSIQ. The earlier publication of Kieffer-Renaux et al., however, does report on FSIQ. Therefore, the study by Grill et al. was included in the meta-analysis of PIQ and VIQ, whereas the study by Kieffer-Renaux et al. was included in the meta-analysis of FSIQ.
Figure 1. Flow chart of study selection n, number of studies; PBTS, paediatric brain tumour survivors; WISC Wechsler Intelligence Scale for Children; CPT, Conners’ Continuous Performance Test.
Download figure to PowerPoint
For five of the 22 WISC-III studies we aggregated data on two or more PBTS subgroups into one mean and SD per study: (1) Patel et al.12 compared PBTS according to their tumour location; (2) Callu et al.39 compared patients with low-grade gliomas and malignant cerebellar tumours; (3) Lacaze et al.24 studied three samples of patients with optic pathway tumours who received three different treatments; (4) Kieffer-Renaux et al.47 compared patients with medulloblastoma who received two different doses of radiotherapy; and (5) Mulhern et al.48 compared patients with medulloblastoma with those having low-grade glioma.
Table I displays details of the studies incorporated in this meta-analysis. Some studies reported insufficient details to allow calculation of effect sizes. In these cases, authors were contacted to provide the missing data.50,32–34,51 For some studies, data were unavailable on one or more of the dependent variables, leading to unequal numbers of studies for these dependent variables.
Table I. Characteristics of studies included in the meta-analysis
| ||Study||Number of participants and diagnosis||Treatment||Location||Time/age||Quality|
|% cRT||% Chemo||% Infra||Mean age at dx||Time since dxa||NOS score|
|WISC-III||Hazin et al.49||13 LGG, 7 MB||35||35||100||8.3||1.9||100|
|Patel et al.12||70 BT||76||71||49||7.8||3.4||100|
|Saury and Emanuelson36||8 MB||100||100||63||7.8||5.1||50|
|Sands et al.37||24 BT||29.2||100||65.4||3.0||3.3||100|
|Aukema et al.29||6 MB||100||100||100||4.7||8.9||50|
|Bonner et al.38||101 BT||74||NA||56||7.0||3.9||100|
|Callu et al.39||20 HGG, 19 LGG||44||36||100||5.4||3.4||100|
|Briere et al.40||12 MB, 6 GNS||94||78||NA||6.3||NA||0|
|Ris et al.26||83 LGG||0||0||16||NA||NA||100|
|Sanders et al.41||5 HGG||80||100||40||0.9||11.4||100|
|Jalali et al.23||7 LGG||100b||NA||<42||NA||NA||100|
|Khong et al.42||12 MB||100||100||100||8.5||3.4||50|
|Beebe et al.43||92 LGG||0||0||100||8.1||0.4||100|
|Grill et al.44||76 PF||100||72||100||5.7||6.1||50|
|Spiegler et al.34||34 MB, EP||100||70||100||5.5||2.5||100|
|Lacaze et al.24||21 LGG||38||100||0||NA||NA||100|
|Carey et al.46||15 BT||60||53||NA||NA||NA||100|
|Kieffer-Renaux et al.47||36 MB||100||100||100||8.0||5.0||50|
|Merchant et al.25||8 GCT||100||0||0||NA||NA||100|
|Grill et al.9||19 MB, EP||100||NA||100||6.1||5.3||100|
|Mulhern et al.48||18 MB, 18 LGG||50||50||100||NA||NA||100|
|CPT||Butler et al.50||131 BTNS||NA||NA||NA||6.5||5.3||50|
|Mabbott et al.51||64 MB, EP||50||NA||50||5.8||5.6||100|
|Conklin et al.32||61 BTNS||NA||NA||NA||6.5||5.0||50|
|Stargatt et al.52||16 BT||62||69||100||9.9||4.1||100|
|Reeves et al.53||38 MB||100||100||100||8.3||2.0||100|
|Mulhern et al.54||37 BT||100||49||62||NA||NA||100|
|Mulhern et al.33||25 MB||100||NA||100||8.2||5.2||100|
Figures 2 and 3 display the studies’ effect sizes as well as the overall effect sizes for each of the dependent variables and the accompanying 95% confidence intervals. Effect sizes of all dependent variables were heterogeneously distributed and a random effect analysis was used in all analyses. There was no significant association between the study quality ratings and effect sizes (all ps>0.08) for any of the dependent variables.
Figure 2. Wechsler Intelligence Scale for Children, 3rd edition (WISC-III) study results. CI, confidence interval; FSIQ, Full-scale IQ; PIQ, Performance IQ; VIQ, Verbal IQ.
Download figure to PowerPoint
Figure 3. Conners’ Continuous Performance Test (CPT) study results. Higher scores indicate worse performance for all three dependent variables. CI, confidence interval; EC, errors of commission; EO, errors of omission; MHRT, mean hit reaction time.
Download figure to PowerPoint
Wechsler Intelligence Scale for Children-III
PBTS had lower FSIQ scores than their peers, as indicated by a combined random effect size of d=−0.79 (p<0.001), translating into a large effect. Of the 21 studies that reported on FSIQ, 14 reported scores significantly (p<0.05) below the average FSIQ of 100,9,12,26,34,36–41,45–48 whereas none of the studies reported scores significantly higher than average.
PIQ scores were significantly lower in PBTS than in the normative sample, as indicated by a combined random effect size of d=−0.90 (p<0.001), again translating into a large effect. Fifteen of the 19 studies found PIQ scores of PBTS significantly (p<0.05) below average,9,12,24,26,34,36–40,43,44,47–49whereas in none of the studies were scores significantly higher than average found. [Correction added on 27 December 2012, after first online publication: effect size of PIQ corrected to d=0.90].
VIQ scores of PBTS were significantly below average. The combined random effect size was d=−0.54 (p<0.001), which represents a medium effect size. Eleven of the 19 studies reported scores that were significantly (p<0.05) below the mean.9,12,23,34,36–38,40,44,47,48 Eight studies reported VIQ scores that did not differ significantly from the mean of the normative sample.24–26,29,39,42,43 The combined effect size for PIQ was significantly higher than the combined effect size for VIQ (d=−0.29, p<0.001), indicating greater impairments in PIQ than in VIQ.
There was no evidence for publication bias for any of the WISC-III measures, as we found high fail-safe N values and non-significant (ns) positive correlations between sample size and effect size (FSIQ: fail-safe N=871, r=0.36, ns; PIQ: fail-safe N=1030, r=0.16, ns; and VIQ: fail-safe N=406, r=0.25, ns).
Conners’ Continuous Performance Test
The studies were ambiguous about the scores of PBTS on MHRT of the CPT. Three of seven studies found significantly slower MHRT,51,53,54 whereas two studies found responses of the PBTS to be significantly faster than average.32,33 Two other studies found PBTS scores in the average range.50,52 Across studies a non-significant combined random effect size of d=0.15 was found for MHRT.
The number of errors of omission on the CPT committed by PBTS was higher than the normative sample, as indicated by a combined random effect size of d=0.82 (p<0.001), which is considered to be a large effect. All but one study found significantly higher errors of omission rates in PBTS than the normative sample.32,33,51–53 Fail-safe N was 64 and there was a positive non-significant correlation between sample sizes and effect sizes (r=0.85), together indicating that there was no evidence for publication bias.
PBTS did not differ from the normative sample on the number of errors of commission, as indicated by a non-significant combined random effect size of d=0.03. Five of seven studies found no performance differences between PBTS and the normative sample;32,33,50,51,54 one study reported PBTS to make fewer errors of commission than the normative sample,52 and another study found that more errors were made by the PBTS than the normative sample.53
Table II reports the results for the meta-regression analysis for the five moderating variables. Cranial radiotherapy was a strong predictor of lower intellectual functioning, accounting for 26%, 32%, and 19% of the variance in FSIQ, PIQ, and VIQ respectively, with cranial radiotherapy leading to lower scores as opposed to no cranial radiotherapy. Chemotherapy accounted for 22% of the variance in FSIQ and 29% of the variance in PIQ scores, with chemotherapy leading to lower scores as opposed to no chemotherapy. There was no association between chemotherapy and VIQ. Furthermore, we found no predictive value of tumour location or age at diagnosis for intelligence scores. Longer time since diagnosis, however, was highly predictive of lower scores on all WISC-III scales, accounting for large proportions of the variance (FSIQ 41%; PIQ 44%; VIQ 25%). As expected, there was a strong association between cranial radiotherapy and chemotherapy (r=0.54, p<0.05), and between age at diagnosis and time since diagnosis (r=−0.66, p<0.05), not allowing us to distinguish between the effects of these moderating variables.
Table II. Meta-regression analyses, Wechsler Intelligence Scale for Children (3rd edition) studies
| n || β || R2 || p || n || β || R2 || p || n || β || R2 || p |
|Age at diagnosis||15||0.40||0.16||0.061||14||0.15||0.02||0.568||14||0.35||0.12||0.114|
|Time since diagnosis||14||−0.64||0.41||<0.001||13||−0.67||0.44||<0.001||13||−0.50||0.25||0.014|
- Top of page
- What this paper adds
This meta-analysis summarized neurocognitive functioning of 710 (WISC-III) and 372 (CPT, CPT II) PBTS. We found substantial impairments in intellectual functioning and attentional abilities. PBTS scored on average −0.54SD to −0.90SD lower on the WISC-III scales than the normative sample, with PIQ scores being even more depressed than VIQ scores. The number of PBTS in this meta-analysis that were at grade level and succeeding academically is unknown. However, a large body of research has demonstrated that intellectual functioning, as assessed with intelligence tests, is a powerful predictor of academic achievement and vocational success.55,56 In PBTS this association was substantiated by Reddick et al.,57 who found a positive relation between intelligence scores of PBTS, as measured with the Wechsler intelligence scales, and academic achievement, as measured with the abbreviated Wechsler Individual Achievement Test. Moreover de Boer et al.58 report academic failure and lower intelligence to be risk factors for decreased employment rates in PBTS, with PBTS being nearly five times more likely to be unemployed than healthy peers.
This meta-analysis found performance intelligence to be more vulnerable to the detrimental effects of a brain tumour and its associated treatment than verbal intelligence, with the observed PIQ scores in this meta-analysis being on average −0.29SD lower than the VIQ scores. Similar findings have been reported in other populations at risk for brain damage, including children born very preterm and those with traumatic brain injury.59,60 The discrepancy between PIQ and VIQ findings in PBTS might be related to the high rate of cerebellar tumours in this group, as half of the paediatric brain tumours are located within the posterior fossa.34 PIQ subtests are of a multi-component nature and draw more heavily on motor functions, visuomotor integration, visual attention, abstract reasoning, and working memory than the VIQ subtests. Information about how many PBTS were excluded from the analyses because of a persistent motor or sensory disability was lacking. However, it is well known that many of these neurocognitive functions depend heavily on cerebellar functioning.61–63 Therefore damage to the cerebellum is expected to be related to depressed PIQ scores. Secondly, lower PIQ than VIQ scores may be related to a general slowing in information processing resulting from white matter damage. Compared with VIQ subtests, many PIQ subtests draw heavily on processing speed owing to their timed character. Although grey matter development peaks during childhood, white matter continues to develop gradually until early adulthood.59 The maturation of white matter is therefore challenged in PBTS because cancer therapy damages the healthy cells of the central nervous system. Glial progenitor cells are especially vulnerable to the effects of chemotherapy and radiotherapy.64 These cells are responsible for the formation of oligodendrocytes and astrocytes, both myelinating cells that are crucial for white matter integrity. White matter integrity plays a major role in information processing speed with decreased white matter integrity, resulting in slower processing speed.65
In the present meta-analysis, severe attentional problems in PBTS were revealed, as indicated by the high mean number of errors of omission committed by PBTS on the CPT compared with the normative sample. In addition to the errors of omission of the CPT, attention problems in PBTS have been reported using other measures of attention. PBTS perform below the norm on the Freedom of Distractibility Index, the attentional factor of the WISC-III.40 Also, measuring attention using the Gordon Diagnostic System, PBTS scores on focused attention have been reported to be almost 1SD below the mean of the normative sample and more than 1SD below the mean on selective attention.66 Other researchers have used the Conners’ Rating Scale for Parents and Teachers to identify attention deficits in PBTS; they found more than half of the patients obtained scores above the 75th centile, indicating attentional difficulties.67 Attentional abilities develop early in life and form the basis for other emerging and proliferating neurocognitive functions.59 As a result, attentional abilities are an important prerequisite for scholastic development and are strongly associated with academic achievement.59 Reddick et al.57 found worse attentional functioning in PBTS, measured with the CPT, to be associated with lower academic achievement scores, as measured with the abbreviated Wechsler Individual Achievement Test.
Interestingly, this meta-analysis found no evidence for slow processing speed in PBTS as measured with MHRT on the CPT. Individual studies revealed conflicting findings, with some researchers reporting worse performance of PBTS than the normative sample and others reporting no difference or even better performance by PBTS. Our failure to find evidence for slowed information processing as measured by the MHRT on the CPT does not fit with the frequently observed impairments in white matter integrity in PBTS. Such impairments are expected to result in slower processing speed.65 Perhaps MHRT is a limited measure for processing speed. It is possible that slowed information processing becomes evident only in tasks that place demands on the integration of multiple stimulus modalities and involve more complex mental operations, which would engage widespread neural networks in the brain. Speed on such complex tasks, for example the performance subtests of the WISC-III, draw more heavily on white matter integrity than simple stimulus response tasks such as the CPT. Indeed, several researchers have reported decreased scores of PBTS on the processing measures obtained with more complex tasks such as the speed index of the Wechsler Scale and the Trail Making Task.12,29,38
Inhibitory control seems to be spared in PBTS, as the number of errors of commission on the CPT was in the average range. This result converges with results of earlier studies into inhibitory control, which failed to find evidence for inhibitory problems.68,69 The absence of an effect on inhibitory control may be explained in terms of tumour location. Inhibition of responses is primarily mediated by the prefrontal cortex, and paediatric brain tumours in the frontal lobe are rare.59,70 On the other hand, there are a multitude of connections between the cerebellum and the frontal lobes, which can be damaged by local radiotherapy, so causing inhibitory problems.61 Also, frontal lobes are irradiated in craniospinal radiotherapy, for example for medulloblastomas. Indeed Aukema et al.29 found vulnerability of the white matter in the frontal lobes in survivors of medulloblastoma, which was associated with slower processing speed. The fact that we did not find inhibitory control problems might also be caused by the relatively slow development of the prefrontal cortex at the ages the patients were tested.59
As described above, cancer treatment inevitably causes cell damage, challenging the normal maturation and myelination of the neural pathways in the young brain, and consequently challenging the development of cognitive, motor, behavioural, and emotional functioning.42,57,71,72 In our exploratory analyses we found that cranial radiotherapy and chemotherapy are indeed strong predictors of worse intellectual functioning. In our analyses it was impossible to make the distinction between different doses and volumes of radiotherapy, but it is known from the literature that higher doses and larger volumes are associated with worse neurocognitive outcomes.5 The combination of radiotherapy and chemotherapy in adults is believed to play an important role in neurocognitive deficits, owing to the induced damage to neural progenitor cells for hippocampal neurogenesis and the maintenance of subcortical white matter integrity.64 Moreover, we found that longer time since diagnosis is highly associated with worse intellectual outcomes. This finding is probably explained in terms of the neurocognitive problems such as reduced attentiveness, slower processing speed, and memory problems exhibited by PBTS that challenge the learning process, causing an increasing gap between PBTS and their peers in intellectual functioning. No effect was found for tumour location on intellectual functioning. Although the adverse effects of the tumour and its treatment are considered detrimental for the developing brain, interestingly no relationship was found between mean age at diagnosis and WISC-III scores of PBTS. Longer follow-ups might reveal effects of age at diagnosis.
This meta-analysis has some limitations that need to be taken into account. A limited number of studies were available for this meta-analysis. Also, the available studies did not allow a distinction to be made between the different brain tumour diagnoses, tumour locations, and treatment intensities. This may have contributed to heterogeneity in the study findings. For the quality analysis, we used the most relevant quality rating that we found. Nevertheless, not all criteria were applicable to the included studies, which potentially decreased the reliability of the quality ratings. Furthermore, it has been reported in the literature that there is an effect of sex in outcome, to the disadvantage of female children.72 However, owing to lack of variability in proportions of male and female children in the included studies, it was not possible to include sex in the analyses. It would have been interesting to compare scores of patients on the WISC-III and the CPT; unfortunately none of the included studies reported on both a full WISC-III and the CPT. In addition, with one single exception, the included studies used an uncontrolled study design or compared PBTS with other patient groups.29 Also, working memory and other important cognitive areas have not been addressed in this meta-analysis, despite their interdependence with intelligence, attention, and processing speed. Even though the WISC-III and the CPT are well-validated tests and normative data are available, the use of a healthy comparison group, matched on demographic background characteristics, would have been preferable.