The impact of therapy for childhood acute lymphoblastic leukaemia on intelligence quotients; results of the risk-stratified randomized central nervous system treatment trial MRC UKALL XI

Advances in the treatment of paediatric acute lymphoblastic leukaemia (ALL) have resulted in 5 year event-free survival rates of over 80% [1]. With such good survival, efforts are now focused on minimising treatment-related morbidity. One area of concern is the possible long-term effects of central nervous system (CNS) directed therapy on children.

Whilst CNS-directed treatments result in few long-term neurocognitive impairments in adults [2], they may adversely affect children whose neurocognitive systems are still in the process of maturing [3]. The first reports of adverse neuropsychological outcomes emerged in the 1970s and 80s after the introduction of universal CNS directed therapy - usually in the form of cranial irradiation (XRT) [4, 5]. These initial observations led to attempts to identify the causative agents, any additional risk factors and the exact nature of the impairment. There followed numerous studies examining neurocognitive outcomes after various forms of CNS-directed treatment (for recent reviews see [6, 7]) but drawing definitive conclusions from these studies is compromised by small patient numbers, differences in study design, the vast range of tests employed, use of historical cohorts, lack of proper control groups, non-random assignment of different CNS-directed treatments and changes in accompanying systemic therapy and supportive care over time [8, 9].

Debate still exists over the most important causative agents, and in particular the relative impact of different CNS-directed treatments on neuropsychological outcomes. Early studies using global measures of intellectual functioning, such as intelligence quotients (IQs) and academic attainment, showed fairly consistent declines in patients treated with XRT [5, 10‐13]. This led to increasing avoidance of radiotherapy in many treatment protocols and, as a result, recent data are sparse. The outcome with chemotherapy-only regimens is more variable with some showing almost normal cognitive functioning [14‐18], and others reporting reduced IQs [19]. A large meta-analysis [20] suggests chemotherapy alone is associated with modest declines in IQ and other neurocognitive functions. The relative impact of intrathecal methotrexate (IT MTX) versus high-dose systemic methotrexate (HDM) on CNS morbidity remains an important unanswered question especially since their equivalence in terms of overall survival means that any adverse side-effects are increasingly important.

An emerging view is that the mode of CNS-directed therapy may have little influence on adverse outcomes which may instead reflect the impact of the underlying disease and/or global manifestations of treatment: Two meta-analyses confining analysis to neuropsychological outcomes in studies which included valid control groups have shown that patients with ALL fare worse than controls regardless of their mode of CNS-directed therapy [20, 21]. The choice of control group is vital since IQ is highly correlated with socioeconomic status [22] making comparison with population means inappropriate in most small to medium scale studies. Until now a sufficiently large randomised trial including an appropriate control group has been lacking to definitively address this question.

If the mode of treatment is not the main determinant of adverse outcome then the search for additional risk factors becomes even more important. A number of small studies have identified younger age [4, 10, 15, 23, 24] and female sex [10, 25] as likely candidates. In an early meta-analysis [5], an age of 5 years or under at initial diagnosis was a significant factor but this study did not examine gender differences. Girls may fare worse, particularly in some areas such as verbal IQ [26] but existing meta-analyses are not sufficiently powered to answer this question [20]. Moreover, age and gender factors may interact to give rise to adverse outcome.

Against this background, the MRC UKALLXI psychometric study aimed to compare prospectively the neurocognitive effects of three different types of CNS-directed therapy (HDM vs. IT MTX and HDM vs. XRT). The study of a large cohort of children randomly allocated to different treatment regimens, and comparison with an appropriate control group allowed this study to address a number of important questions not yet reliably answered in the literature i.e. 1) In modern treatment protocols (with avoidance of radiotherapy in children under 2 years of age) is the use of cranial irradiation still associated with adverse neuropsychometric outcomes compared to high dose methotrexate? 2) Is high dose systemic methotrexate associated with different psychometric outcomes compared to intrathecal methotrexate? 3) Does age or gender influence susceptibility to adverse psychometric outcomes? 4) Can a subset of children at high risk of neurological adverse outcomes be identified to enable targeted intervention? 5) Is treatment for ALL associated with reduction in IQ test scores in patients compared to scores in age matched relatives?

No differences in event free survival (EFS) were seen between the two randomised treatment arms [27], thus increasing the importance of identifying any adverse effects of treatment. Here, we report the results of intelligence tests for patients and controls assessed at baseline and at 3 and 5 year time-points after initiation of treatment.

During the study period 866 children had an IQ test (555 patients; 311 controls). As shown in the accompanying CONSORT diagram (Figure 1) the numbers of eligible patients tested at the three time points were 305/876 (35%), 369/1137 (32%) and 289/728 (39%), respectively. Thus, the proportion tested did not decrease as a function of time from diagnosis. Psychologists were asked to give priority to testing the high risk group, and this is reflected in the proportion of tests done (65% high risk, 30% low risk). Although a small proportion of the eligible patients were not tested due to specific reasons such as refusal, failure to attend for testing or practical problems (language, relocation, relapse prior to test etc.), the vast majority of eligible but untested patients were untested due to the time constraints of the psychologist's workload. There were no differences between randomised arms in the proportions tested.

Further information on the tests employed, standardisation of tests, statistical power calculations and choice of controls is detailed at the end of the paper.

We present here the largest study of neuropsychological outcomes in children treated for ALL. In addition to patient numbers, this study benefits from being randomized with respect to treatment regimes, a prospective design, and the inclusion of a control group of healthy children. Despite the recognised problems of using different tests and standardizations for different age groups this study has produced clear results.

Firstly, there were no significant differences between patients randomised to continuing intrathecal methotrexate alone compared with those randomised to additional high dose methotrexate. This was true for both the under, and the over 5-year old age groups, and for both sexes. The numbers of participants in these comparisons were large allowing reasonable confidence that important differences do not exist. These findings are consistent with the majority of smaller studies [14‐18] and meta-analyses [20] in the literature.

Similarly, we found no significant differences in IQ scores in those randomised to cranial irradiation compared with those randomised to high dose methotrexate. Although possibly unexpected, our results mirror those of another recent study showing that with modern protocols the neuropsychological outcomes for XRT and chemotherapy-only groups are very similar [28]. Importantly, the UKALL XI protocols used a relatively high dose of cranial irradiation (24 Gy) further strengthening results of Waber [28] whose protocols only used 18 Gy. In addition, relatively early folinic acid rescue (commencing 36 hours after the start of the HDM infusion) may have reduced late effects of HDM. Both of these factors would have been expected to widen any gap between HDM and XRT in terms of adverse effects. These results contrast with earlier reports of significant impacts of cranial irradiation on IQ and other measures of intellectual functioning [4, 13, 14, 18, 24, 29]. Several possibilities may explain these discrepant results. Firstly, the majority of studies showing adverse effects of cranial radiotherapy included very young children, and in many the adverse effects of radiotherapy were strongly associated with the younger age groups [10, 14, 18, 24, 25, 29, 30]. Since radiotherapy is thought to cause neurotoxicity predominantly by demyelination [31] and myelination is not complete until much later in childhood, younger children would be expected to be particularly vulnerable. Our study avoided all radiotherapy in children under 2 years of age and in addition the XRT randomisation was confined to children with a WCC > 50 - a biological feature associated with older age. Secondly, the use of an adequate control group is vital since studies that showed a detrimental effect of radiotherapy may have been demonstrating a detrimental effect of ALL and its treatment rather than a specific effect of XRT alone [5]. This is supported by carefully controlled longitudinal studies from the St Jude group which showed no difference between XRT and chemotherapy groups at a single time point [32], but subsequent longitudinal follow-up showed a decline in both treatment groups over time [26, 33]. Thirdly, most reports of XRT effects pre-date the current treatment era and therefore changes in accompanying systemic therapy, supportive care or improved delivery methods may have either reduced the morbidity from cranial radiotherapy or narrowed the gap by increased neurotoxicity with intensified systemic therapy. This is supported by data from animal models [34] and patients [35] suggesting that systemic chemotherapy can have synergistic or protective effects when combined with XRT. Finally, the majority of previous reports involved non-randomised, retrospective studies of small numbers of patients and may have therefore been inadvertently biased towards recruitment of children with poorer outcomes and/or publication bias.

Although overall our data do not suggest that age is a significant risk factor for mean IQ values, children aged less than 5 years at initial diagnosis are more likely to have IQ below 80 at 3 years compared to children aged over 5 years at diagnosis, irrespective of treatment allocation. This is consistent with models of brain development suggesting that younger children are likely to be particularly vulnerable to neurotoxic insults. These data also highlight that mean IQ values may mask significant individual declines in IQ as discussed below.

The finding of similar outcomes in males and females is reassuring. Initial reports of inferior outcomes in girls came from relatively small studies using combinations of methotrexate and cranial radiotherapy [10, 25]. More recently a number of chemotherapy-only protocols have also shown inferior outcomes in girls [17], although a meta-analysis of chemotherapy-only protocols could not reach a firm conclusion [20]. The possible underlying mechanisms for gender differences in neuropsychological outcome in ALL are poorly understood and in fact in other areas of acute brain injury, such as head injury, girls usually have better outcomes than boys. Again, changes in therapy protocols such as lack of co-administration of high-dose methotrexate and avoidance of radiotherapy in young patients may explain the lack of difference in IQ in our studies.

Importantly, despite the lack of effect of randomised treatment allocation on IQ, patients definitely fared worse than controls, with a lower mean IQ of between 5 and 7 points. The effect was seen for FSIQ, VIQ and PIQ. A reduction in IQ score of this magnitude may be of only modest impact in children with average or above average initial IQ scores but importantly this effect also translates into a larger proportion of children with IQ scores less than 80 - a level consistent with low intellectual functioning. These results suggest that children treated for ALL are at risk of neurodevelopmental morbidity regardless of which of these randomised CNS-directed therapies they received. This has previously been suggested by smaller studies [33, 36], a meta-analysis [21], and a recent larger study [28] which reported some selective weaknesses in verbal IQ and mathematics fluency in all children with ALL regardless of their treatment allocation. It is also supported by a lack of dose response for both radiotherapy [26] (18 Gy vs. 24 Gy) and methotrexate [18] (HDM vs. very high dose MTX). It is known that even intrathecal methotrexate alone can be associated with white matter changes, calcifications, leukoencephalopathy, cortical atrophy, and seizures in some patients [37].

Some important limitations of our study should be acknowledged. A cross-sectional design was necessary to maximise patient recruitment in order to answer the main study questions but this design makes it impossible to track the unfolding of impairments in individual patients over time. The large numbers of participants, balanced randomisation and inclusion of a socioeconomically matched control group makes substantial demographic differences in the tested cohorts in the different arms at 3 time-points unlikely, but some alteration in the demography of the groups over time cannot be excluded, and it is possible that this may explain improvement in scores at 5 years. Secondly, IQ tests are a relatively global measure of intelligence. A multitude of more specific defects have been reported in the literature with a particular propensity for domains such as attention, arithmetic fluency, non-verbal reasoning, to be affected [13, 15, 38]. We chose to investigate IQ as a primary outcome measure because it was so well standardised and relatively robust but these results do not exclude the possibility of specific influences of our randomised treatment arms on more subtle but important neuropsychological measures. Whilst investigation of these additional measures would obviously add to our findings it does not detract from our major observation of a difference in mean IQ between patients and controls.

Overall these data suggest that factors other than the mode of CNS directed treatment determine the likelihood of CNS morbidity and that there may be vulnerable groups of children who manifest large declines in IQ whilst others are relatively unaffected. That mean IQ scores comfortably fall in the average range will be a huge reassurance to most parents and patients - attention now needs to be focussed on identifying the smaller subset of vulnerable children. Study of these children (alongside matched unaffected controls) should allow identification of possible risk-factors. Candidates include; inherent genetic susceptibility, drug toxicity, time out of full-time education or particular vulnerability of certain individuals to the impact of chronic illness. Pharmacogenomic and genome wide association studies comparing severely affected children with those with persistently normal IQs should help identify genetic and drug-related risk factors. Indeed a recent report implicates polymorphisms in folate metabolism pathways as a risk factor for CNS morbidity [39]. Correlative neuro-imaging may also help identify aetiology, as it is possible to quantify leukoencephalopathy using MRI [40] and functional MRI offers an exciting new approach [41]. Systemic drugs used in all children with ALL include anti-folates, steroids and nucleoside analogues all of which have documented neurotoxic side effects [30, 42, 43]. The equivalent results in pre-school and older children argue against frequent and/or prolonged absence from school being the primary cause for the observed reduction in IQ.

In summary, with modern protocols and avoidance of XRT for very young children, the neuropsychological outcomes for XRT and chemotherapy-only groups are very similar. We are unable to confirm female gender as a risk factor, but children aged below 5 years may be more vulnerable to treatment related neurotoxic effects. The most striking finding of this study is the difference observed between patients and controls, regardless of the CNS treatment delivered. This supports the view that ALL itself, and the necessity for intensive treatment, has a detrimental effect on IQ in some children. Detailed longitudinal neuropsychological assessments should allow individualised risk factors for neurocognitive morbidity to be examined. We predict that improvements in neuropsychological outcomes for children with ALL will depend more on individualised therapy for children at high risk of CNS morbidity than on avoidance of specific CNS-directed therapy regimens in unselected patient cohorts.