Long-term neurotoxicity among childhood acute lymphoblastic leukaemia survivors enrolled between 1971 and 1998 in EORTC Children Leukemia Group studies

The study participants included childhood ALL patients (less than 18 years at diagnosis) who were treated in one of the EORTC CLG clinical trials 58741 (accrual period: 1971–1978), 58831/2 (1983–1989), and 58881 (1989–1998) and were alive two months after the end of the first line therapy (Figure S1). Exclusion criteria in these initial studies were B-cell non-Hodgkin lymphomas, B-cell ALL, (except in 58741 study), previously treated patients (corticoids and/or vincristine for less than 8 days were allowed in 58831/2 studies, corticoids for less than 8 days were allowed in 58881 study), severe encephalopathy, severe heart disease or trisomy 21 (in 58831/2 studies), initial CNS involvement (in 58832 study) and criteria related to risk-group definitions (in 58831/2 studies). Patients who were not eligible for participation in these trials were excluded from the present neurotoxicity study, except in case the initial exclusion criteria was a central nervous system (CNS) involvement at baseline. Patients with a history of trisomy 21 were excluded as well, irrespective of whether it was an exclusion criterion.

Between 2011 and 2018, 24 hospitals from France and Belgium participated in the 58LAE study and provided updated follow-up information about eligible patients by filling in an electronic case report form (eCRF), based on medical records of the patients. New clinical evaluation was recommended but not mandatory. The eCRF comprised information about clinical outcomes, treatments, and long-term toxicities, including neurotoxicity (cognitive disturbance, seizures, strokes and leukoencephalopathy).

In this section, we provide details on definitions (i.e., CNS involvement), treatments (i.e., CRT, hematopoietic stem cell transplantation [HSCT] and high-dose cytarabine [HD Cytarabine]) and risk group (very high-risk [VHR] in study 58881) considered for the study of association with neurotoxicity (Table 1). Other details regarding studies 58741, 58831/2 and 58881 are reported in Supplementary Table S1) [15].

CNS involvement was defined as the presence of ≥ 1 blast on cytospin with any white blood cells count (WBC) in the cerebrospinal fluid. CRT was administered to all patients in study 58741. Then, studies 58831/2 investigated whether CRT could be omitted with systemic and intrathecal CNS prophylaxis in CNS-negative patients. Based on these results, CRT was fully omitted in the 58881 study. In 58881 study, patients with VHR criteria (≥ 1000 blasts/mm³ of blood at the end of prephase, and/or who did not achieve complete remission after induction and/or with t(4;11)/MLL-AF4) received an intensified treatment according to the BFM relapse protocol. HSCT was reserved to VHR patients enrolled in study 58881 in first remission if a donor was available. Finally, the value of HD Cytarabine (1 g/m²/day intravenously on days 8–9, 22–23, 36–37 and 50–51 of interval therapy) was evaluated in a randomized way in increased risk patients in complete remission in 58881 study.

At the time of the enrolment in studies 58741, 58831/2 and 58881, informed consent was sought from all participants or, if participants were under 18, from a parent and/or legal guardian, according to local practice of each participating centre and in accordance with the Declaration of Helsinki. The EORTC study 58LAE was approved by the Ethical Committees of the participating institutions and informed consent was obtained from all alive patients at the time of the latest follow-up, in accordance with the applicable national legislation.

Late adverse effects were defined as toxicities occurring during first line therapy and still present two months after the end of first line therapy, or appearing two months or later after the end of first line therapy. Definitions of the adverse effects (and grades for cognitive disturbance) were based on the Common Terminology Criteria for Adverse Events (CTCAE), Version 4.0 [16].

Time to events of interest (leukoencephalopathy, seizure and stroke) was defined as the time between two months after the end of the first line therapy and a diagnosis of the corresponding event.

For all estimated parameters, point estimates and 2-sided 95% confidence intervals (CI) are presented. All tests were performed at a two-sided significance level of 0.05.

The cumulative incidence of leukoencephalopathy, seizures and stroke was estimated using the Aalen-Johansen estimator [17]. The proportional sub-distribution hazards model was used to investigate the association between protocol, age at diagnosis, and CNS involvement at diagnosis, and the cumulative incidence of the neurotoxicities [18]. Patients with an event of interest before two months after the end of the first line therapy were counted as events at day 1. Death without the event of interest was considered as a competing event. The follow-up of patients was censored at the time of last visit. In case of a missing event time in the analysis of leukoencephalopathy, seizures and stroke, the cumulative incidence functions were estimated using multiple imputation. In the analyses of cumulative incidences, time to event was imputed using the estimated cumulative incidence function for the group of interest based on complete cases. The estimates based on the imputed data were then combined using the Rubin’s rules [19]. Multiple imputation with the asymptotic normal data augmentation scheme was used to obtain estimates from the Fine and Gray model [20]. The number of imputations was 100 in the estimation of the cumulative incidence functions and 10 in the estimation of the subdistribution hazard ratio. Logistic regression was used to investigate the association between age at diagnosis, CNS involvement at diagnosis, HSCT, HD Cytarabine dose (among patients from the 58881 study only), and VHR (among patients from the 58881 study only) and cognitive disturbance. Both in the logistic regression and the Fine and Gray model analyses, for each covariate of interest a separate model was fitted. The analyses of age, CNS involvement at diagnosis, and HSCT were adjusted by protocol. Patients with missing covariates that were required for a specific analysis were excluded from such analysis.

A Fine and Gray model stratified by protocol with a time-varying covariate for HSCT taking a value 1 after HSCT and a value 0 otherwise was used to investigate the association between HSCT and seizures. Since the relapse status and HSCT were correlated, a sensitivity analysis was performed in which relapse modelled as a time-varying covariate taking a value 1 after relapse and a value 0 otherwise was added to this model.

Among 3228 patients included in the EORTC 58741, 58831/2, and 58881 trials, 2329 patients were eligible for participation in the current study (Figure S1). Among those eligible, data for at least one neurotoxicity outcome was available for 890 patients, who were included in the analysis. Patients with and without available data were similar in terms of the National Cancer Institute (NCI) risk group and exposure to CRT and HSCT (Table S2). However, death and relapse were reported more often for patients without than those with available data.

Among the 890 patients included in this study, the median follow-up was 19 years and the interquartile range of the follow-up was 15–22 years. Relapse occurred in 20% of the included patients, CNS relapse in 10%, 16% received CRT and 11% underwent HSCT (Table 1).

Data on cognitive disturbance was available for 584 patients. Approximately 66% of patients from the 58741 trial and approximately 15% from the more recent trials had cognitive disturbance grade 1 or higher (Table 2). Adjusting for protocol, the associations of age at diagnosis, CNS involvement at diagnosis, and HSCT with the risk of cognitive disturbance were not statistically significant (Table 3). There was also no evidence of an association between VHR after consolidation in the 58881 study or being randomized to the HD Cytarabine arm in the 58881 study and the risk of cognitive disturbance.

The cumulative incidence of seizures at 20 years from end of treatment was 45% (95% CI 33–62%) in the 58741 study, 13% (95% CI 9–19%) in the 58831/2 studies and 8% (95% CI 6–10%) in the 58881 study (Fig. 1A). In a model adjusting for protocol, HSCT was associated with a higher incidence of seizures (HR = 3.09, 95% CI 1.50–6.35). In a model including protocol, relapse, and HSCT, relapse was strongly associated with a higher incidence of seizures (HR = 4.12, 95% CI 2.27–7.48) but the association for HSCT became weaker and not statistically significant (HR = 1.47, 95% CI 0.64–3.37). Table 4 presents the incidence of seizures by age at diagnosis, CNS involvement at diagnosis, VHR after consolidation, and HD Cytarabine randomization.

Additionally, we observed that seizures still occur beyond 5 years of follow-up. The cumulative incidence continues to increase in particular in patients from the 58741 study but also in patients from the 58831/2 study, with some events observed even beyond 20 years (Fig. 1A).

The cumulative incidence of stroke at 20 years from end of treatment was 16% (95% CI 8–35%) in the 58741 study, 2% (95% CI 1–5%) in the 58831/2 studies and 2% (95% CI 1–4%) in the 58881 study (Fig. 1B). Table 5 presents the incidence of stroke by age at diagnosis, CNS involvement at diagnosis, VHR after consolidation, and HD Cytarabine randomization. Patients who were 10–17 years of age at diagnosis had a higher incidence of stroke as compared to those < 6 years of age (HR = 3.35, 95% CI 1.31–8.55). The difference between those 6–9 and those < 6 years of age was small and not statistically significant. Stroke continued to occur beyond 5 years of follow-up, in particular in patients from the 58741 study (Fig. 1B).

The cumulative incidence of leukoencephalopathy at 20 years from end of treatment was 62% (95% CI 43–91%) in the 58741 study, 5% (95% CI 2–10%) in the 58831/2 studies and 3% (95% CI 2–5%) in the 58881 study (Fig. 1C). Table 6 presents the incidence of leukoencephalopathy by age at diagnosis, CNS involvement at diagnosis, VHR after consolidation, and HD Cytarabine randomization. Patients who were 10–17 years of age at diagnosis had a higher incidence of leukoencephalopathy as compared to those < 6 years of age (HR = 3.09, 95% CI 1.54–6.22, Table 6). The difference between those 6–9 and those < 6 years of age was small and not statistically significant. Additionally, leukoencephalopathy continued to occur beyond 5 years of follow-up particularly in patients from the 58741 study but also in those from the other trials, with some events observed beyond 20 years from end of treatment (Fig. 1).

Neurocognitive disturbance rates reported here are in concordance with previous work published by St. Jude’s team [3]. Impairment rates across neurocognitive domains at mean 26 years since diagnosis ranged from 28.6 to 58.9% in a large cohort of SCALL [3]. Neurocognitive disturbance may occur in several domains such as executive function, attention, memory and processing speed [3, 4].

We found that neurotoxic complications are far more prevalent in the 58741 cohort, and this could be explained by the CRT applied to treat every patient.

In others previous studies, the predominant role of CRT in long-term neurotoxicity was already demonstrated [4, 5]. Perhaps, CRT may be a key risk factor of neurotoxicity, but the high cumulative dose of steroid therapy (Table S2) and higher single dose of HD MTX may have also played a role. Unfortunately, the design of this study and the number of events do not allow to confirm the impact of each therapy as risk factor.

We also previously reported the long-term outcomes of the randomization comparing no CRT versus CRT in medium/high-risk patients included in study 58832, at a median follow-up of 20 years (range 4–32 years) [22]. Omission of CRT was associated with significant decrease in the rate of CNS toxicities: leukoencephalopathy and cognitive disturbance were halved, seizures were reduced fourfold, and no patients randomized to the no CRT arm had strokes compared with almost 8% of the patients in the CRT arm.

Neurotoxicity is widely attributed to CRT [3] but the intensified intravenous administration of chemotherapeutics drugs and ITTs have gradually replaced CRT for the treatment of ALL in children. It was shown that even patients treated with chemotherapy only can also present neurocognitive disturbance mainly in the executive and attention domains [3, 4].

The reported prevalence of seizure/stroke in irradiated patients (25 Gy) from the cohort 58741 is particularly high (45%/16%, respectively) but this study is, to our knowledge, the first to report incidence of seizure/stroke in a SCALL cohort who have all been irradiated. Rate of seizure/stroke in 58831/2 and 58881 cohorts is concordant with previous publications [9, 10, 25]. The 10-year cumulative incidence of subsequent seizure and/or stroke in SCALL was estimate at 4.3% [9]. Neuropsychological performance in patients who experience seizure indicated problems in attention; working memory and processing speed and treatment related seizure were associated with leukoencephalopathy [25]. Seizures can be caused by MTX (such as transient stroke like syndrome) but also due to asparaginase treatment [10]. The long term neurological consequence of this acute toxicity remains to be clarified and is difficult to evaluate separately from other neurotoxic drugs.

We reported here a high rate of leukoencephalopathy in the cohort 58741 and to our knowledge, there is no other study including only irradiated SCALL patients. On the contrary, quite low rates of leukoencephalopathy are reported here, in cohort 58831/2 and 58881 in comparison with previous work [12]. In this study, the authors reported leukoencephalopathy in 23.6% of survivors of SCALL (87 out of 369) among which 14 were symptomatic. Leukoencephalopathy persisted at end of therapy in 74% and 58% of asymptomatic and symptomatic patients respectively. Leukoencephalopathy was prospectively assessed by MRI at four timepoints through treatment.

The difference may be explained by the fact that here the event of leukoencephalopathy was retrieved from medical records and has not been prospectively assessed by magnetic resonance imaging. We may have missed asymptomatic leukoencephalopathy, including in the non-irradiated patient.

Patients above 10 years old at the time of treatment seemed to be more at risk of leukoencephalopathy. This is also consistent with previous work [12]. Methotrexate can also cause long-term leukoencephalopathy. Leukoencephalopathy secondary to methotrexate persisted in 74% of asymptomatic and 58% of symptomatic patients at end of therapy [12] and was associated with decreasing neurocognitive performance [24]. Higher cumulative number of ITTs or high dose IV cytarabine are other risk factors for leukoencephalopathy. [12, 13]

In our previous work, we also showed that decrease in MTX dosage was associated with lower rates of leukoencephalopathy and cognitive disturbance, highlighting the role of chemotherapy in CNS toxicities [24].

Finally, if de-escalation in ALL treatment in particular the abandonment of CRT therapy has led to less neurological complications, survivors who received chemotherapy still suffered from apparent cognitive impairment in particular in their attention and executive function domains [4]. In the future, late neurological consequences of new therapy in acute lymphoblastic leukemia such as CAR-T cell therapy or blinatumomab should be carefully monitored. Additionally, the use of dexamethasone in current trial (e.g. https://​clinicaltrials.​gov/​study/​NCT04307576) may provide evidence about its additional neurotoxicity.

Due to insufficient data related to second neoplasms (SN) in the 890 patients included, we could not include an analysis of secondary CNS tumours. We previously reported the incidence of SN in 58832 and 58881 EORTC CLG studies and found low incidence of SN and secondary CNS tumours, in particular in patients treated without CRT [22, 25]

Other limitations include the retrospective design of the study, based on medical records and without systematically reassessment of the patients for the purpose of this study. Some late neurological events may therefore have been underreported. Among 2329 eligible patients, only 890 patients were included in the analysis and information about some endpoints was missing among patients included in the analysis. Although patients with and without available data were similar in terms of the risk group and exposure to most toxic treatments, such a large extent of missing data involves a significant risk of a selection bias. Finally, due to the study design, we were not able to investigate the effects of particular treatments.

This retrospective observational study highlights the frequency of neurological late effects in SCALL. With the increase of the overall survival of ALL patients in the past decades, late effect surveillance and their management became an important public health challenge. With this regard, our observation that neurotoxicity continues to occur beyond 5 years of follow-up is of particular importance for clinicians involved in late effects clinics. This is a further argument in favour of comprehensive, multidisciplinary long-term follow-up, not limited to the period when initial cancer is treated and most relapses occur.

The role and potential benefit of longitudinal neurological screening such as MRI and/or neuropsychological tests should be evaluated in SCALL, as neurological late effects continue to occur beyond 5 years after end of treatment.