Use of Morphological Early Response Data to Predict Relapse in Teenage and Young Adult (TYA) Patients with Acute Lymphoblastic Leukaemia (ALL)

Introduction

Treatment outcomes for teenage and young adult (TYA) patients with Acute Lymphoblastic Leukaemia (ALL) lag behind those of younger patients when treated on equivalent protocols, with higher rates of toxicity and relapse reported for those aged >10 years. Outcomes of relapsed ALL are poor so more accurate identification of those destined to relapse is required. Whilst minimal residual disease (MRD) is a powerful tool to predict those at low risk and in whom treatment can be de-escalated, improved strategies are required for early identification of high risk patients.  This study aimed to determine if the morphological early response assessment could be used as a predictor of relapse in TYA patients with ALL.

Methods

All patients treated on the national prospective UKALL2003 trial aged 10 – 24 years who had not relapsed by day 35 (i.e. could potentially have treatment escalated) were included in the study.  Early response in patients aged >10 years was defined as percentage of blasts by morphology at day 8: rapid early responders had ≤ 25% blasts and slow early responders >25% blasts. We analysed the following patient characteristics which were previously associated with differential outcomes: age at diagnosis, presenting white cell count, rapid or slow early response (RER or SER); MRD status at day 28; immunophenotype (B/T/other) and cytogenetic risk group (high risk, good risk, poor risk, T-ALL, other). Cox regression was used to assess the association of these factors with time to relapse and logistic regression (LR) was used to assess their association with relapse by the end of treatment (EOT:  2.5 years for girls and 3.5 years for boys). Using backwards selection (cut off for inclusion: p=0.1) a multivariable LR model was established and a receiver operating characteristic curve (ROC) was used to assess how well it could predict relapse before the EOT.    

Results

820 consecutive patients were included in the analysis of whom 597 were aged 10 - 15 years and 223 were aged 16 – 24 years. 332 patients were classified as MRD high risk and 184 as SER on the basis of morphology. RER was significantly associated with a reduced risk of relapse even after adjustment for other factors including MRD, age, WBC and risk status (Hazard Ratio (HR) 0.64 (95% CI 0.42 – 0.97, p<0.001).

697 patients with an outcome available at the EOT could be included in the LR analysis. The final multivariable model included day 28 MRD, day 8 response, age, WBC and cytogenetic risk group. The predictor generated by this model gives a ROC area under the curve of 81.2% (95% CI: 76.8 – 85.6) but there were no cut points which resulted in acceptable detection (DR) and false positive rates (FPR).  As the intervention for the poor prognosis group would potentially include an allogeneic transplant with associated high mortality and morbidity a low FPR would be key to clinical utility.  In order to achieve a clinically useful DR of 79.5% we would have to accept a false positive rate of 30.5%. As only 11% pf patients relapsed by EOT but 36% of patients fell into our bad prognosis group this would mean that for every 100 patients, ~27 would be transplanted unnecessarily.   In the subgroup analysis of those aged 16 – 24 early response was not significantly associated with either time to relapse (HR RER: 0.74 (0.36 – 1.51, p = 0.4) or relapse by the EOT (Odds ratio RER: 0.63 (0.26 – 1.49, p =0.3)

Conclusions

Slow early response is significantly associated with both time to relapse and relapse by EOT in this patient cohort aged 10 – 24 years and this effect is independent of other prognostic factors including MRD. However, even when over-fit the prognostic model generated (which included SER) resulted in FPRs which were too high to be clinically useful.  Future work should focus on novel alternative methods of relapse risk prediction for the adolescent patient with ALL.

Figure 1