Prediction of non-adherence to oral mercaptopurine in pediatric ALL

Acute lymphoblastic leukemia (ALL) is the most commonly diagnosed childhood cancer with 95% of children successfully achieving remission within four weeks of induction treatment. However, relapse occurs in approximately 20% of cases within 5 years. Relapse carries a poor prognosis, and the second line therapies available are toxic. Prolonged maintenance treatment with self-administered mercaptopurine is necessary to achieve durable remission, though low or variable mercaptopurine exposure increases risk of relapse, with adherence rates of <90% associated with a 3.9-fold increased risk of relapse.¹

Patient-level factors, such as age, ethnic background, single-parent, number of children in household, household income, parental education, and ingestion of mercaptopurine at varying times of day have been shown to be associated with an increased risk of non-adherence.¹  

The COG AALL0331 study enrolled 5,377 children aged 1–9 years with standard-risk B-ALL, of which 1,857 were eligible to continue. The study aimed to assess whether pegaspargase intensification in combination with low-dose chemotherapy would improve complete remission rates.² In a secondary analysis of data from these patients, Hoppmann, et al. aimed to identify patients being treated with mercaptopurine who were non-adherent using a risk prediction model based on patient-level factors. They recently published their study in Cancer.¹

Study design and patient characteristics

The COG AALL0331 study design has been reported previously,² and our summary can be found here. For this secondary analysis, patients included

• were aged ≤21 at diagnosis,
• were in their first remission,
• were receiving mercaptopurine as maintenance therapy,
• had received ≥6 months maintenance therapy; and,
• had a minimum of ≥6 months remaining on maintenance therapy.

During the first 6 months of therapy, adherence was monitored using the Medication Event Monitoring System, which records the date and time of each bottle opening. Patients/parents were also asked to complete a questionnaire around medication consumption habits on Days 29, 57, 113, and 141. Patient demographics, dosage information, and the genotype of thiopurine methyltransferase were collected at study entry. Red cell thioguanine nucleotide levels were collected at 6 consecutive monthly time points. The risk prediction model was developed at Month 3 as there was the least amount of missing data at this time point. The characteristics of patients included in this analysis are detailed in Table 1. A total of 407 patients were included, with a mean age of 7.7 years, patients were mostly male (68%), 33% did not take their mercaptopurine at the same time of day, and 28% were non-adherers (mean adherence rates of <90%). The characteristics of patients included in the prediction model training set and test set were comparable (Table 1).

Table 1. Characteristics of patients at 3 months who were included in risk prediction model development*

Results

Patient characteristics by adherence status are detailed in Table 2. Non-adherers were more likely to be Hispanic (44% vs 30%) or African American (25% vs 12%; p < 0.0001), were more likely to have a single parent household with multiple children (20% vs 6%; p < 0.001) and were less likely to take their mercaptopurine at the same time every day (54% vs 71%; p = 0.0007).

Table 2. Characteristics of study population by adherence status*

Included in the risk prediction model were: age at study year, race/ethnicity, absolute neutrophil count, mercaptopurine dose intensity, household structure, and mercaptopurine ingestion pattern (whether it was taken at the same time of day). The odds ratio of each of these variables are detailed in Table 3.

Table 3. Association of variables with non-adherence to oral mercaptopurine*

The training set model yielded an area under the curve (AUC) of 0.79 (95% confidence interval [CI], 0.71–0.85) and when assessed in the test set, the AUC was 0.74 (95% CI, 0.63–0.85). The utility of the model was also tested at other time points, and yielded AUCs of 0.63 at 1 month, 0.72 at 2 months, 0.71 at 4 months, and 0.69 at 5 months. The model performed better in the older patient cohort (≥12 years), with an AUC of 0.79 (95% CI, 0.59–0.99), than in the younger patient cohort (<12 years) which had an AUC of 0.70 (95% CI, 0.58–0.81).

Assessing different predictive probabilities, a cut point of 0.3 was chosen by the group as the point at which patients, high- or low-risk of adherence, could be separated with 71% sensitivity and 76% specificity. Using the binary risk classifier, Hoppmann, et al. found that the 5-year cumulative incidence of relapse was 11.9% for patients at high risk of non-adherence vs 4.5% for those at low risk of non-adherence (p = 0.006). When taking into account the National Cancer Institute risk status, patients at high risk of non-adherence were found to have a 2.2-fold increased risk of relapse (95% CI, 0.94–5.07; p = 0.07).

Conclusion

The group felt that their analysis was important given the challenges that clinicians face when identifying children with ALL at risk of mercaptopurine non-adherence. Limitations of the study include use of the Medication Event Monitoring System to measure adherence as it only measures bottle opening, not medication ingestion, lack of information on minimal residual disease and impact on relapse, and the varying time points from start of maintenance prior to enrollment. Despite these limitations, the study was strengthened by the diverse patient population included. Hoppmann, et al. felt that their study could enable the targeting of interventions such as education and personalized text message reminders to patients at higher risk of non-adherence. As such, the group are now building an online tool to enable clinicians to tailor recommendations accordingly.