Abstract Background Infant MB (iMB) treatment strategies may include adjuvant focal RT (fRT) for local control, to minimize neurocognitive side-effect of craniospinal irradiation (CSI). From a large international cohort of 380 relapsed iMB, we previously described fRT was associated with worse post relapse survival (PRS) compared to chemotherapy alone, in univariate analysis. Methods We performed a sub-analysis of patients initially treated with adjuvant fRT, to describe their pattern of relapse and salvage practices. Results The cohort included 51 patients who relapsed after initial fRT (median dose: 54 Gy) along with conventional in 47 patients or high-dose chemotherapy (HDC) in 4 patients. Median age at diagnosis was 34.2 months (range 11-55). At initial diagnosis, 42(85.7%) achieved gross total resection and 46(90.2%) were localized. Maintenance therapy or intrathecal/intraventricular chemotherapy were added in 14 patients. Molecular subgrouping was available in 31 patients (8 SHH, 17 group 3, 6 group 4). Median time of relapse was 13 months from diagnosis (5-51.9). Relapse was disseminated or combined (local and distant) in 96% of the cases. Twelve patients (24%) underwent palliative management. Thirty-nine (76%) patients received curative intent salvage therapy, of which 31 (79%) received CSI (median dose of 36 Gy, range 18-39.6). CSI was administered alone in 22% or with conventional chemotherapy or HDC in 65% and 13% . The 5 years PRS was 20.8% (±9%). Female sex, age <24 months at diagnosis, late relapse ≥12 months from diagnosis, asymptomatic relapse, salvage CSI, CR post-salvage therapy were associated with better PRS. Patients salvaged with CSI had a 5-year PRS of 26.5% vs 0% without. Conclusion Following adjuvant fRT, almost all patients who relapsed presented with dissemination, despite initial localized disease, and dismal PRS. Focal RT in combination with upfront therapy is ineffective to prevent leptomeningeal relapses and the success of salvage CSI is limited.

ABSTRACT:A disadvantaged neighborhood, as represented by area-level socioeconomic status (SES), has been associated with adverse outcomes among children with acute lymphoblastic leukemia (ALL) in the United States, but the duration of impact after ALL diagnosis is not well understood. This retrospective cohort study used the National Cancer Database to evaluate the impact of area-level SES on overall survival among children with ALL. Median income and education quartiles based on residential zip code were used to create a composite area-level SES variable. Individual-level variables included age, sex, race, year of diagnosis, primary payer, distance to care, rurality, time to treatment, and comorbidity index. Cox proportional hazard models were created overall and conditional on surviving 2 and 5 years from diagnosis. This cohort of 17 044 children with ALL consisted of 57% males, 61% non-Hispanic White, 24% Hispanic, 8% non-Hispanic Black, 5% Asian, and 2% other race/ethnicities. Furthermore, 12% (1971) resided in the lowest SES neighborhoods, whereas 18% resided in the highest SES neighborhoods. Compared with those in the highest SES neighborhoods, children in the lowest SES neighborhoods had a 55% increased hazard of all-cause mortality (95% confidence interval [CI], 1.29-1.88; P < .001). The increased hazard of mortality among those from the lowest SES neighborhoods persisted among those who survived 2 years (hazard ratio [HR], 1.70; 95% CI, 1.28-2.26; P = .001) and 5 years (HR, 1.87; 95% CI, 1.17-3.00; P = .032) after ALL diagnosis. These findings suggest the need for long-term strategies to support at-risk patients beyond completion of frontline therapy.

Childhood cancer survivors carry a high burden of late-occurring treatment-related morbidity. Long-term risk-based anticipatory surveillance allows for early detection and management of complications. We sought to examine demographic, clinical, and social characteristics associated with survivorship clinic attendance at the Taking on Life after Cancer (TLC) Clinic at the Children’s Hospital of Alabama. The cohort included 1122 TLC-eligible patients diagnosed with cancer between 2000 and 2016. The outcome of interest was ≥1 TLC visit. Univariable logistic regression modeling assessed cancer type, treatment era, age, sex, race/ethnicity, payer type, rural/urban residency, and distance from clinic. Significant variables (P<0.1) were retained in multivariable modeling. The median age at diagnosis was 7 years old (0–19); 47

Background Obesity at diagnosis of childhood acute lymphoblastic leukemia (ALL) is associated with greater risk of relapse; whether this association extends to obesity during maintenance is unstudied. Methods This study used data from AALL03N1 to calculate median body mass index (BMI) for 676 children over 6 consecutive months during maintenance therapy; BMI percentile (BMI%ile) were operationalized as normal/underweight (<85%ile), overweight/obese (85%-98%ile), and extreme obesity (>= 99%ile). Hazard of relapse was estimated using multivariable proportional subdistributional hazards regression after adjusting for all relevant demographic and clinical predictors. Results Median age at study enrollment was 6 years and median length of follow-up was 7.9 years. Overall, 43.3% of the cohort was underweight/normal weight, 44.8% was overweight/obese, and 11.8% had extreme obesity. Cumulative incidence of relapse at 4 years from study enrollment was higher among those with extreme obesity (13.6% +/- 4.5%) compared to those with underweight/normal weight (9.0% +/- 2.1%). Multivariable analysis revealed that children with extreme obesity had a 2.4-fold (95% confidence interval [CI], 1.1-5.0; p = .01) greater hazard of relapse compared to those who were underweight/normal weight. Overweight/obese patients were at comparable risk to those who were underweight/normal weight (hazard ratio, 0.8; 95% CI, 0.4-1.6). Erythrocyte thioguanine nucleotide (TGN) levels were significantly lower among children with extreme obesity compared to those with underweight/normal weight (141.6 vs. 168.8 pmol/8 x 10(8) erythrocytes; p = .0002), however, the difference in TGN levels did not explain the greater hazard of relapse among those with extreme obesity. Conclusions Extreme obesity during maintenance therapy is associated with greater hazard of relapse in children with ALL. Underlying mechanisms of this association needs further investigation. Lay summary Findings from this study demonstrate that extreme obesity during maintenance therapy is associated with a greater hazard of relapse among children with acute lymphoblastic leukemia. We show that children with obesity have lower levels of erythrocyte thioguanine nucleotides even after adjusting for adherence to oral chemotherapy. However, these lower levels do not explain the greater hazard of relapse, paving the way for future studies to explore this association.

The association between individual-level poverty and relapse in children receiving maintenance treatment for acute lymphoblastic leukemia (ALL) remains unclear. In a secondary analysis of COG-AALL03N1, we used data from US Census Bureau to categorize patients living below year-specific federal poverty thresholds, calculated using self-reported annual household income and size of household. Participants living 120% below federal poverty thresholds were categorized as living in extreme poverty. Hazard of relapse was estimated using multivariable proportional subdistributional hazards regression for patients living in extreme poverty while receiving ALL maintenance therapy after adjusting for relevant predictors. Among 592 patients in this analysis, 12.3% of the patients were living in extreme poverty. After a median follow-up of 7.9y, the cumulative incidence of relapse at 3y from study enrollment among those living in extreme poverty was significantly higher (14.3%, 95% confidence interval [CI]= 7.3-23.6) compared to those not living in extreme poverty (7.6%, 95%CI=5.5-10.1, P=0.04). Multivariable analysis demonstrated that children living in extreme poverty had a 1.95-fold greater hazard of relapse (95%CI=1.03-3.72, P=0.04) compared to those not living in extreme poverty; this association was mitigated after inclusion of race/ethnicity in the model (hazard ratio=1.68, 95%CI=0.86-3.28, P=0.1), likely due to collinearity between race/ethnicity and poverty. A greater proportion of children living in extreme poverty were non-adherent to mercaptopurine (57.1% vs 40.9%, P=0.04); however, poor adherence did not completely explain the association between poverty and relapse risk. Future studies need to understand the mechanisms underlying the association between extreme poverty and relapse risk. Clinical Trial number: NCT00268528.

Introduction: There is conflicting evidence regarding the association between poverty and survival in children with cancer. Most studies have relied on surrogate indicators of poverty, such as zip code-based median income for area of residence or on insurance status (Bona, PBC 2016; Kehm, Cancer 2019; Bona, JNCI 2021). We hypothesized that household poverty (measured at the individual level, using self-reported annual household income) during maintenance would be associated with a greater hazard of relapse among children with ALL. Methods: In this secondary analysis of COG-AALL03N1 (primary aim: adherence to oral mercaptopurine during maintenance), we examined the association between household poverty during maintenance and relapse risk in children with ALL. Eligibility for enrollment on AALL03N1 included (1) age <21y at ALL diagnosis; (2) receiving maintenance therapy in first remission. The current analysis was restricted to patients living in US who provided information on annual household income and the number of household members. We used national yearly median poverty thresholds provided by CDC; these take into account the annual household income and the size of family unit (i.e., number of adults and children <18y). For example, the annual income of a three-person household (two parents and one child [patient]) would need to be >$14,480 to be above the poverty threshold in the year 2002 (or >$21,831 in the year 2021). Among those below poverty threshold, we classified families with annual income under 120% federal threshold (median percent below poverty threshold for this cohort) as living in extreme poverty. Hazard of relapse from study enrollment during ALL maintenance therapy was estimated using proportional subdistribution hazard regression models, treating extreme poverty as the primary exposure and adjusting for age at study, sex, NCI risk group, blast cytogenetics, average 6MP and methotrexate dose intensity and time from start of maintenance to study entry. Results: We summarize sociodemographic and disease characteristics of 592 eligible patients in the Table. Median household income of the cohort was $35,000/year (range $10,000->$150,000). Overall, 73 (12.3%) patients met criteria for extreme poverty. Families living in extreme poverty were more likely to self-report their race/ethnicity as Hispanic or African-American and report parental education as high school (HS) or lower. Of note, clinical factors did not differ by poverty level. In a logistic regression, non-white race/ethnicity (odds ratio [OR]=7.9, 95%CI=2.4-26.1, P=0.007; reference: non-Hispanic white) and low parental education (≤HS: OR=4.2, 95%CI=2.4-7.4, P=<0.001; reference: college degree or higher) were associated with greater odds of living in extreme poverty. Cumulative incidence of relapse at 3y (Figure) from study entry among patients living in extreme poverty was higher compared to those without extreme poverty (14.3% [95%CI, 7-24] vs 7.6% [95%CI, 5-10], P=0.04). Patients living in extreme poverty had a 1.97-fold higher hazard of relapse compared to those not living in extreme poverty after adjusting for above mentioned covariates (95%CI=1.03-3.76, P=0.04). The reasons for the association between individual level poverty and relapse risk are likely multifactorial and possibly include poor adherence to oral chemotherapy. We found that patients living in extreme poverty had lower adherence to 6MP compared to those not living in extreme poverty (85.2±19.6% vs 90.6±14.8%, P=0.03). Conclusions: Extreme household poverty is associated with increased hazard of relapse in children receiving maintenance therapy for ALL compared to those living without extreme household poverty. Mechanisms investigating this association need to be explored in future studies. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Background: One-fifth of U.S. counties are designated persistent child poverty counties (& GE;20% of children in poverty since 1980). The association between a persistent child poverty environment and mortality in children with cancer is unknown.Methods: Our cohort includes 2,089 children with cancer (2000- 2016) in Alabama. We used multivariable Cox proportional hazards modeling (adjusted for sociodemographics/clinical characteristics) to assess mortality by persistent child poverty designation at 1, 5, and 10 years from diagnosis. Distance to treatment was subse-quently explored.Results: Forty-two percent of the cohort lived in a persistent child poverty county; they were more likely to be African American (P < 0.0001), have public/no insurance (P = 0.0009), and live >100 miles to treatment (P < 0.0001). Children in persistent child poverty counties were 30% more likely to die by 5 years [95% confidence interval (CI) = 1.06-1.59; P = 0.012]. Distance (per 20-mile increase) to treatment was associated with a 9% increased mortality risk (P < 0.0001). Children with both exposures (distance >100 miles and persis-tent child poverty) faced the highest mortality risk at 5 years (HR = 1.80; 95% CI = 1.39-2.33; P < 0.0001). In subanalysis, children exposed to persistent child poverty were at higher risk for cancer-related mortality. However, the risk of health-related mortality did not differ.Conclusions: Among children with cancer from the Deep South, persistent child poverty was a prevalent exposure associated with inferior overall survival. Distance to treatment was independently associated with inferior survival. Children with both exposures had the highest risk of mortality.Impact: Persistent child poverty is associated with inferior survival among children with cancer; mechanisms underlying this disparity warrant investigation.See related commentary by Orjuela-Grimm and Beauchemin, p. 295

PURPOSE:Infant and young childhood medulloblastoma (iMB) is usually treated without craniospinal irradiation (CSI) to avoid neurocognitive late effects. Unfortunately, many children relapse. The purpose of this study was to assess salvage strategies and prognostic features of patients with iMB who relapse after CSI-sparing therapy.METHODS:We assembled a large international cohort of 380 patients with relapsed iMB, age younger than 6 years, and initially treated without CSI. Univariable and multivariable Cox models of postrelapse survival (PRS) were conducted for those treated with curative intent using propensity score analyses to account for confounding factors.RESULTS:The 3-year PRS, for 294 patients treated with curative intent, was 52.4% (95% CI, 46.4 to 58.3) with a median time to relapse from diagnosis of 11 months. Molecular subgrouping was available for 150 patients treated with curative intent, and 3-year PRS for sonic hedgehog (SHH), group 4, and group 3 were 60%, 84%, and 18% (P = .0187), respectively. In multivariable analysis, localized relapse (P = .0073), SHH molecular subgroup (P = .0103), CSI use after relapse (P = .0161), and age ≥ 36 months at initial diagnosis (P = .0494) were associated with improved survival. Most patients (73%) received salvage CSI, and although salvage chemotherapy was not significant in multivariable analysis, its use might be beneficial for a subset of children receiving salvage CSI < 35 Gy (P = .007).CONCLUSION:A substantial proportion of patients with relapsed iMB are salvaged after initial CSI-sparing approaches. Patients with SHH subgroup, localized relapse, older age at initial diagnosis, and those receiving salvage CSI show improved PRS. Future prospective studies should investigate optimal CSI doses and the role of salvage chemotherapy in this population.

Background Poor mercaptopurine (6MP) adherence (mean adherence rate < 90%) increases the relapse risk among children with acute lymphoblastic leukemia (ALL). 6MP adherence remains difficult to measure in real time. Easily measured patient-level factors could identify patients at risk for poor adherence. Methods The authors measured 6MP adherence via electronic monitoring for 6 months per patient. Using data from month 3, they created a risk prediction model for 6MP nonadherence in 407 children with ALL (mean age, 7.7 +/- 4.4 years); they used receiver operating characteristic analyses in the training set (n = 250) and replicated this in the test set (n = 157). Results Age, race/ethnicity, 6MP dose intensity, absolute neutrophil count, 6MP ingestion patterns, and household structure were retained in the prediction model. The model yielded areas under the receiver operating characteristic curve (AUCs) of 0.79 (95% confidence interval [CI], 0.71-0.85) and 0.74 (95% CI, 0.63-0.85) in the training and test sets, respectively. The model performed better for those who were >= 12 years old (AUC, 0.79; 95% CI, 0.59-0.99) than those <12 years old (AUC, 0.70; 95% CI, 0.58-0.81). Using the predicted probability of nonadherence based on receiver operating characteristic analysis, the authors developed a binary risk classifier to classify patients with a high or low probability of nonadherence. The sensitivity and specificity of the binary risk classifier were 71% and 76%, respectively. Adjusted for clinical prognosticators, the risk of relapse was 2.2-fold higher (95% CI, 0.94-5.1; P = .07) among patients with a high probability of nonadherence in comparison with those with a low probability, as identified by the risk prediction model. Conclusions The risk prediction model identified patients with a high probability of nonadherence and could be used in real time to personalize recommendations and interventions in the clinic. Lay Summary The vast majority of children with acute lymphoblastic leukemia, the most common childhood cancer, are cured. The treatment of acute lymphoblastic leukemia includes taking an oral chemotherapy medicine (mercaptopurine) for approximately 2 years. Children who miss doses of this medicine (specifically children who take the medicine less than 90% of the time that it is prescribed) are more likely to suffer leukemia relapse. The authors of this article have measured mercaptopurine adherence with electronic bottle caps to determine characteristics of patients that predict nonadherence, and they have created a prediction tool that could allow physicians to identify and intervene with patients at high risk of nonadherence.

Background The outcomes of deep vein thrombosis (DVT) in children with May-Thurner Syndrome (MTS) remain unclear. Objectives This systematic review and patient-level meta-analysis aims to describe the outcomes of children with MTS presenting with DVT. Methods A systematic review of the published literature was performed. Data related to patients <18 years diagnosed with MTS and DVT was extracted. Risk of bias was assessed using the Murad criteria. Outcomes included vessel patency post-treatment, DVT recurrence, and post-thrombotic syndrome (PTS). Predictive and explanatory models were developed for these outcomes. Results In total, 109 cases were identified (age range 4-17 years; 77 females) in 28 studies; 75% of patients had >= 1 additional risk factor for DVT. PTS was seen in 61% of patients, DVT recurrence in 38%, and complete vessel patency post-treatment in 65%. The models developed to predict and explain PTS performed poorly overall. Recurrent thrombosis (adjusted for age and patency) predicted PTS (odds ratio [OR] 3.36, 95% confidence interval [CI] 1.28-8.82). DVT management strategies (adjusted for age and DVT characteristics) predicted vessel patency (OR 2.10, 95% CI 1.43-3.08). Lack of complete vessel patency (adjusted for age and thrombophilia) predicted recurrent DVT (OR 2.70, 95% CI 1.09-6.67). Sensitivity analyses showed the same direction of effects for all outcomes. Conclusions PTS and DVT recurrence occur frequently in pediatric MTS. PTS prediction is complex and it was not possible to identify early predictors to guide clinical practice. Use of imaging-guided therapy and thrombus burden predicted venous patency, and lack of patency predicted DVT recurrence.

Abstract Introduction: Higher BMI at ALL diagnosis is associated with an increased risk of post-induction residual leukemia (Orgel, Blood 2014) and relapse (Butturini, JCO 2007). However, children may experience significant changes in BMI during the pre-maintenance phases of ALL treatment (Withycombe, Pediatr Blood Cancer 2009), necessitating an examination of the association between BMI during maintenance and relapse risk. We hypothesized that higher BMI during maintenance would be associated with a greater risk of relapse. We also explored the association between BMI and red cell thioguanine (TGN) levels to understand whether BMI-associated variations in TGN biodistribution explained the BMI-relapse association. Methods: We used data from COG-AALL03N1 (primary aim was 6MP adherence) to examine the association between BMI during maintenance and relapse risk. Eligibility for enrollment on AALL03N1 included age ≤21y at ALL diagnosis and receiving maintenance therapy in first remission. The current analysis was limited to patients with wild-type thiopurine methyltransferase genotype. BMI (exposure variable) was calculated as sex- and age-based percentile per CDC normative data, and operationalized as normal/underweight [<85%ile], overweight/obese [85-98%ile] and morbidly obese [≥99%ile]). Hazard of relapse (any site) was estimated using multivariable proportional subdistributional hazards regression after adjusting for age at study enrollment, sex, race/ethnicity, NCI risk group, cytogenetics, 6MP dose intensity (6MPDI), and time from initiation of maintenance. We compared fitted means of red cell TGN levels by BMI groups after adjusting for age at enrollment, sex, race/ethnicity, 6MPDI and time from initiation of maintenance using generalized estimated equations. The association between BMI and relapse risk, as well as between BMI and TGN levels, was also examined in a sub-cohort of patients with available 6MP adherence. Results: The sociodemographic and disease characteristics of the 676 study participants are summarized in the Table. Median BMI%ile was 88.5 (range, 0-100); normal/underweight: 43%, overweight/obese: 45%, and morbidly obese: 12%. As shown in the Figure, cumulative incidence of relapse at 2y from start of maintenance therapy was significantly greater among patients with morbid obesity (11.9±4.3%) when compared to those who were overweight/obese (5.4±1.6%) and underweight/normal weight (4.1±1.4%). After adjusting for the variables listed above, we found that patients with morbid obesity had a 3.2-fold greater hazard of relapse (95%CI=1.4-7.5, P=0.008) when compared to patients who were normal/underweight. After adjusting for age at enrollment, sex, race/ethnicity, 6MPDI, and time from initiation of maintenance, patients with morbid obesity had lower mean red cell TGN levels compared to normal/underweight patients (mean difference: -27.2±7.4 pmol/8 x 10 8 erythrocytes, P=0.0002). However, inclusion of TGN in the model did not alter the association between BMI and hazard of relapse (HR=3.8, 95%CI, 1.6-9.0, P=0.003). These findings did not change after adjusting for 6MP adherence in the sub-cohort with available adherence data (n=435). Conclusion: Morbid obesity during maintenance for childhood ALL is associated with relapse as well as lower systemic exposure to 6MP. However, lower TGN levels do not explain the relation between BMI and relapse risk. Therefore, there is a need to understand the mechanism of the relation between morbid obesity during maintenance and relapse risk in children with ALL. Figure 1 Figure 1. Disclosures No relevant conflicts of interest to declare.

"We share this case study to underscore the difficulty of caring for a terminally ill child when parents wish to withhold diagnostic/prognostic information, review the evidence for and against prognostic disclosure to children, and highlight guiding principles that can help clinicians and other medical staff navigate these difficult circumstances."

To the Editor: As a first-year pediatric oncology fellow, I had a patient with an aggressive, metastatic tumor. Like other oncology trainees and clinicians, I had received and benefited from training in delivering bad news.1 Still, I felt unprepared to navigate my first conversation about the end of life with my patient’s mother. Though my patient’s prognosis was poor, he had a remarkable initial response to chemotherapy. He was discharged home and celebrated Christmas among family. His grandfather shook my hand as they left the hospital, saying, “This is the best day of my life.” I continued to see him as an outpatient for 2 months—until his cancer rapidly progressed. There is really no good way to deliver the worst news to a parent. I wanted my patient’s mom to understand her child was dying and, despite our best efforts, we could not save him. I wanted to tell her I had consulted oncologists around the country, and we were out of meaningful options. I wanted her to know her son’s remaining days were limited. As a parent with my own toddler at home, I could not imagine the indescribable heartbreak of losing a child. I worried about getting this conversation wrong—either by not sharing the gravity of his clinical status or doing so without compassion. I practiced in my car and with our chaplain. Still, sitting with his mom in their hospital room, I couldn’t find the right words. After I tried a few different ways, she looked up and said, “Mom to mom, I need you to tell me what is going on. Is he going to make it?” I told her the truth while we both cried. I think back on this conversation often—how difficult it was to say those words and knowing his mom’s life would be forever changed. What I hadn’t anticipated was telling her directly that her son was dying allowed her to think through what that day might look like. In the coming days, she told me she wanted him to feel peaceful, surrounded by family, with music playing. Though we were not able to save her child, we were able to honor her wishes and give him a peaceful death. Acknowledgments: The author is grateful for this patient’s care team, including Dr. Susie Buckingham, Jolene Ethridge, and Dr. Kim Whelan. Anna Hoppmann, MD, MPHSecond-year fellow, Pediatric Hematology and Oncology, Children’s of Alabama/University of Alabama at Birmingham, Birmingham, Alabama; [email protected]

10530 Background: Poor adherence to 6MP (measured electronically [MEMS]) increases relapse risk in children with ALL (Bhatia et al. JAMA Oncol 2015). Adherence is difficult to assess clinically and non-adherers are more likely to over-report 6MP intake (Landier et al. Blood 2017). Key sociodemographic/clinical factors (Bhatia et al. JCO 2012) and red cell methyl-mercaptopurine (MMP, a 6MP metabolite) levels (Hoppmann et al. ASCO 2017) are associated with non-adherence and could potentially identify non-adherers. Methods: We developed a prediction model for 6MP non-adherence (MEMS adherence rate < 90%), using receiver operating characteristic (ROC) analyses in 407 children with ALL receiving 6MP (mean age 7.7±4.4y; 68% males; 35% Caucasians, 34% Hispanics, 16% African Americans, 15% Asians). The cohort was divided into a training set (n = 250) and test set (n = 157) using stratified random sampling (stratified by race/ethnicity, gender, age and 6MP non-adherence). We used logistic regression with backward variable elimination, guided by change in area under ROC (AUC), to create a prediction model in the training set, using only clinical and sociodemographic variables (Clinical Model). We then generated a model that added 6MP dose-intensity (6MPDI)-adjusted red cell MMP levels to the Clinical Model (Final Model). All models were validated in the test set. Results: Predictors retained in the Training Clinical Model included: age, race/ethnicity, absolute neutrophil count, 6MPDI, family structure, and taking 6MP at the same vs varied time of day (AUC = 0.79; 95%CI 0.72-0.85). The Training Final Model (adding 6MPDI-adjusted MMP to the Clinical Model) yielded an AUC = 0.79 (95%CI 0.72-0.86). The Test Final Model (AUC = 0.79, 95%CI 0.69-0.88) showed significantly superior discrimination compared to the Test Clinical Model (AUC = 0.74, 95%CI 0.63-0.85; P = 0.002). Using a binary classifier with predicted probability of non-adherence ≥0.5, the Test Final Model had an accuracy of 79%, and positive and negative predictive values of 71% and 80%, respectively. Conclusions: We created, validated, and compared 2 risk-prediction models for 6MP non-adherence in children undergoing maintenance chemotherapy. While inclusion of red cell MMP levels provided superior discrimination in identifying non-adherent patients, the Clinical Model (without MMP levels) performed adequately well, and could be used in the clinical setting.

The authors declare that there is no conflict of interest.

10514 Background: Non-adherence to 6MP (monitored with medication event monitoring system [MEMs]) is associated with an increased risk of relapse in children with ALL.(JAMA Oncol, 2015) Self-report over-estimates true medication intake, particularly in non-adherent patients.(Blood, 2017) However, monitoring adherence using MEMs is logistically difficult. We investigated whether red cell 6MP metabolite levels (thioguanine nucleotide [TGN] and methylated mercaptopurine [MMP]) taken together, could identify non-adherent patients. Methods: The analysis included children with ALL in maintenance. To minimize variability in TGN and MMP levels due to pharmacogenetics, we excluded TPMT heterozygotes and homozygote mutants. We also excluded Asians to remove variability due to NUDT15. TGN and MMP levels were drawn at 6 consecutive monthly time points for each patient and averaged. TGN and MMP levels (pmol/8 x 108red cells) were standardized, adjusted for 6MP dose intensity, and then analyzed using cluster analysis (Spath, H. [1980]). Results: The 373 patients eligible for analysis yielded 5 clusters. Cluster #1 (n = 119; mean MMP: 15,656; mean TGN: 158); Cluster #2 (n = 211; MMP: 6,042; TGN: 135); and 3 very small outlying clusters (total N = 43). Adjusting for age, sex, race/ethnicity, cytogenetics and NCI risk, we found that patients in Cluster #2 were 2.6 times as likely to be non-adherent (MEMs-based adherence < 95%) compared to Cluster #1 (95% CI 1.5-4.4; P= 0.0007). Mean MEMs-based adherence was significantly higher for patients in Cluster #1 (94.3%) when compared to those in Cluster #2 (87.8%, p = 0.0002). Using Fine-Gray proportional subdistribution hazards models for competing risks and adjusting for clinical and sociodemographic factors, we found that patients in Cluster #2 were at a 2.3-fold higher risk of relapse compared with those in Cluster #1 (95%CI, 1.0-6.4, p = 0.058). Conclusions: These findings illustrate the potential for using a combination of red cell TGN and MMP levels in identifying non-adherent patients. We propose to use these and clinical and demographic factors associated with non-adherence in creating an adherence calculator.