High-Dimensional Precision Medicine From Patient-Derived Xenografts.

The complexity of human cancer often results in significant heterogeneity in response to treatment. Precision medicine offers the potential to improve patient outcomes by leveraging this heterogeneity. Individualized treatment rules (ITRs) formalize precision medicine as maps from the patient covariate space into the space of allowable treatments. The optimal ITR is that which maximizes the mean of a clinical outcome in a population of interest. Patient-derived xenograft (PDX) studies permit the evaluation of multiple treatments within a single tumor, and thus are ideally suited for estimating optimal ITRs. PDX data are characterized by correlated outcomes, a high-dimensional feature space, and a large number of treatments. Here we explore machine learning methods for estimating optimal ITRs from PDX data. We analyze data from a large PDX study to identify biomarkers that are informative for developing personalized treatment recommendations in multiple cancers. We estimate optimal ITRs using regression-based (Q-learning) and direct-search methods (outcome weighted learning). Finally, we implement a superlearner approach to combine multiple estimated ITRs and show that the resulting ITR performs better than any of the input ITRs, mitigating uncertainty regarding user choice. Our results indicate that PDX data are a valuable resource for developing individualized treatment strategies in oncology. Supplementary materials for this article are available online.

Assessment of a Precision Medicine Analysis of a Behavioral Counseling Strategy to Improve Adherence to Diabetes Self-management Among Youth: A Post Hoc Analysis of the FLEX Trial.

Importance: The Flexible Lifestyles Empowering Change (FLEX) trial, an 18-month randomized clinical trial testing an adaptive behavioral intervention in adolescents with type 1 diabetes, showed no overall treatment effect for its primary outcome, change in hemoglobin A1c (HbA1c) percentage of total hemoglobin, but demonstrated benefit for quality of life (QoL) as a prespecified secondary outcome. Objective: To apply a novel statistical method for post hoc analysis that derives an individualized treatment rule (ITR) to identify FLEX participants who may benefit from intervention based on changes in HbA1c percentage (primary outcome), QoL, and body mass index z score (BMIz) (secondary outcomes) during 18 months. Design, Setting, and Participants: This multisite clinical trial enrolled 258 adolescents aged 13 to 16 years with type 1 diabetes for 1 or more years, who had literacy in English, HbA1c percentage of total hemoglobin from 8.0% to 13.0%, a participating caregiver, and no other serious medical conditions. From January 5, 2014, to April 4, 2016, 258 adolescents were recruited. The post hoc analysis excluded adolescents missing outcome measures at 18 months (2 participants [0.8%]) or continuous glucose monitoring data at baseline (40 participants [15.5%]). Data were analyzed from April to December 2018. Interventions: The FLEX intervention included a behavioral counseling strategy that integrated motivational interviewing and problem-solving skills training to increase adherence to diabetes self-management. The control condition entailed usual diabetes care. Main Outcomes and Measures: Subgroups of FLEX participants were derived from an ITR estimating which participants would benefit from intervention, which would benefit from control conditions, and which would be indifferent. Multiple imputation by chained equations and reinforcement learning trees were used to estimate the ITR. Subgroups based on ITR pertaining to changes during 18 months in 3 univariate outcomes (ie, HbA1c percentage, QoL, and BMIz) and a composite outcome were compared by baseline demographic, clinical, and psychosocial characteristics. Results: Data from 216 adolescents in the FLEX trial were reanalyzed (166 [76.9%] non-Hispanic white; 108 teenaged girls [50.0%]; mean [SD] age, 14.9 [1.1] years; mean [SD] diabetes duration, 6.3 [3.7] years). For the univariate outcomes, a large proportion of FLEX participants had equivalent predicted outcomes under intervention vs usual care settings, regardless of randomization, and were assigned to the muted group (HbA1c: 105 participants [48.6%]; QoL: 63 participants [29.2%]; BMIz: 136 participants [63.0%]). Regarding the BMIz univariate outcome, mean baseline BMIz of participants assigned to the muted group was lower than that of those assigned to the intervention and control groups (muted vs intervention: mean difference, 0.48; 95% CI, 0.21 to 0.75; P = .002; muted vs control: mean difference, 0.86; 95% CI, 0.61 to 1.11; P < .001); this group also had a higher proportion of individuals with underweight or normal weight using weight status cutoffs (95 [69.9%] in muted group vs 24 [54.6%] in intervention group and 11 [30.6%] in control group; χ24 = 24.67; P < .001). The approach identified subgroups estimated to benefit based on HbA1c percentage (54 participants [25.0%]), QoL (89 participants [41.2%]), and BMIz (44 participants [20.4%]). Regarding the HbA1c percentage outcome, participants expected to benefit from the intervention did not have significantly higher baseline HbA1c percentages than those expected to benefit from usual care (9.4% vs 9.2%; difference, 0.2%; 95% CI, -0.16% to 0.56%; P = .44). However, participants in the muted group had higher mean HbA1c percentages at baseline than those assigned to the intervention or control groups (muted vs intervention: 9.9% vs 9.4%; difference, 0.5%; 95% CI, 0.13% to 0.89%; P = .02; muted vs control; 9.9% vs 9.2%; difference, 0.7%; 95% CI, 0.34% to 1.08%; P = .001). No significant differences were found between subgroups estimated to benefit in terms of the composite outcome from the FLEX intervention (91 participants [42.1%]) vs usual care (125 participants [57.9%]). Conclusions and Relevance: The precision medicine approach represents a conceptually and analytically novel approach to post hoc subgroup identification. More work is needed to understand markers of positive response to the FLEX intervention. Trial Registration: ClinicalTrial.gov identifier: NCT01286350.

Tobacco use and psychosis risk in persons at clinical high risk.

AIM: To evaluate the role of tobacco use in the development of psychosis in individuals at clinical high risk. METHOD: The North American Prodrome Longitudinal Study is a 2-year multi-site prospective case control study of persons at clinical high risk that aims to better understand predictors and mechanisms for the development of psychosis. The cohort consisted of 764 clinical high risk and 279 healthy comparison subjects. Clinical assessments included tobacco and substance use and several risk factors associated with smoking in general population studies. RESULTS: Clinical high risk subjects were more likely to smoke cigarettes than unaffected subjects (light smoking odds ratio [OR] = 3.0, 95% confidence interval [CI] = 1.9-5; heavy smoking OR = 4.8, 95% CI = 1.7-13.7). In both groups, smoking was associated with mood, substance use, stress and perceived discrimination and in clinical high risk subjects with childhood emotional neglect and adaption to school. Clinical high risk subjects reported higher rates of several factors previously associated with smoking, including substance use, anxiety, trauma and perceived discrimination. After controlling for these potential factors, the relationship between clinical high risk state and smoking was no longer significant (light smoking OR = 0.9, 95% CI = 0.4-2.2; heavy smoking OR = 0.3, 95% CI = 0.05-2.3). Moreover, baseline smoking status (hazard ratio [HR] = 1.16, 95% CI = 0.82-1.65) and categorization as ever smoked (HR = 1.3, 95% CI = 0.8-2.1) did not predict time to conversion. CONCLUSION: Persons at high risk for psychosis are more likely to smoke and have more factors associated with smoking than controls. Smoking status in clinical high risk subjects does not predict conversion. These findings do not support a causal relationship between smoking and psychosis.