Reinforced risk prediction with budget constraint using irregularly measured data from electronic health records.

Uncontrolled glycated hemoglobin (HbA1c) levels are associated with adverse events among complex diabetic patients. These adverse events present serious health risks to affected patients and are associated with significant financial costs. Thus, a high-quality predictive model that could identify high-risk patients so as to inform preventative treatment has the potential to improve patient outcomes while reducing healthcare costs. Because the biomarker information needed to predict risk is costly and burdensome, it is desirable that such a model collect only as much information as is needed on each patient so as to render an accurate prediction. We propose a sequential predictive model that uses accumulating patient longitudinal data to classify patients as: high-risk, low-risk, or uncertain. Patients classified as high-risk are then recommended to receive preventative treatment and those classified as low-risk are recommended to standard care. Patients classified as uncertain are monitored until a high-risk or low-risk determination is made. We construct the model using claims and enrollment files from Medicare, linked with patient Electronic Health Records (EHR) data. The proposed model uses functional principal components to accommodate noisy longitudinal data and weighting to deal with missingness and sampling bias. The proposed method demonstrates higher predictive accuracy and lower cost than competing methods in a series of simulation experiments and application to data on complex patients with diabetes.

Cost-Effectiveness Analysis Evaluating Delivery Strategies for Pain Coping Skills Training in Women With Breast Cancer.

Pain coping skills training (PCST) is efficacious in patients with cancer, but clinical access is limited. To inform implementation, as a secondary outcome, we estimated the cost-effectiveness of 8 dosing strategies of PCST evaluated in a sequential multiple assignment randomized trial among women with breast cancer and pain (N = 327). Women were randomized to initial doses and re-randomized to subsequent doses based on their initial response (ie, ≥30% pain reduction). A decision-analytic model was designed to incorporate costs and benefits associated with 8 different PCST dosing strategies. In the primary analysis, costs were limited to resources required to deliver PCST. Quality-adjusted life-years (QALYs) were modeled based on utility weights measured with the EuroQol-5 dimension 5-level at 4 assessments over 10 months. A probabilistic sensitivity analysis was performed to account for parameter uncertainty. Implementation of PCST initiated with the 5-session protocol was more costly ($693-853) than strategies initiated with the 1-session protocol ($288-496). QALYs for strategies beginning with the 5-session protocol were greater than for strategies beginning with the 1-session protocol. With the goal of implementing PCST as part of comprehensive cancer treatment and with willingness-to-pay thresholds ranging beyond $20,000 per QALY, the strategy most likely to provide the greatest number of QALYs at an acceptable cost was a 1-session PCST protocol followed by either 5 maintenance telephone calls for responders or 5 sessions of PCST for nonresponders. A PCST program with 1 initial session and subsequent dosing based on response provides good value and improved outcomes. PERSPECTIVE: This article presents the results of a cost analysis of the delivery of PCST, a nonpharmacological intervention, to women with breast cancer and pain. Results could potentially provide important cost-related information to health care providers and systems on the use of an efficacious and accessible nonmedication strategy for pain management. TRIALS REGISTRATION: ClinicalTrials.gov: NCT02791646, registered 6/2/2016.

Behavioral cancer pain intervention dosing: results of a Sequential Multiple Assignment Randomized Trial.

ABSTRACT: Behavioral pain management interventions are efficacious for reducing pain in patients with cancer. However, optimal dosing of behavioral pain interventions for pain reduction is unknown, and this hinders routine clinical use. A Sequential Multiple Assignment Randomized Trial (SMART) was used to evaluate whether varying doses of Pain Coping Skills Training (PCST) and response-based dose adaptation can improve pain management in women with breast cancer. Participants (N = 327) had stage I-IIIC breast cancer and a worst pain score of > 5/10. Pain severity (a priori primary outcome) was assessed before initial randomization (1:1 allocation) to PCST-Full (5 sessions) or PCST-Brief (1 session) and 5 to 8 weeks later. Responders ( > 30% pain reduction) were rerandomized to a maintenance dose or no dose and nonresponders (<30% pain reduction) to an increased or maintenance dose. Pain severity was assessed again 5 to 8 weeks later (assessment 3) and 6 months later (assessment 4). As hypothesized, PCST-Full resulted in greater mean percent pain reduction than PCST-Brief (M [SD] = -28.5% [39.6%] vs M [SD]= -14.8% [71.8%]; P = 0.041). At assessment 3 after second dosing, all intervention sequences evidenced pain reduction from assessment 1 with no differences between sequences. At assessment 4, all sequences evidenced pain reduction from assessment 1 with differences between sequences ( P = 0.027). Participants initially receiving PCST-Full had greater pain reduction at assessment 4 ( P = 0.056). Varying PCST doses led to pain reduction over time. Intervention sequences demonstrating the most durable decreases in pain reduction included PCST-Full. Pain Coping Skills Training with intervention adjustment based on response can produce sustainable pain reduction.

Hand Grip Strength Predicts Mortality and Quality of Life in Heart Failure: Insights From the Singapore Cohort of Patients With Advanced Heart Failure.

BACKGROUND: Frailty is prevalent among patients with heart failure (HF) and is associated with increased mortality rates and worse patient-centered outcomes. Hand grip strength (GS) has been proposed as a single-item marker of frailty and a potential screening tool to identify patients most likely to benefit from therapies that target frailty so as to improve quality of life (QoL) and clinical outcomes. We assessed the association of longitudinal decline in GS with all-cause mortality and QoL. Decline in GS is associated with increased risk of all-cause mortality and worse overall and domain-specific (physical, functional, emotional, social) QoL among patients with advanced HF. METHODS: We used data from a prospective, observational cohort of patients with New York Heart Association class III or IV HF in Singapore. Patients' overall and domain-specific QoL were assessed, and GS was measured every 4 months. We constructed a Kaplan-Meier plot with GS at baseline dichotomized into categories of weak (≤ 5th percentile) and normal (> 5th percentile) based on the GS in a healthy Singapore population of the same sex and age. Missing GS measurements were imputed using chained equations. We jointly modeled longitudinal GS measurements and survival time, adjusting for comorbidities. We used mixed effects models to evaluate the associations between GS and QoL. RESULTS: Among 251 patients (mean age 66.5 ± 12.0 years; 28.3% female), all-cause mortality occurred in 58 (23.1%) patients over a mean follow-up duration of 3.0 ± 1.3 years. Patients with weak GS had decreased survival rates compared to those with normal GS (log-rank P = 0.033). In the joint model of longitudinal GS and survival time, a decrease of 1 unit in GS was associated with a 12% increase in rate of mortality (hazard ratio: 1.12; 95% confidence interval: 1.05-1.20; P = < 0.001). Higher GS was associated with higher overall QoL (ß (SE) = 0.36 (0.07); P = < 0.001) and higher domain-specific QoL, including physical (ß [SE] = 0.13 [0.03]; P = < 0.001), functional (ß [SE] = 0.12 [0.03]; P = < 0.001), and emotional QoL (ß [SE] = 0.08 [0.02]; P = < 0.001). Higher GS was associated with higher social QoL, but this was not statistically significant (ß [SE] = 0.04 [0.03]; P = 0.122). CONCLUSIONS: Among patients with advanced HF, longitudinal decline in GS was associated with worse survival rates and QoL. Further studies are needed to evaluate whether incorporating GS into patient selection for HF therapies leads to improved survival rates and patient-centered outcomes.

Understanding heterogeneity of responses to, and optimizing clinical efficacy of, exercise training in older adults: NIH NIA Workshop summary.

Exercise is a cornerstone of preventive medicine and a promising strategy to intervene on the biology of aging. Variation in the response to exercise is a widely accepted concept that dates back to the 1980s with classic genetic studies identifying sequence variations as modifiers of the VO2max response to training. Since that time, the literature of exercise response variance has been populated with retrospective analyses of existing datasets that are limited by a lack of statistical power from technical error of the measurements and small sample sizes, as well as diffuse outcomes, very few of which have included older adults. Prospective studies that are appropriately designed to interrogate exercise response variation in key outcomes identified a priori and inclusive of individuals over the age of 70 are long overdue. Understanding the underlying intrinsic (e.g., genetics and epigenetics) and extrinsic (e.g., medication use, diet, chronic disease) factors that determine robust versus poor responses to various exercise factors will be used to improve exercise prescription to target the pillars of aging and optimize the clinical efficacy of exercise training in older adults. This review summarizes the proceedings of the NIA-sponsored workshop entitled, "Understanding Heterogeneity of Responses to, and Optimizing Clinical Efficacy of, Exercise Training in Older Adults" and highlights the importance and current state of exercise response variation research, particularly in older adults, prevailing challenges, and future directions.

Telehealth and racial disparities in colorectal cancer screening: A pilot study of how virtual clinician characteristics influence screening intentions.

Introduction: Racial disparities in colorectal cancer (CRC) can be addressed through increased adherence to screening guidelines. In real-life encounters, patients may be more willing to follow screening recommendations delivered by a race concordant clinician. The growth of telehealth to deliver care provides an opportunity to explore whether these effects translate to a virtual setting. The primary purpose of this pilot study is to explore the relationships between virtual clinician (VC) characteristics and CRC screening intentions after engagement with a telehealth intervention leveraging technology to deliver tailored CRC prevention messaging. Methods: Using a posttest-only design with three factors (VC race-matching, VC gender, intervention type), participants (N = 2267) were randomised to one of eight intervention treatments. Participants self-reported perceptions and behavioral intentions. Results: The benefits of matching participants with a racially similar VC trended positive but did not reach statistical significance. Specifically, race-matching positively influenced screening intentions for Black participants but not for Whites (b = 0.29, p = 0.10). Importantly, perceptions of credibility, attractiveness, and message relevance significantly influenced screening intentions and the relationship with race-matching. Conclusions: To reduce racial CRC screening disparities, investments are needed to identify patient-focused interventions to address structural barriers to screening. This study suggests that telehealth interventions that match Black patients with a Black VC can enhance perceptions of credibility and message relevance, which may then improve screening intentions. Future research is needed to examine how to increase VC credibility and attractiveness, as well as message relevance without race-matching.

Reinforcement Learning Methods in Public Health.

PURPOSE: Reinforcement learning (RL) is the subfield of machine learning focused on optimal sequential decision making under uncertainty. An optimal RL strategy maximizes cumulative utility by experimenting only if and when the information generated by experimentation is likely to outweigh associated short-term costs. RL represents a holistic approach to decision making that evaluates the impact of every action (ie, data collection, allocation of resources, and treatment assignment) in terms of short-term and long-term utility to stakeholders. Thus, RL is an ideal model for a number of complex decision problems that arise in public health, including resource allocation in a pandemic, monitoring or testing, and adaptive sampling for hidden populations. Nevertheless, although RL has been applied successfully in a wide range of domains, including precision medicine, it has not been widely adopted in public health. The purposes of this review are to introduce key ideas in RL and to identify challenges and opportunities associated with the application of RL in public health. METHODS: We provide a nontechnical review of the theoretical and methodologic underpinnings of RL. A running example of RL for the management of an infectious disease is used to illustrate ideas. FINDINGS: RL has the potential to make a transformative impact in a range of sequential decision problems in public health. By allocating resources if, when, and where they are most impactful, RL can improve health outcomes while reducing resource consumption. IMPLICATIONS: Public health researchers and stakeholders should consider RL as a means of efficiently using data to inform optimal evidence-based decision making.

A spatiotemporal recommendation engine for malaria control.

Malaria is an infectious disease affecting a large population across the world, and interventions need to be efficiently applied to reduce the burden of malaria. We develop a framework to help policy-makers decide how to allocate limited resources in realtime for malaria control. We formalize a policy for the resource allocation as a sequence of decisions, one per intervention decision, that map up-to-date disease related information to a resource allocation. An optimal policy must control the spread of the disease while being interpretable and viewed as equitable to stakeholders. We construct an interpretable class of resource allocation policies that can accommodate allocation of resources residing in a continuous domain and combine a hierarchical Bayesian spatiotemporal model for disease transmission with a policy-search algorithm to estimate an optimal policy for resource allocation within the pre-specified class. The estimated optimal policy under the proposed framework improves the cumulative long-term outcome compared with naive approaches in both simulation experiments and application to malaria interventions in the Democratic Republic of the Congo.

Estimation and Optimization of Composite Outcomes.

There is tremendous interest in precision medicine as a means to improve patient outcomes by tailoring treatment to individual characteristics. An individualized treatment rule formalizes precision medicine as a map from patient information to a recommended treatment. A treatment rule is defined to be optimal if it maximizes the mean of a scalar outcome in a population of interest, e.g., symptom reduction. However, clinical and intervention scientists often seek to balance multiple and possibly competing outcomes, e.g., symptom reduction and the risk of an adverse event. One approach to precision medicine in this setting is to elicit a composite outcome which balances all competing outcomes; unfortunately, eliciting a composite outcome directly from patients is difficult without a high-quality instrument, and an expert-derived composite outcome may not account for heterogeneity in patient preferences. We propose a new paradigm for the study of precision medicine using observational data that relies solely on the assumption that clinicians are approximately (i.e., imperfectly) making decisions to maximize individual patient utility. Estimated composite outcomes are subsequently used to construct an estimator of an individualized treatment rule which maximizes the mean of patient-specific composite outcomes. The estimated composite outcomes and estimated optimal individualized treatment rule provide new insights into patient preference heterogeneity, clinician behavior, and the value of precision medicine in a given domain. We derive inference procedures for the proposed estimators under mild conditions and demonstrate their finite sample performance through a suite of simulation experiments and an illustrative application to data from a study of bipolar depression.

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.

A Pilot Study Examining the Efficacy of Delivering Colorectal Cancer Screening Messages via Virtual Health Assistants.

INTRODUCTION: Patients are more likely to complete colorectal cancer screening when recommended by a race-concordant healthcare provider. Leveraging virtual healthcare assistants to deliver tailored screening interventions may promote adherence to colorectal cancer screening guidelines among diverse patient populations. The purpose of this pilot study is to determine the efficacy of the Agent Leveraging Empathy for eXams virtual healthcare assistant intervention to increase patient intentions to talk to their doctor about colorectal cancer screening. It also examines the influence of animation and race concordance on intentions to complete colorectal cancer screening. METHODS: White and Black adults (N=1,363) aged 50-73 years and not adherent to colorectal cancer screening guidelines were recruited from Qualtrics Panels in 2018 to participate in a 3-arm (animated virtual healthcare assistant, static virtual healthcare assistant, attention control) message design experiment. In 2020, a probit regression model was used to identify the intervention effects. RESULTS: Participants assigned to the animated virtual healthcare assistant (p<0.01) reported higher intentions to talk to their doctor about colorectal cancer screening than participants assigned to the other conditions. There was a significant effect of race concordance on colorectal cancer screening intentions but only in the static virtual healthcare assistant condition (p=0.04). Participant race, age, trust in healthcare providers, health literacy, and cancer information overload were also significant predictors of colorectal cancer screening intentions. CONCLUSIONS: Animated virtual healthcare assistants were efficacious compared with the static virtual healthcare assistant and attention control conditions. The influence of race concordance between source and participant was inconsistent across conditions. This warrants additional investigation in future studies given the potential for virtual healthcare assistant‒assisted interventions to promote colorectal cancer screening within guidelines.

Receiver operating characteristic curves and confidence bands for support vector machines.

Many problems that appear in biomedical decision-making, such as diagnosing disease and predicting response to treatment, can be expressed as binary classification problems. The support vector machine (SVM) is a popular classification technique that is robust to model misspecification and effectively handles high-dimensional data. The relative costs of false positives and false negatives can vary across application domains. The receiving operating characteristic (ROC) curve provides a visual representation of the trade-off between these two types of errors. Because the SVM does not produce a predicted probability, an ROC curve cannot be constructed in the traditional way of thresholding a predicted probability. However, a sequence of weighted SVMs can be used to construct an ROC curve. Although ROC curves constructed using weighted SVMs have great potential for allowing ROC curves analyses that cannot be done by thresholding predicted probabilities, their theoretical properties have heretofore been underdeveloped. We propose a method for constructing confidence bands for the SVM ROC curve and provide the theoretical justification for the SVM ROC curve by showing that the risk function of the estimated decision rule is uniformly consistent across the weight parameter. We demonstrate the proposed confidence band method using simulation studies. We present a predictive model for treatment response in breast cancer as an illustrative example.

Bayesian Nonparametric Policy Search with Application to Periodontal Recall Intervals.

Tooth loss from periodontal disease is a major public health burden in the United States. Standard clinical practice is to recommend a dental visit every six months; however, this practice is not evidence-based, and poor dental outcomes and increasing dental insurance premiums indicate room for improvement. We consider a tailored approach that recommends recall time based on patient characteristics and medical history to minimize disease progression without increasing resource expenditures. We formalize this method as a dynamic treatment regime which comprises a sequence of decisions, one per stage of intervention, that follow a decision rule which maps current patient information to a recommendation for their next visit time. The dynamics of periodontal health, visit frequency, and patient compliance are complex, yet the estimated optimal regime must be interpretable to domain experts if it is to be integrated into clinical practice. We combine non-parametric Bayesian dynamics modeling with policy-search algorithms to estimate the optimal dynamic treatment regime within an interpretable class of regimes. Both simulation experiments and application to a rich database of electronic dental records from the HealthPartners HMO shows that our proposed method leads to better dental health without increasing the average recommended recall time relative to competing methods.

Estimating Dynamic Treatment Regimes in Mobile Health Using V-learning.

The vision for precision medicine is to use individual patient characteristics to inform a personalized treatment plan that leads to the best possible health-care for each patient. Mobile technologies have an important role to play in this vision as they offer a means to monitor a patient's health status in real-time and subsequently to deliver interventions if, when, and in the dose that they are needed. Dynamic treatment regimes formalize individualized treatment plans as sequences of decision rules, one per stage of clinical intervention, that map current patient information to a recommended treatment. However, most existing methods for estimating optimal dynamic treatment regimes are designed for a small number of fixed decision points occurring on a coarse time-scale. We propose a new reinforcement learning method for estimating an optimal treatment regime that is applicable to data collected using mobile technologies in an out-patient setting. The proposed method accommodates an indefinite time horizon and minute-by-minute decision making that are common in mobile health applications. We show that the proposed estimators are consistent and asymptotically normal under mild conditions. The proposed methods are applied to estimate an optimal dynamic treatment regime for controlling blood glucose levels in patients with type 1 diabetes.

Evaluation of a Stepped-Care eHealth HIV Prevention Program for Diverse Adolescent Men Who Have Sex With Men: Protocol for a Hybrid Type 1 Effectiveness Implementation Trial of SMART.

BACKGROUND: Adolescent men who have sex with men (AMSM), aged 13 to 18 years, account for more than 80% of teen HIV occurrences. Despite this disproportionate burden, there is a conspicuous lack of evidence-based HIV prevention programs. Implementation issues are critical as traditional HIV prevention delivery channels (eg, community-based organizations, schools) have significant access limitations for AMSM. As such, eHealth interventions, such as our proposed SMART program, represent an excellent modality for delivering AMSM-specific intervention material where youth are. OBJECTIVE: This randomized trial aimed to test the effectiveness of the SMART program in reducing condom-less anal sex and increasing condom self-efficacy, condom use intentions, and HIV testing for AMSM. We also plan to test whether SMART has differential effectiveness across important subgroups of AMSM based on race and ethnicity, urban versus rural residence, age, socioeconomic status, and participation in an English versus a Spanish version of SMART. METHODS: Using a sequential multiple assignment randomized trial design, we will evaluate the impact of a stepped-care package of increasingly intensive eHealth interventions (ie, the universal, information-based SMART Sex Ed; the more intensive, selective SMART Squad; and a higher cost, indicated SMART Sessions). All intervention content is available in English and Spanish. Participants are recruited primarily from social media sources using paid and unpaid advertisements. RESULTS: The trial has enrolled 1285 AMSM aged 13 to 18 years, with a target enrollment of 1878. Recruitment concluded in June 2020. Participants were recruited from 49 US states as well as Puerto Rico and the District of Columbia. Assessments of intervention outcomes at 3, 6, 9, and 12 months are ongoing. CONCLUSIONS: SMART is the first web-based program for AMSM to take a stepped-care approach to sexual education and HIV prevention. This design indicates that SMART delivers resources to all adolescents, but more costly treatments (eg, video chat counseling in SMART Sessions) are conserved for individuals who need them the most. SMART has the potential to reach AMSM to provide them with a sex-positive curriculum that empowers them with the information, motivation, and skills to make better health choices. TRIAL REGISTRATION: ClinicalTrials.gov Identifier NCT03511131; https://clinicaltrials.gov/ct2/show/NCT03511131. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): DERR1-10.2196/19701.

Use of standardized bioinformatics for the analysis of fungal DNA signatures applied to sample provenance.

The use of environmental trace material to aid criminal investigations is an ongoing field of research within forensic science. The application of environmental material thus far has focused upon a variety of different objectives relevant to forensic biology, including sample provenance (also referred to as sample attribution). The capability to predict the provenance or origin of an environmental DNA sample would be an advantageous addition to the suite of investigative tools currently available. A metabarcoding approach is often used to predict sample provenance, through the extraction and comparison of the DNA signatures found within different environmental materials, such as the bacteria within soil or fungi within dust. Such approaches are combined with bioinformatics workflows and statistical modelling, often as part of large-scale study, with less emphasis on the investigation of the adaptation of these methods to a smaller scale method for forensic use. The present work was investigating a small-scale approach as an adaptation of a larger metabarcoding study to develop a model for global sample provenance using fungal DNA signatures collected from dust swabs. This adaptation was to facilitate a standardized method for consistent, reproducible sample treatment, including bioinformatics processing and final application of resulting data to the available prediction model. To investigate this small-scale method, 76 DNA samples were treated as anonymous test samples and analyzed using the standardized process to demonstrate and evaluate processing and customized sequence data analysis. This testing included samples originating from countries previously used to train the model, samples artificially mixed to represent multiple or mixed countries, as well as outgroup samples. Positive controls were also developed to monitor laboratory processing and bioinformatics analysis. Through this evaluation we were able to demonstrate that the samples could be processed and analyzed in a consistent manner, facilitated by a relatively user-friendly bioinformatic pipeline for sequence data analysis. Such investigation into standardized analyses and application of metabarcoding data is of key importance for the future use of applied microbiology in forensic science.

Assessing Tuning Parameter Selection Variability in Penalized Regression.

Penalized regression methods that perform simultaneous model selection and estimation are ubiquitous in statistical modeling. The use of such methods is often unavoidable as manual inspection of all possible models quickly becomes intractable when there are more than a handful of predictors. However, automated methods usually fail to incorporate domain-knowledge, exploratory analyses, or other factors that might guide a more interactive model-building approach. A hybrid approach is to use penalized regression to identify a set of candidate models and then to use interactive model-building to examine this candidate set more closely. To identify a set of candidate models, we derive point and interval estimators of the probability that each model along a solution path will minimize a given model selection criterion, for example, Akaike information criterion, Bayesian information criterion (AIC, BIC), etc., conditional on the observed solution path. Then models with a high probability of selection are considered for further examination. Thus, the proposed methodology attempts to strike a balance between algorithmic modeling approaches that are computationally efficient but fail to incorporate expert knowledge, and interactive modeling approaches that are labor intensive but informed by experience, intuition, and domain knowledge. Supplementary materials for this article are available online.

Efficient augmentation and relaxation learning for individualized treatment rules using observational data.

Individualized treatment rules aim to identify if, when, which, and to whom treatment should be applied. A globally aging population, rising healthcare costs, and increased access to patient-level data have created an urgent need for high-quality estimators of individualized treatment rules that can be applied to observational data. A recent and promising line of research for estimating individualized treatment rules recasts the problem of estimating an optimal treatment rule as a weighted classification problem. We consider a class of estimators for optimal treatment rules that are analogous to convex large-margin classifiers. The proposed class applies to observational data and is doubly-robust in the sense that correct specification of either a propensity or outcome model leads to consistent estimation of the optimal individualized treatment rule. Using techniques from semiparametric efficiency theory, we derive rates of convergence for the proposed estimators and use these rates to characterize the bias-variance trade-off for estimating individualized treatment rules with classification-based methods. Simulation experiments informed by these results demonstrate that it is possible to construct new estimators within the proposed framework that significantly outperform existing ones. We illustrate the proposed methods using data from a labor training program and a study of inflammatory bowel syndrome.

Precision Medicine.

Precision medicine seeks to maximize the quality of healthcare by individualizing the healthcare process to the uniquely evolving health status of each patient. This endeavor spans a broad range of scientific areas including drug discovery, genetics/genomics, health communication, and causal inference all in support of evidence-based, i.e., data-driven, decision making. Precision medicine is formalized as a treatment regime which comprises a sequence of decision rules, one per decision point, which map up-to-date patient information to a recommended action. The potential actions could be the selection of which drug to use, the selection of dose, timing of administration, specific diet or exercise recommendation, or other aspects of treatment or care. Statistics research in precision medicine is broadly focused on methodological development for estimation of and inference for treatment regimes which maximize some cumulative clinical outcome. In this review, we provide an overview of this vibrant area of research and present important and emerging challenges.

Identifying optimal dosage regimes under safety constraints: An application to long term opioid treatment of chronic pain.

There is growing interest and investment in precision medicine as a means to provide the best possible health care. A treatment regime formalizes precision medicine as a sequence of decision rules, one per clinical intervention period, that specify if, when and how current treatment should be adjusted in response to a patient's evolving health status. It is standard to define a regime as optimal if, when applied to a population of interest, it maximizes the mean of some desirable clinical outcome, such as efficacy. However, in many clinical settings, a high-quality treatment regime must balance multiple competing outcomes; eg, when a high dose is associated with substantial symptom reduction but a greater risk of an adverse event. We consider the problem of estimating the most efficacious treatment regime subject to constraints on the risk of adverse events. We combine nonparametric Q-learning with policy-search to estimate a high-quality yet parsimonious treatment regime. This estimator applies to both observational and randomized data, as well as settings with variable, outcome-dependent follow-up, mixed treatment types, and multiple time points. This work is motivated by and framed in the context of dosing for chronic pain; however, the proposed framework can be applied generally to estimate a treatment regime which maximizes the mean of one primary outcome subject to constraints on one or more secondary outcomes. We illustrate the proposed method using data pooled from 5 open-label flexible dosing clinical trials for chronic pain.