Technological Innovation in Health Care Delivery

Health data have unique characteristics in how fragmented, sensitive, complex, and regulated they are. These characteristics create challenges in developing impactful solutions and in adopting new technological breakthroughs. Using examples of real-world implementations, this session will highlight some of the emerging solutions to address the challenges with large scale health data integration and analysis. Federal policies like TEFCA (Trust Exchange Framework and Common Agreement) and EHR (electronic health records) certification requirements, along with new technologies like Web 3.0, blockchain, and common data standards for interoperability are creating possibilities that need to be tested, adopted, and further refined.

Biomedical breakthroughs such as gene and cell therapies are delivering cures to serious illnesses such as spinal muscular atrophy, hemophilia, certain types of blindness, and other previously untreatable conditions. However, they often come with eye-popping price tags, imposing challenges to healthcare insurers and patients. In this talk, Prof. Lo will review these challenges, describe some of the subtleties of the healthcare delivery ecosystem—including drug pricing practices—and propose a solution. Often called the “Netflix” model, the proposal involves allowing healthcare payers to subscribe to the lowest-cost provider of re-insurance for these expensive one-time therapeutics: the drug manufacturer itself. The subscription model for cures is among the most efficient methods of health insurance, and has advantages for all stakeholders, including payers, patients, and drug companies.

We consider the problem of sequentially allocating sample observations to learn personalized treatment strategies, motivated by the design of adaptive clinical trials that aim to learn the best treatment as a function of patient covariates. In such settings there may be clinical knowledge of which covariates are predictive (they may interact with the treatment choice) and which are prognostic (they may influence the outcome independent of treatment choice). We extend the expected value of information (EVI)/knowledge gradient framework to develop useful heuristics for a context with predictive and prognostic covariates and a delay in observing outcomes. We also propose and analyze closely related Monte Carlo-based allocation policies to enhance our proposal’s computational efficiency and applicability for adaptive contextual learning. We show that several of our proposed allocation policies are asymptotically optimal in learning treatment strategies. We run simulation experiments motivated by an application for clinical trial design to assess potential treatments of sepsis. We illustrate that the proposed EVI-based allocation policies, with knowledge about which covariates are predictive and prognostic, can improve the rate of inference relative to some existing approaches to adaptive contextual learning.

Healthcare decision makers, such as the National Institute of Health and Care Excellence in the UK, use cost-effectiveness analysis and modelling to compare the costs and effects of disease management strategies. These analyses rely on limited evidence and decisions are often uncertain. Value of information (VoI) analysis quantifies the monetary value to decision makers of gathering further evidence. VoI requires nested Monte Carlo simulation to estimate the uncertain benefits of further research, which is computationally impractical for all but the simplest of cost-effectiveness models. Model regression and approximation approaches, including Gaussian processes, generalised additive models, and integrated nested Laplace approximation (INLA), have come into use as short-cut approaches to estimating VoI. However, these approaches may be unreliable for realistic models.

In my talk, I will explain these issues in greater detail and highlight problems with model regression and approximation when applied to realistic economic models. As an alternative, I will present adaptations of Multilevel and Quasi Monte Carlo sampling schemes from computational finance to the estimation of VoI. These achieve the same accuracy and precision of standard Monte Carlo with lower computational cost by minimising the variance and bias of their VoI estimators. I will apply both methods to example cost-effectiveness models, including a model used in the UK national guidelines on directly acting oral anticoagulants (DOACs) for prevention of stroke in atrial fibrillation. Results will be compared with estimates from model regression and approximation.

Transfer learning is an emerging and popular paradigm for utilizing existing knowledge from  previous learning tasks to improve the performance of new ones. In this talk, we will first present transfer learning in the early diagnosis of eye diseases: diabetic retinopathy and retinopathy of prematurity.  We will discuss how this empirical  study leads to the mathematical analysis of the feasibility issue in transfer learning, for which we build for the first time, to the best of our knowledge, a mathematical framework for the general procedure of transfer learning. Within this framework,  we establish  the feasibility of transfer learning by showing its  equivalence to the well-definedness of an associated optimization problem.  

Health care markets have consolidated in recent decades, with increases in both horizontal and vertical ownership ties. We study the implications of shared ownership along both of these dimensions in the U.S. market for outpatient dialysis using a new dataset of mergers, acquisitions, and joint ventures between dialysis chains and local partners such as physicians. We first provide novel evidence of the growth and prevalence of joint ventures in dialysis facilities, which nearly tripled from 9.8% in 2005 to 29.8% in 2017. Using a difference-in-differences framework, we find that joint ventures result in much larger gains in market share compared to acquisitions but relatively similar changes in practices. We also provide evidence that these gains in market share stem largely from business stealing and that patient steering at joint ventures may serve as a barrier to potential entrants. We conclude by connecting these results to technological and analytical innovations within the dialysis industry.

Health literacy is a widely recognized challenge worldwide, with many adults lacking the requisite skills to engage successfully in the management of their health and healthcare. Affecting both individual and societal health outcomes, limited health literacy particularly exacerbates the increasing physical and psychological burden for patients with multiple health conditions as well as pediatric, elderly and disadvantaged populations. Recent developments in digital therapeutic solutions offer an opportunity to apply systems thinking and perspectives combined with artificial intelligence, machine learning and natural language processing methods to synthesize the myriad components of a multi-pronged approach to improving societal health literacy at scale. This talk will highlight some of these developments with a focus on digital platforms and algorithmic artifacts in the healthcare delivery setting, recognizing the challenges of misinformation, disinformation and inclusivity in identifying and disseminating authoritative and accurate content for educating and empowering patients and the public.

The complexity of healthcare delivery frequently requires methodological innovations to methods developed for other settings. The need in healthcare to rapidly model disruptive system changes (e.g., a novel disease) is gaining awareness.  Electronic Medical Records  implementations are increasingly harvested through decision support systems.  Operations Research, Machine Learning, and AI methods are being tailored to better address the unique needs of healthcare.  Supported by practice-based collaborations with several hospitals, we examine advances in personalized bed admissions and placement, including consideration of capacity and other constraints.  Incorporating the prediction and control of unique patient outcomes  following a shock, we present research on adaptive learning and decision making to assign a bed unit type (ICU, PCU, General) with the goal of reducing harmful individual outcomes (i.e., unplanned 30-day hospital readmissions). COVID 19 elevated the need for joint adaptive machine learning and optimization for the allocation of reusable resources in the face of health outcome feedback that is delayed. We present a stratified adaptive learning and optimization approach to balance the management of care under resource constraints in terms of efficiency versus effectiveness. 

In this talk we will discuss how machine learning algorithms and network modeling applied to refined EMR message inbox data could be used to better understand heterogeneity of work composition and team dynamics in primary care clinics, and how they may correlate with physician well-being. The work provide some important insights with respect to the importance of process design and operations of outpatient clinics in mitigating physician burnout.

This is joint work with Celia Escribe and multiple physicians at MGH.

Global health delivery continues to be one of the greatest challenges to our world community, with over 400 million of the world’s population lacking access to basic medical care and 5 billion to safe and affordable surgical care.  Each year 143 million life-enhancing surgeries go unperformed in low and middle income countries due to lack of local healthcare specialists. Yet, recent developments in data science are presenting an unprecedented opportunity to unlock access to critical healthcare by creating efficient markets in healthcare delivery. In this talk Dr. LaRovere will describe how data science is utilized to identify medical deserts and areas of need and the opportunity that data science provides to match the optimal healthcare professionals and resources to those areas of need. 

Healthcare payers often make reimbursement decisions regarding new medical treatments under uncertainty. Conditional approval schemes (e.g., Cancer Drugs Fund, Innovative Medicines Fund) postpone reimbursement decisions until after the collection of post-marketing data to mitigate uncertainty regarding a treatment’s cost-effectiveness. The design of conditional approval schemes has not received much attention in the literature, however. Our game-theoretic model examines when to use a conditional approval scheme, how to design the trial and market access, and how to negotiate reimbursement during and after the post-marketing trial. We find that the interim reimbursement price offered during a conditional approval scheme’s period of post-market data collection can drastically affect equilibrium outcomes. We illustrate the potentially negative impact of policy constraints regarding interim pricing and offer a new risk-sharing mechanism to mitigate those constraints’ potentially adverse consequences.  We also show that, contrary to the common view, price reduction and uncertainty reduction might not be substitutes.

 In recent years, online learning has gained widespread attention in clinical trials and treatment recommendations due to its suitability for handling the bandit feedback nature of treatment outcomes. However, healthcare decision-making problems are more complex than the traditional applications of online learning, such as advertisement recommendations, dialogue response selection, and influence maximization. Two key features set healthcare problems apart. Firstly, doctors cannot engage in random exploration and change prescriptions frequently as this may cause excessive anxiety for patients. Secondly, treatment outcomes may be observed with delayed feedback, as some pharmaceutical ingredients take time to show their effects. In this talk, we will demonstrate how existing online learning algorithms can be modified to account for these unique features. We will also discuss the theoretical results and numerical performance of the proposed algorithms.