Decision Making in Health and Medical Care

There are an estimated 7,000 different rare diseases, most of which are single gene disorders potentially amenable to gene-targeted therapies. Given that individual rare diseases affect small numbers of patients – a few hundred to a few thousands each (sometimes fewer) – and that less than 5% of rare diseases have an approved therapy, new strategies are needed to accelerate rare diseases therapeutics development. Many diseases at a time approaches, platforms, and shared molecular etiology targeting are currently being piloted for rare genetic disorders, and hold promise for accelerating therapeutics development for many more rare diseases.

Pitt Hopkins Syndrome is a serious lifelong neurological disorder that is caused by random mutations in a gene and transcription factor called TCF4 on the 18th Chromosome. Symptoms typically appear when the first milestones are missed around 6 months. Many children with Pitt Hopkins are unable to speak, walk, or purposefully use their hands. Breathing problems, feeding tubes, seizures, anxiety, gastrointestinal, and orthopedic issues are common. Audrey Davidow is the President of the Pitt Hopkins Research Foundation and has spent the last 9 years fighting for funds, science, awareness, treatment and hopefully a cure for her son’s debilitating disease. Here she shares her journey—the highs, the lows, the successes and failures along the way.

Presented by CEO, Neil Kumar: How BridgeBio built a biopharmaceutical company designed to discover, create, test and deliver transformative medicines to treat patients who suffer from genetic diseases and cancers with clear genetic drivers by using innovative financial engineering solutions to fund biomedical breakthroughs.

Rare disorders are estimated to afflict more than 25 million Americans. US Congress passed the Orphan Drug Act of 1983 to provide financial incentives and accelerated FDA regulatory pathways to incentivize companies to develop much needed therapies to treat rare disorders. However, among the more than 7,000 rare disorders, majority of the disorders afflict fewer than 1,000 patients in the US. With “once-a-lifetime” treatment modalities such as gene therapy becoming the mainstay of rare disease treatments, a roadmap for innovative private-public partnerships to finance drug development will be helpful for underserved patient populations afflicted by ultra-rare diseases.

Dr. Irony will present on the topic of FDA engagement in assisting the development of treatments for rare diseases, through the use of various regulatory authorities available.

Global initiatives that independently address Infectious Diseases (IDs) and non-communicable diseases (NCDs) miss the opportunity to enhance capabilities for both missions. The synergies extend across the spectrum from the molecular etiology to the macro-level capacity-building (space, staff and stuff) for the provision of healthcare. In addition to a biomanufacturing industrial base for vaccines and other medical countermeasures, critical infrastructure for protection from epidemics such as Ebola and pandemics, such as COVID-19, includes a sustainable healthcare system to meet sudden surge capacity. An approach that provides an integrated health system using the flex-competence model supports routine care for cancer and other NCDs, while enabling rapid adaptation for response to emerging infectious diseases.

This presentation introduces a framework for prioritizing investments in emerging infectious disease (EID) vaccine development within a newly established, international multi-stakeholder setting. Drawing from established approaches in the fields of Health Research Priority Setting, Decision Analysis and Operations Research, a conceptual frame is proposed for addressing interconnected problems of strategic objective setting, investment boundary setting, project and portfolio selection. Specific prioritization models are then presented to illustrate how solutions can be supported to each of these problems in a real-life setting, including multi-criteria decision analysis, Monte Carlo simulation and simulation-optimization methods.

This presentation will cover the rapid flow of information, decision-making, and risk management that enabled the development of the COVID-19 vaccine.

The year-long special issue of the Harvard Data Science Review provides a glimpse into the massive stress tests COVID-19 has created for many ecosystems in our global communities. This talk samples the challenges and opportunities generated by such tests that can directly affect our ability to make hard and quality decisions in medical and health care. Questions addressed include: -Did individual acceptance of the use of personal health data for public benefit change during COVID-19? (Frederic Gerdon, Helen Nissenbaum, Ruben L. Bach, Frauke Kreuter, and Stefan Zins) -Did COVID-19 change the level of public support to health care reform in the United States? (John Sides, Chris Tausanovitch, and Lynn Vavreck) -How can we ensure responsible data science and AI innovations for combating COVID-19 and pandemics in general? (David Leslie) -How reliable are our estimates of case fatality rates (and other rates) for COVID-19? (Anastasios Nikolas Angelopoulos, Reese Pathak, Rohit Varma, and Michael I. Jordan) The audience is also invited to examine if these studies themselves pass the stress test in the sense of maintaining high standards under stringent time constraints.

Medicine is, in its essence, decision-making under uncertainty. Decisions about the tests to be performed and interventions, pharmacological or surgical, to be administered. These decisions have to be made probabilistically due to uncertainty resulting from the complexity of the human body, a complex system made of a vast number of elements intensely interacting with each other in a non-linear manner. The human body’s complex structure is responsible for millions of unique combinations of risk factors and protective characteristics that individually define our health and disease. It also leads to redundancies within the human body, which result in our robustness to disease and rarity of the essential severe adverse outcomes, such as death. Unfortunately, rare events are difficult to predict and prevent. However, the computation is very well suited to modeling both vast numbers of unique combinations of risk factors and protective characteristics, the individualization of medicine, and the modeling of rare events. These advantages of computational medicine promise to simultaneously improve health outcomes and lower healthcare costs through computationally driven individualization of medicine.

Modeling and incorporating idiosyncratic preferences of patients (risk aversion, bounded rationality, asymmetric aversion to information, risk communication, rational inattention, etc.) is ubiquitous in building multi-period/dynamic decision-making models for personalized medical and health care. Additional difficulties for the construction of such models come from the interaction of uncertainty and time, as updated information becomes progressively available over the course of a treatment, recovery, pregnancy, etc.. In my talk, I will introduce representative decision-making models for cases arising in OB-GYN, present the related optimization problems and discuss the associated conceptual, methodological and computational difficulties in solving them.

In this talk I will present research on the efficient computation of EVPPI (Expected Value of Partial Perfect Information) and EVSI (Expected Value of Sample Information). I will start by explaining how the need for these arises from decision making under uncertainty, for example in deciding whether to fund medical research to reduce the uncertainty in identifying the best treatment for a particular condition. I will then outline our development of a Multilevel Monte Carlo method for the resulting nested expectation problems. This is joint work with Dr Howard Thom of the University of Bristol, and Dr Takashi Goda of the University of Tokyo. M.B. Giles, T. Goda. “Decision-making under uncertainty: using MLMC for efficient estimation of EVPPI”. Statistics and Computing, 29(4):739-751, 2019. T. Hironaka, M.B. Giles, T. Goda, H. Thom. “Multilevel Monte Carlo estimation of the expected value of sample information”. SIAM/ASA Journal on Uncertainty Quantification, 8(3):1236-1259, 2020.

Over the years we have learned to make consequential decisions from some of the world’s most complex computer simulations. The approaches are far from turn-key and bracketing confidence requires people to intervene and evaluate throughout. We have worked to co-develop these tools and methods together with the problems. That is, one cannot think of uncertainty quantification as an afterthought. In medically assisted decision making, as well as decision making in general from increasingly rich and complex data, there are yet no parallel methods that can provide quantifiable confidence in predictions. In this talk I will discuss how we can learn from where methods have worked to get us to a future where we can make decisions with an understanding of how certain we are.

Academic cancer centers are a relatively new invention, closely tied historically to the National Cancer Act of 1971. They were designed to marry optimal in-person care of diagnosed patients with laboratory science intended to advance cancer management. In-person clinical trials grew in importance as the field matured, some asking purely scientific questions but most, especially lately, including government supported ones, involving products in commercial development by industry. The design of cancer centers has thus always evolved to meet the needs and expectations of the public, and to take advantage of novel opportunities. However, major changes in technology and society are now accelerating the process in meaningful ways. Some of these changes involve an emphasis on prevention and diagnosis over just treatment, a shift toward expansion of activities beyond physical contact as by remote access to trials of investigational medicines, the use of telemedicine in both research and care, the establishment of broad based biobanks, the emergence of massive collaborative efforts in both data acquisition and analysis, and the development of knowledge bases to guide clinical decisions using molecular information. What remains to be better addressed is an improved relationship between the centers and industry. A more efficient mechanism for commercially-minded very early development of discoveries and concepts derived from center research, now constrained by limitations in both financial and human resources, would benefit all parties, including the public. Ideally, all these evolutionary advances will not only improve the quality of care, research, education, and but also promote health care equity in diverse populations.

A bandit problem is classic in sequential decision making. The explicit goal in the context of medicine is to treat patients as effectively as possible, those in a clinical trial and those who come after the trial. The latter patients benefit from the results of the trial but so do the former patients. The optimal decision (clinical trial design) balances the conflicting desiderata of treating the current patient effectively (earning) while simultaneously getting information for treating subsequent patients (learning). I will describe the introduction of bandit strategies and their modifications into medicine, via what I call platform clinical trials.