PRATIK SHAH

Emergent neural networks to redesign clinical testing of medicines and learning from real world evidence

Real World Data (RWD) and Real World Evidence (RWE) are playing an increasing role in healthcare decisions to support innovative use of Electronic Health Records and other digital sources to help patients and physicians with experimental and existing treatments [R1]. For example: phase 3 clinical trials evaluating new therapies and vaccines are among the most complex experiments performed in medicine. Around 50% of phase 3 trials fail and in some cases lead to adverse events [R2]. This high failure rate also counteracts patients consenting to treatments by experimental drugs as a last resort. The United States Food and Drug Administration (FDA) states that another significant challenge is the difficulty of predicting clinical results in a wider patient base in the real world vs. controlled randomized trials [R1]. In collaboration and with guidance from the United States FDA, we invent explainable AI and machine learning systems that learn from diverse and inclusive datasets for ethical clinical use and testing of experimental medicines [R3].

  • For example, we have described novel and non-trivial reward functions for self-learning Reinforcement Learning (RL) algorithms for dose de-escalation studies during clinical trials to alleviate chemotherapy toxicity. These algorithms learn reward contribution from physician actions and patient states/health without future survival/outcomes information to solve fundamental problems in clinical development of medicines [1].
  • Recently, MIT and the United States FDA signed a Memorandum of Understanding “Health 0.0” to engender AI and ML research for computational medicine and clinical development and accompanying regulatory framework to improve health outcomes for patients. We focus on three key themes for development of next-generation medicines by adoption of digital evidence generated by AI and ML: (1) validation and modernizing the clinical trials process, (2) strategies for rational use of AI- and ML-driven learning from real-world data and evidence and, (3) required regulatory oversight for integration, explanation, and de-risking of AI/ML digital analytics in medical care to patients [2, R3].
  • In collaboration with regulatory agencies and clinical partners, we are initiating a regulatory path for AI and ML software as a medical device and digital medicines developed in our laboratory from this research [2, 4, 6]. Our research classifies, predicts and enriches novel digital endpoints to benefit patient health, eliminate adverse events, and improve outcomes. This work has significant impact on the ethical decisions facing patients and their families, and regulatory decisions for US FDA and European Medical Agency [2, R1].
  • In another effort, we are leveraging the use of observational RWD and RWE to supplement randomized data to train neural networks cognizant of causal inference and clinical trial design [3, 5].

Peer-Reviewed Publications

  1. Reinforcement learning with action-derived rewards for chemotherapy and clinical trial dosing regimen selection. 2018. [Abstract] [Full Paper]

    Gregory Yauney, Shah P*

    (*Senior author supervising research)

    Proceedings of the 3rd Machine Learning for Healthcare Conference. PMLR 85:161-226

  2. Artificial intelligence and machine learning in clinical development: a translational perspective. 2019. [Abstract] [Full Paper]

    Pratik Shah* et al.

    (*Corresponding author)

    Nature Digital Medicine. DOI: 10.1038/s41746-019-0148-3

  3. Artificial Intelligence for clinical trial design. 2019. [Abstract] [Full Paper]

    Stefan Harrer, Shah P, Antony P, Hu J

    Trends in Pharmacological Sciences. PMID: 31326235

  4. Improving cancer diagnosis and care: clinical application of computational methods in precision oncology. 2019. [Abstract] [Full Paper]

    Pratik Shah

    Proceedings of a Workshop, National Academies of Sciences, Engineering, and Medicine. Washington, DC: The National Academies Press. PMID: 31386317

  5. Artificial neural networks detect and classify hematologic malignancies or solid tumors using electronic medical records prior to clinical diagnosis. 2020.

    Yujia Zhou, Barnes C, Magoc T, Lipori G, Shah P, Cogle CR

    Published at 61st American Society of Hematology Annual Meeting, Orlando, FL

  6. Data science in clinical pharmacology and drug development for improving health outcomes in patients. 2020. [Abstract] [Full Paper]

    Pratik Shah, Peck R, Vamvakas S, van der Graaf, Piet H

    Clinical Pharmacology and Therapeutics. PMID: 32202650

AI for drug discovery sketch

Resources

  • R1 - Framework for FDA’s Real-World Evidence Program [PDF]
  • R2 - 22 Case Studies Where Phase 2 and Phase 3 Trials had Divergent Results [PDF]
  • R3 - FDA MOU [Link]

Select Talks

Press

Honors

Other MIT Research Areas