Menu

TMDU-MIT-NUS-ANZICS-JSICM Critical Data Workshops and Datathon 2023

Overview of the hands-on workshops

HWS01

Carrel, Adrien

- Topological deep Learning: l A new direction for artificial intelligence with healthcare applications

click here for further information

HWS01 Carrel, Adrien
Topological deep Learning: l A new direction for artificial intelligence with healthcare applications
Topological Deep Learning and Geometric Deep Learning (TDL/GDL) are emerging fields at the intersection of mathematics, computer science, and artificial intelligence. This workshop aims to introduce participants to the principles and applications of these two fields in the context of healthcare.
First, an introduction to the fundamental concepts will be made. This approach focuses on extending traditional deep learning techniques to handle different data structures such as graphs, meshes, and point clouds. Participants will learn about graph neural networks, higher-order networks, and other architectures. The second part of the workshop will delve into the potential applications these fields in healthcare. The applications include drug discovery, molecule properties prediction and diagnosis of some diseases. Participants will have the opportunity to engage in hands-on activities through a notebook and interactive discussions.
By the end of the workshop, attendees will be equipped with the knowledge to explore these exciting fields further and apply its techniques to tackle real-world healthcare challenges.
Bringing a laptop will be essential for experimenting with the notebook. While some experience in computer science and/or mathematics is beneficial to understand some concepts, the workshop will be designed to be accessible to a wide audience.

HWS02

Ebner, Daniel

- The potentials of generative AI in healthcare

click here for further information

HWS02 Ebner, Daniel
The potentials of generative AI in healthcare
An example abstract: In this workshop, we will explore the transformative potential of Generative AI within the context of the Intensive Care Unit (ICU). Participants will delve into the practical applications of Generative AI models, such as GPT-4, to simulate patient health scenarios, predict health outcomes, and generate intervention strategies. We will also cover the ethical implications, data privacy concerns, and decision-making challenges presented by the use of these technologies in the ICU environment. With a balanced blend of theoretical discussions and hands-on exercises, our goal is to inspire innovative approaches for integrating AI into healthcare, ultimately improving ICU patient care and outcomes.
———————
Of note, ChatGPT is able to process Japanese text, but it is not quite as strong at Japanese as it is in English. The workshop can be either English or Japanese, but if a member of your team or one of the AI members in Japan is familiar with a large-language model (LLM) like ChatGPT that focuses specifically on Japan or Japanese healthcare, that could be helpful.
If one does not exist today, it is possible that one will be created between now and the datathon, so it might be something that we have to keep our eyes out for.

HWS03

Aoki, Tomonoshin

- Healthcare digital transformation (DX): International comparative analysis of health systems and cases, accompanied by hands-on research demonstrations <Japanese session>
医療分野におけるデジタルトランスフォーメーション(DX):医療システムと事例の国際比較分析及びハンズオンのリサーチ・デモンストレーション"<日本語開催>

click here for further information

HWS03 Aoki, Tomonoshin
Healthcare digital transformation (DX): International comparative analysis of health systems and cases, accompanied by hands-on research demonstrations <Japanese session>
Healthcare digital transformation, or DX, is an emerging global phenomenon with profound implications not just for individual medical practices, but for entire healthcare systems. This transformative process, however, is heavily influenced by the unique structure and policies of each country's healthcare system.
In this interactive workshop, we will delve into the role of DX within various healthcare systems across the globe, with a focus on countries such as the UK, US, and Japan. Drawing on cases, we will analyze how data sharing has been implemented in these different contexts.
This workshop will not only deepen participants' understanding of these processes, but also provide insights into empirical research on institutional design related to DX. Through practical demonstrations and discussion, participants are expected to gain a better understanding of the policy perspectives on DX, its potential, and the nature and challenges of policy research.
This workshop is ideal for policy researchers, healthcare professionals, data scientists, and anyone interested in the intersection of healthcare and technology.
Prerequisites: All necessary materials and code will be provided. While foundational knowledge related to healthcare systems, policy-oriented empirical research, and programming (ideally in R) will amplify your learning experience, the workshop is designed to be comprehensive and beneficial even without prior knowledge in these areas.

HWS04

Kinoshita, Takahiro

- What do you mean by “adjusting for” confounders? <Japanese session>
交絡を”調整する”ってどういう意味ですか?<日本語開催>

click here for further information

HWS04 Kinoshita, Takahiro
What do you mean by “adjusting for” confounders? <Japanese session>
Adjusting for confounders is a pivotal concept in causal inference with real-world data. The first crucial step involves identifying the essential variables required for a balanced and fair comparison-referred to as confounders to achieve conditional exchangeability in the formal language of epidemiology. Numerous causal methods have been proposed to “adjust for” these confounders, including multivariable linear or logistic regression using confounders as covariates, propensity score analyses such as covariate adjustment, stratification, and matching, inverse-probability of treatment weighting, and the g-formula. However, the critical distinctions in underlying assumptions as well as result interpretations are frequently overlooked.
In this workshop, our objective is to define the necessary assumptions for obtaining unbiased effect estimates from non-randomized observational data. Furthermore, we will explore the disparity between “conditional effects” and “marginal effects” and introduce reliable methods that address your causal questions.
We expect all participants to have a basic understanding of R language (not GUIs such as EZR or R commander) or Python. The hands-on session will be conducted using RStudio.

HWS05

Minegishi, Yu/Hase, Takeshi/Uchimido, Ryo/Yamada, Tomoaki

- Learning Python with hands-on experience <Japanese session>
ハンズオンで学ぶPython入門<日本語開催>

click here for further information

HWS05 Minegishi, Yu/Hase, Takeshi/Uchimido, Ryo/Yamada, Tomoaki
Learning Python with hands-on experience <Japanese session>
This workshop is designed to provide a foundation in Python programming. Participants will learn Python syntax, data types, control structures, functions, and modules through practical exercises. The main objective is to grasp fundamental programming concepts and acquire the skills to create simple programs using Python. The workshop follows a hands-on approach, with instructors providing explanations and participants engaging in exercises and tasks. It is suitable for programming beginners and those new to Python, and attendees are required to bring their own laptops. By participating in this Python Introductory Hands-on, participants will gain proficiency in programming basics and acquire the ability to develop basic programs using Python.

HWS06

Shimizu, Sayuri

- Analyzing administrative data (DPC data) of acute care hospitals in Japan  <Japanese session>
日本の急性期医療機関の管理データ(DPCデータ)を用いた分析演習<日本語開催>

click here for further information

HWS06 Shimizu, Sayuri
Analyzing administrative data (DPC data) of acute care hospitals in Japan  <Japanese session>
Japanese acute care hospitals submit data on all discharged patients (DPC data) to the Ministry of Health, Labour and Welfare (MHLW), and a database covering almost all Japanese in acute care has been established. DPC data, which includes inpatient medical record information such as age, gender, disease information, and severity of illness, as well as information on all medical procedures received during hospitalization, is used for clinical epidemiological studies, hospital management, health care quality indicators, and health care policy. In this hands-on seminar, participants will learn about the characteristics of Japanese medical data through analysis of sample data from individual DPC forms and open data.

HWS07

Feng, Mornin/Huang, Ling

- Trust deep learning model in healthcare: From accuracy to reliability

click here for further information

Feng, Mornin/Huang, Ling
Trust deep learning model in healthcare: From accuracy to reliability
As AI continues to shape various sectors, including healthcare, establishing robust ethics protocols becomes crucial to fully harness its potential benefits. This workshop focuses on a fundamental ethical consideration in healthcare: reliability. Ensuring the reliability of AI models in healthcare is vital to instill trust among clinicians, patients, and regulatory authorities.
In this workshop, we aim to discuss critical challenges and emerging solutions in developing and implementing trustworthy AI models within healthcare settings. Key topics include defining reliability, strategies for developing reliable AI healthcare models, and techniques for validating and evaluating reliability performance. Real-world case studies will illustrate the entire process from data collection to clinical application, providing valuable insights. We invite researchers, clinicians, data scientists, industry professionals, and policymakers to join this workshop and contribute to the advancement of reliable AI models in healthcare.

HWS08

Kimes, Patric/Nakajima, Lui/Motos, Joao

- AI for real world clinical applications: Considerations and pitfalls

click here for further information

HWS08 Kimes, Patric/Nakajima, Lui/Motos, Joao
AI for real world clinical applications: Considerations and pitfalls
Over the past two decades, exponential growth in data availability, computational power, and newly available modeling techniques has led to an expansion in interest, investment, and research in AI applications for medicine and clinical practice. However, before AI can be widely deployed in clinical settings, further thought must be given to acknowledge the several pitfalls within the AI lifecycle.
This workshop will guide participants through examples of AI in real-world applications. We will explore topics including the use of AI in drug-diagnostic co-development and the seven-step AI lifecycle, using case studies and examples from oncology and ophthalmology. We will explore challenges, biases, and pitfalls that can arise during this process and discuss possible solutions. No prior experience is necessary; all backgrounds are welcome.

HWS09

Tsuji, Shingo

- Ask anything about practical data Science--- Know your present situation, set your own goal, clarify the process --- <Japanese session>
データサイエンスの水先案内---現状を知り、ゴールを定め、プロセスを明確にしよう---<日本語開催>

click here for further information

HWS09 Tsuji, Shingo
Ask anything about practical data Science--- Know your present situation, set your own goal, clarify the process --- <Japanese session>
Data science is one of the most important components for life science research. However, data science itself consists of a bunch of knowledge such as computer science, machine learning, mathematics etc., so the researchers without a computer science background often lose their way to go. In this workshop, I will listen to your any kind of issues about data science and try to solve the problems through intensive discussion. You don't need any preparation except your research goals or current stuck point.

HWS10

Tagawa, Koshiro/Tohyama, Takeshi/Soko Setoguchi

- Data analysis using propensity score matching <Japanese session>
傾向スコアマッチングを用いたデータ分析<日本語開催>

click here for further information

HWS10 Tagawa, Koshiro/Tohyama, Takeshi/Soko Setoguchi
Data analysis using propensity score matching <Japanese session>
In recent years, the use of propensity score matching in analyses has become more prevalent. In this course, the participants will learn the basics of propensity score matching and have hands-on practice.
This workshop is designed to be accessible to beginners. Even if you have no previous analytical experience with statistical software, we encourage you to attend if you are interested.
We will use SAS software for analysis. Please prepare to install the commercial version of SAS software or register to access SAS OnDemand via the cloud free of charge (https://welcome.oda.sas.com/).

HWS11

Ichihara, Nao

- ICU outcome assessment using matching methods <Japanese session>
マッチングに基づくICUアウトカム評価<日本語開催>

click here for further information

HWS11 Ichihara, Nao
ICU outcome assessment using matching methods <Japanese session>
For clinical registries to be of value for improving quality of care, it is important to define an outcome measure that appropriately addresses differences in patient risks across populations.
The O/E ratio (observed/expected ratio) is used for such purposes.  One can obtain a prediction formula for patient death using the entire ICU registry data, apply this formula to predict risks of death for each patient in a specific ICU, sum them up to estimate number of expected deaths at the ICU, and the ratio of the number of observed deaths to that of expected deaths can be used as the O/E ratio.  However, even if ICU-A has a higher O/E ratio than ICU-B, it doesn’t mean ICU-A has a lower quality of care than ICU-B (Simpson’s paradox).  The O/E ratio is imperfect as an outcome measure that addresses differences in patient risk across ICUs (risk-adjusted outcome measure).
Alternatively, evaluating outcome of care at an ICU can be understood as causal inference in observational study, i.e., a measure that represents “average effect” of “treatment at the ICU” on “the patients’ risk of death.” As such, a meaningful outcome measure can be defined using matching method, an analytical approach widely used in observational medical research.
This workshop expects participants to be familiar with R. Matching-based outcome measures are defined, measured, and summarized using JIPAD data and synthetic data.  In addition, O/E ratios will be calculated, and compared with matching-based outcome measures.