Postdoc in causal inference for medical AI

Spurred by the advancements in Artificial Intelligence (AI), a wave of prediction models is reaching healthcare, with some already employed in Dutch hospitals. Many such models are used at Amsterdam University Medical Center. Research and regulation on safety and robustness have largely focused on proving that an AI application is reliable before its release. However, recent scandals – especially in the healthcare domain, but not limited to it – have revealed that many AI applications fail to perform as intended after implementation in a clinical context for decision support. Notable failures include an algorithm used to prioritize patients needing medical attention in the US, and the QRisk algorithm used for cardiovascular risk management. Analysis of these failures has top priority, since they may directly lead to more adverse outcomes.

A growing community of researchers advocates for a causal inference perspective on prediction modeling for decision support. Some recent work has shown that the implementation of prediction models can be damaging for downstream patient outcomes even if model performance is outstanding. A start has been made to inventorize the blind spot of a non-causal approach to prediction modeling. This research line has implications for the whole field of predictive modeling, as evidenced by a response to the TRIPOD+AI guidelines on reporting of clinical prediction models. As a consequence, more work is needed to analyze the potential risks associated with the use of prediction models for decision support, to develop metrics that actually track what we want, and use these theoretical tools to review the prediction models currently used at Amsterdam UMC.

The project linked to this vacancy will take place in the Methodology group of the Medical Informatics department under the supervision of Dr. Giovanni Cinà, for a duration of two years.

We are looking for a postdoc who is eager to contribute to the safer implementation of machine learning (ML) methods in healthcare settings. You will investigate how causal inference techniques can be leveraged to enhance the monitoring of deployed medical AI applications. This exploration will address a number of methodological topics intersecting causal inference, such as distribution shift and interpretability.

You will provide efficient and scalable implementations of your methods and will integrate them with popular open-source systems. Via your affiliation with the Medical informatics department, you will have access to several large datasets comprising tens of thousands of patient records.

Your focus will be on research, with a potential involvement in teaching such as student thesis supervision. As part of a training for senior roles, you will help out in organizational and managerial roles such as co-supervision of PhDs or organization of seminars.

As a postdoc, the following is required:

• You hold a PhD in computer science, epidemiology, econometrics, machine learning, artificial intelligence, mathematics, data science, medical informatics, or a related field.
• You have experience in causal inference, in the form of projects and-publications.
• You have hands-on experience in machine learning and related libraries (e.g., PyTorch, Sklearn) and strong programming skills in Python or R.
• You have a creative and independent mindset and work well in multidisciplinary environments.
• You are fluent in English.

• A flying start to your career in scientific research.
• Plenty of room for your drive to shape tomorrow's healthcare.
• Working on large-scale and in-house research, with motivated colleagues from all over the world.
• You will be employed by Amsterdam UMC Research BV.
• A contract for 24 months (extension possible, subject to funding).
• Salary scale 10: € 3.493 to € 5.504 gross based on full-time employment (depending on education and experience) and a year-end bonus of 8.3%. Calculate your net salary here.
• Holiday hours: 190,4 per year for fulltime and a possibility to save additional hours.
• Pension accrual with BeFrank, a modern, comprehensible and fairly priced pension.
• For >7 km each way, 100% reimbursement for public transport travel costs and, for private transport, €0.18 per km up to a maximum of 40 km each way.
• Do you prefer walking or cycling? Take advantage of our good bike scheme. Moreover, you will receive a reimbursement of €0.18 per km.

Watch this video with more information about joining Amsterdam UMC Research BV.

Amsterdam UMC Research BV supports non-profit scientific research. In doing so, we provide researchers with everything they need to excel. Our principal investigators (PIs) and project leaders offer support in the field of project management, finance and human resources. In medical scientific research projects, legal support is also provided.

Watch the video to find out more.

You will be integrated in an ongoing collaboration with industrial partners, with concrete possibilities for your research to influence the development of existing medical AI products. You will be embedded in the Methodology group of the Medical informatics department, a vibrant and diverse team of >15 researchers dedicated to the study and implementation of AI solutions for healthcare problems.

Moreover, you will be appointed at the Medical informatics department of Amsterdam UMC at the University of Amsterdam. You will join the team of dr. Giovanni Cinà, within the Methodology group lead by prof. dr. Ameen Abu-Hanna.

Dr. Giovanni Cinà is assistant professor in responsible medical AI and works on the reliability and robustness of medical AI applications, and specifically on topics such as OOD detection, explainable AI and causal inference. He holds a joint position with the Institute for Logic, Language and Computation in the Faculty of Science. Prof. dr. Abu-Hanna is a principal investigator in Methodology in Medical informatics and has extensive experience in AI, ML, and prognostic modelling and evaluation.

Our team has on-going collaborations with colleagues at the IvI-UvA, the ILLC-UvA and the Amsterdam Business School, and you are encouraged to actively participate in these collaborations.

During the publication period, applications will be handled continuously. If the vacancy is filled, it will be closed prematurely.

Do you have any questions or do you require additional information about the position? Please contact dr. Giovanni Cinà, Assistant Professor of Artificial Intelligence, via g.cina@amsterdamumc.nl.

For more information about the application procedure, please contact Rhiannon Sandfort, Recruitment advisor, via r.e.sandfort@amsterdamumc.nl.

Note: applications should include the following information (all files besides your CV should be submitted in one single pdf file):

• A detailed CV referring to your education and work experience.
• A letter of motivation.
• A list of publications.
• Examples of code written by the applicant (e.g., via a link to a github repository).

Only complete applications received within the response period will be considered.

A reference check, screening and hiring test may be part of the procedure. Read here whether that applies to you. If you join us, we ask you for a VOG (Certificate of Good Conduct). Internal candidates will be given priority over external candidates in case of equal suitability. Acquisition in response to this vacancy is not appreciated.