Michael Wornow’s Post

EHRSHOT: Stanford’s EHR dataset, model and validation all available now!! Read the full paper linked on #brainxcommunity’s along with more such research, publications & open source datasets: https://brainxai.org

A comprehensive set of clinical use cases are available through EHRSHOT released by Stanford, which includes a foundational model pre-trained on longitudinal EHR records. #healthcarellm #genai

Happy to share our new preprint, "Sample Selection Bias in Machine Learning for Healthcare." Sample Selection Bias (SSB) occurs when the study population is less representative of the target population, leading to biased and potentially harmful decisions for underrepresented subpopulations. In our work, we propose a new approach to address SSB in healthcare. Instead of relying on traditional bias correction techniques, we focus on identifying the target population by modeling the differences between the study and target populations. Our proposed techniques demonstrate robustness across various settings, including different dataset sizes, event rates, and selection rates, outperforming existing bias correction techniques. Read more here: https://lnkd.in/gAnabbBt A big thank you to my collaborators, Lei Clifton, Achille Salaün, Huiqi Yvonne Lu, Kim Branson, Patrick Schwab, Dr Gaurav Bhaskar Nigam and Prof. David Clifton, for their invaluable contributions. #MachineLearning #HealthcareAI #BiasCorrection #DataScience #Research

The main goal of machine learning models in healthcare is to accurately estimate patient risk. For a model to be practical, it should not only have good discriminative performance but also be well-calibrated so that the predicted probabilities are meaningful and interpretable. So model development starts by defining the right probability to be estimated. When making predictions for patients, we often encounter outcomes that depend on other conditions. For example, we may want to predict the quality of life two years after survivorship, but this measurement is conditional on the patient surviving for those two years. Traditionally, patients who did not survive are excluded from the model development, which can introduce bias. In the #BD4QoL project, we developed a methodology based on conformal predictions that includes all patients in the model development process, making the algorithm less biased and more fair. This approach allows us to predict not only the probability of having a poor quality of life in the future but also the probability of surviving, and the joint probability of experiencing any adverse event for all patients. This new method can be valuable for informing patients, clinicians, and hospital management. By knowing which patients will need help to improve their quality of life or require other specialised care, resources can be allocated more efficiently. This result was only made possible by the joint effort of partners who shared their valuable data across borders, including Fondazione IRCCS Istituto Nazionale dei Tumori di Milano, Johannes Gutenberg University Mainz and University of Bristol, for researchers at the University of Oslo and University of Deusto. The peer-reviewed publication will soon be available to the whole community.

Over the last 18 months at Sama Therapeutics, our team has designed, built, and tested medical AI/ML incorporating every single bullet point reviewed here (and more besides) into our codebase — creating an unprecedented, scalable R&D platform product to enhance precision clinical trial analytics and healthcare provider workflows in neuropsychiatry. To control for overfitting, data leakage, and stochasticity, we challenged the multimodality, meta-learning model’s treatment outcome prediction performance across 4 different diagnoses (MDD, ADHD, OCD, and PTSD) and 3 different interventions (TMS, neurofeedback, and SSRIs), achieving between 125-150% nPPV, ie positive predictive power normalized for imbalances in base responder rates between cohorts. We further verified our approach in the most rigorous possible fashion, via 3rd parties who conducted blinded, out-of-sample replications — winning Sama Therapeutics’ “transprognostic” algorithm the top 4 positions on the BrainClinics Foundation’s international leaderboard to date https://lnkd.in/exTw3wQn and publications forthcoming. Feel free to get in touch with us if you are 1) a drug or device developer with preclinical through Phase 1-3 trials, 2) interested in our neurotechnology integrations for research purposes, as our FDA 510k “software as medical device” pathway is pending, or 3) an angel investor/venture capitalist looking for that seed round investment to make back your entire fund in 3-5 years — we’re raising.

Machine Learning Algorithms: Transforming Healthcare and Medicine Machine Learning Algorithms: Transforming Healthcare and Medicine Introduction Machine learning algorithms have revolutionized various industries, and healthcare and medicine are no exceptions. With the ability to analyze vast amounts of data and identify patterns, machine learning algorithms are transforming the way healthcare professionals diagnose diseases, predict outcomes, and develop personalized treatment plans. In this article, […] https://lnkd.in/dsw8C6nR

Machine Learning Algorithms: Transforming Healthcare and Medicine Machine Learning Algorithms: Transforming Healthcare and Medicine Introduction Machine learning algorithms have revolutionized various industries, and healthcare and medicine are no exceptions. With the ability to analyze vast amounts of data and identify patterns, machine learning algorithms are transforming the way healthcare professionals diagnose diseases, predict outcomes, and develop personalized treatment plans. In this article, […] https://lnkd.in/d32ZMYyS

Both the NHS and life science have much to gain from the deployment of machine learning and AI. However, I was reminded of the article published by Young & Steele (https://lnkd.in/ezntV-HF) examining the drop in performance in predictive models trained on healthcare data. Their research showed that performance will drop over time, in common with similar models in use in other fields. This creates both an imperative, but also an opportunity. All organisations who wish to capitalise on the promise of AI will need to focus on ensuring that performance is maintained, tools such as these models cannot be off the shelf tools which are used without retraining. Organisations developing models based on health data will need to be able to revisit new up to date data as time progresses. Having said that, this challenge is also an opportunity as UK health data, along with data sharing and governance structures, are mature and present an opportunity for the public and private sector to innovate. The UK presents an ideal environment to develop these technologies. The balance to be maintained is ensuring public trust while also utilising this valuable resource. 

Machine Learning Algorithms: Transforming Healthcare and Medicine Machine Learning Algorithms: Transforming Healthcare and Medicine Introduction Machine learning algorithms have revolutionized various industries, and healthcare and medicine are no exceptions. With the ability to analyze vast amounts of data and identify patterns, machine learning algorithms are transforming the way healthcare professionals diagnose diseases, predict outcomes, and develop personalized treatment plans. In this article, […] https://lnkd.in/dsw8C6nR