NB: this series is still a work in progress.

Healthcare AI Development

Welcome to the second post in our series on the healthcare AI lifecycle. To start at the beginning, go to the overview post on the healthcare AI lifecycle. Having established a general framework for the healthcare AI lifecycle, it’s time to cover the specifics. In the absence of a better starting point1, this post focuses on what I perceive to be the “beginning” of the AI lifecycle: the development phase.

Development encompasses the various processes involved in creating an AI model. This phase is foundational, as the quality and success of the AI system largely depend on how well it is developed. Each step is crucial for ensuring that the AI tool will be effective and reliable in real-world clinical settings, from selecting the right task to accessing and preparing data, training the model, and validating its performance.

By the end of this post, you’ll have a comprehensive understanding of the critical steps in developing healthcare AI models and the challenges and considerations of each step.

Healthcare AI development portion of the lifecycle. Development is the creation of models and involves predictive task selection, data access, data preparation, model training, and model validation.
Healthcare AI Development Portion of the Lifecycle. The development phase encompasses the creation of AI models, beginning with task selection, followed by data access, data preparation, model training, and culminating in model validation. This phase is crucial for building robust and effective AI tools that will be used in clinical care.

Development Steps

The development phase of healthcare AI is a multifaceted process that starts with a target task and ends with a (hopefully) robust and effective AI model. To provide a clear structure, I break down this phase into five discrete steps:

  • Task Selection
  • Data Access
  • Data Preparation
  • Model Training
  • Model Validation

As depicted in the figure above, it’s easiest to illustrate these steps as discrete and chronological. However, this linear representation is disingenuous and doesn’t fully capture the reality of the development process. These steps are semi-continuous and often non-linear. Model developers frequently jump back and forth between these steps or work on them concurrently. Despite this fluidity, these steps are generally present in all model development projects and tend to be finalized in the order presented.

This breakdown reflects my approach to structuring the development phase, providing a framework to understand and navigate the complexities. Each step is crucial for building robust and effective AI models that will be used in clinical care. By understanding and addressing the nuances of these steps, we can ensure that the models developed are technically sound, clinically relevant, and reliable.

We will now delve into a brief discussion of each development step, exploring their importance and associated challenges.

Task Selection

Choosing the right problem for AI to tackle is crucial. The journey begins with identifying the specific task or clinical problem the AI model aims to address. This step involves collaboration between clinicians and data scientists to ensure the model’s relevance and potential impact. It’s not just about finding a gap; it’s about ensuring the AI solution can actually improve outcomes or efficiency in a meaningful way. We’re looking for problems where AI can provide insights or automation that weren’t feasible before.

Engaging in thorough discussions with clinicians is essential to pinpoint where they feel the most pressure and where they think AI could be beneficial. Their firsthand experience and insights are invaluable in identifying tasks that truly matter.

Caution should always be exercised when someone says, “I just want an AI to predict/do X.” There may be deeper or related problems that should be surfaced before jumping directly in the initial direction. A great approach for overcoming this issue is to ask a series of probing questions. Some of my favorite lines of inquiry are:

  • Sequential Why?: Asking why or how repeatedly is often a super fast way of understanding the existing problem or system. This iterative questioning can uncover underlying issues that might not be immediately apparent.
  • Would magic help?: Asking how a “perfect solution” would help (e.g., “If I could give you Y information with 100% accuracy, how would that help?”) gives you a sense of the maximum possible benefit of a solution. This helps in understanding the potential impact and feasibility of the AI model.
  • Do you have data?: If the answer is no, you should think long and hard about whether this project is truly feasible. Data availability is a fundamental prerequisite for any AI development, and its absence can significantly hinder progress.

In addition to these considerations, it’s important to be mindful of potential biases in task selection. For example, if we choose a task such as predicting no-shows, we must recognize that this could be problematic due to inherent systemic biases. Structural issues often prevent certain subpopulations from having consistent access to healthcare, and building a model for this task might inadvertently propagate these biases. Instead of developing an AI model for predicting no-shows, it might be more beneficial to investigate other ways to address the root causes, such as creating programs to improve access to healthcare. In this case the best AI model may be no AI model at all.

By carefully selecting the right task and thoroughly understanding the problem, we set a solid foundation for the subsequent steps in the AI development lifecycle. This ensures that the AI model developed is not only technically sound but also highly relevant and impactful in real-world clinical settings.

Data Access

Getting the right data is often the first big hurdle of AI development. The data needs to be comprehensive, clean(able), and relevant. This step involves negotiating access to medical records while ensuring patient privacy and data security. Its imperative that you consider the following:

  • Provenance: where is the data coming from? Who’s going to get it for you?
  • Protection: how are you going to ensure that the data are properly potected? I recommend working directly in hospital IT systems or working with them to spec out compliant environments.
  • Prospective use: will you have this data available when you are trying to use this system prospectively or in the real-world?

Data Preparation

Having obtained data, model developers may realize that healthcare data, like healthcare itself, is complicated. Processing and transforming data for AI model development requires a unique mix of clinical and technical expertise. Preparing this data for AI involves cleaning it, dealing with missing values, and transforming it into a format that algorithms can work with.

This step is usually pretty labor intensive, my estimate is that 90% of the engineering time will be dedicated to data preparation work. It can be helpful to use tools to help automate the data preparation. I made a tool called TemporalTransformer that can help you quickly convert EMR or claims data into a format ready for processing with neural networks/foundation models. I discuss it in the supplemental of my paper on predicting return to work and you can find code here.

Model Training

Training the model is where the may be the most exciting step for the technical members of the project. But its often one of the shortest parts of the project (in terms of wall-time, not CPU/GPU time). We select algorithms, tune parameters, and iteratively improve the model based on its performance. This step is a mix of science, art, and a bit of luck. The goal is to develop a model that’s both performant and generalizable.

Model Validation

After being developed, models must be validated to assess if they will benefit patients, physicians, or healthcare systems. Validation means testing the model on new, unseen data to ensure it performs well in settings representative of intended real-world usage. Ultimately, it’s about making sure the model isn’t just memorizing the data it’s seen but can actually make good predictions on new data.

This step often involves internal and external validation to ensure robustness. There are varying definitions for internal and external validation, but the distinction I like to use is based on the system generating the underlying data. If the data comes from the same system (e.g., same hospital, just a different timespan) then I would consider it internal validation data. A well conducted external validation is a great way to assess if a model will work in a given environment. However, external validation may be challenging due to data-sharing restrictions. Despite this challenge, its often a great place to get started in engaging with healthcare AI system, especially for physicians. Here are some examples of external validation studies that I’ve worked on:

Wrapping Up

We’ve taken a closer look at the development phase of healthcare AI, covering everything from task selection to model validation. Each step is filled with unique challenges and requiring a blend of clinical insight, data science expertise, and ethical considerations. It’s worth noting that while we’ve covered a lot of ground here, each of these development steps could easily merit its own detailed post. The discussions here have been intentionally brief to provide an overview and establish a foundation. In future posts, I may delve deeper into each aspect, unpacking the complexities and sharing insights on how to navigate these crucial stages in the lifecycle of healthcare AI development.

The journey of developing healthcare AI is a continuous one, with each step uncovering new data, insights, and evolving clinical needs. As we advance, our goal remains clear: to harness the potential of AI in improving patient care, enhancing healthcare operations, and advancing the practice of medicine.

Thank you for joining me on this exploration of healthcare AI development. Stay tuned for more detailed discussions on each step and other facets of the healthcare AI lifecycle in upcoming posts. Until next time, keep pushing the boundaries of what’s possible in healthcare with AI.

Go ÖN Home

Some of this content was adapted from the introductory chapter of my doctoral thesis, Machine Learning for Healthcare: Model Development and Implementation in Longitudinal Settings.

  1. “If you wish to make an apple pie from scratch, you must first invent the universe.” - Carl Sagan