Where to Start with AI in Healthcare
Healthcare organizations can take the following approach when looking to incorporate AI in their workflows:
- Start small. Address a real need in the organization rather than coveting a bright, shiny object. If it’s used for a specific case, there’s a higher likelihood of adoption. And if it’s something that meets a budgeted need and has a calculable ROI, that’s even better because it increases the chance of it getting funded.
- Think platforms rather than point solutions. There are AI solutions that can do one thing very well. But if organizations want to create a foundation for more AI adoption in the future, they’ll want to adopt a platform approach.
Currently, a number of healthcare organizations use AI-powered solutions for administrative tasks and some clinical decision support. Clinicians who see their administrative burdens reduced can reclaim time to focus on their patients and work at the top of their licenses.
Challenges Ahead for AI in Healthcare
For an AI solution to deliver results, it requires large amounts of high-quality, trusted data. As a prerequisite for any AI adoption, organizations will need to have a well-crafted data strategy to ensure that a steady supply of data is available for current and future AI tools. If an organization’s data strategy is deficient in any way, it’s going to be difficult to generate useful results.
An organization’s data strategy also must address considerations around data security and privacy. How much personal or confidential information is going to be required by or even made available by an AI solution? Think about data governance, who has access to data and what makes the most sense given what it’s trying to accomplish at an organizational level with its data strategy. Security is always going to be top of mind.
What if a patient wants personal information to be removed and does not permit its use anymore? How does an organization handle those types of requests, and how is that going to impact already deployed solutions? Let’s say an organization has trained an AI model using that data. Will it need to retrain the model with that data removed? What about any of the decisions or results that came from older models that used that data? This only increases the scrutiny and concern over handling data, privacy and security issues.
AI solutions will also require more movement to the cloud. If a healthcare organization wants to try out an AI solution, it can do it via the cloud without having a large commitment to hardware and infrastructure as a testing phase. If the solution is a fit, the organization can either increase its footprint in the cloud or decide to move some workloads in-house.
A number of cloud vendors are hosting newer AI solutions, especially those connected to generative AI, in their own environments. The expectation there is to move more data to the cloud in order to use those solutions.
Fast, reliable networks are also necessary to use any AI in a healthcare organization. High-bandwidth, low-latency networks are ideal because AI algorithms, especially deep learning ones, need speed and volume. Once a model is trained, any latency from the model to the signal for a clinician who needs to decide on medical treatment could be a life-changing matter.
Ultimately, these three areas will be important prerequisites to consider any kind of AI:
- A strong data strategy
- A modern data platform to handle large workloads, which involves cloud computing
- A clear user experience and adoption process
Data and AI governance are team sports. These aspects do not fall under any one umbrella. They involve IT, compliance, legal, clinical, any advocacy groups and more.
What’s Next for AI in Healthcare?
AI solutions should improve the lives of patients and providers. On the patient side, these solutions should make the healthcare journey seamless. On the provider side, they should reduce repetitive administrative work.
One area where AI can serve patients is in self-service options. How can patients schedule appointments? How can they receive more personalized care through the integration and analysis of multiple data sources for preventative measures?
For providers, the automation of repetitive tasks will be a major area of focus. How can ambient clinical documentation improve? How can better computer vision support a virtual nursing program? AI solutions that can take cumbersome tasks off the plates of busy clinicians so they can focus on patients will be embraced heartily.
There’s a lot of excitement around AI and its potential. There’s also recognition that if it’s used in the wrong way or gives the wrong information out, it could present a safety issue for patients. AI is not meant to do everything. It will be an augmentative tool for providers and patients that supports healthcare decision-makers.
This article is part of HealthTech’s MonITor blog series.