A Counter Argument to Microsoft and Epic’s Partnership: Introducing a New Approach with ChatGPT-4, EHRs, and DICOM Images
Authors: Amina Khalpey, PhD, Parker Wilson, DO, Zain Khalpey, MD, PhD, FACS
Abstract:
This paper critiques the recent partnership between Microsoft and Epic Healthcare Systems to harness the power of generative artificial intelligence (AI) to improve electronic health records (EHRs). We believe this partnership is a step in the wrong direction toward AI in healthcare. We outline the major problems associated with using big software companies to introduce AI in healthcare. As a rebuttal, we propose a novel approach to using generative AI with ChatGPT-4 for EHRs and reading Digital Imaging and Communication in Medicine (DICOM) images. Additionally, we present a ten-step process and timeline for implementation.
Introduction:
The recent announcement of Microsoft and Epic’s partnership to integrate Microsoft Azure OpenAI Service with Epic’s EHR platform, leveraging generative AI, has generated excitement and optimism in the healthcare industry.1 One of the initial generative AI solutions is already in the works at UC San Diego Health, UW Health in Madison, WI, and Stanford Health Care involving enhancements to automatically drafted message responses.
Although, this may seem exciting at first. We believe this partnership should be approached with caution as it may not be the panacea it promises to be. Several problems are associated with using these software companies to elevate AI in healthcare. At a glance, these companies are shrouded in bias and profiteering. They slowly belabor changes and will inevitably produce a product that is not helpful to clinicians because these companies are not built for clinicians. Below we list several of the problems we see with this deal.
Major Problems with Microsoft and Epic’s Partnership:
1. Slow pace of innovation: Large companies often struggle with agility and swift implementation of new technologies.2 This could hinder the advancement and adoption of generative AI in healthcare, a technology that is moving extraordinarily fast.
2. Lack of interoperability: The current partnership may not address the interoperability challenges that persist in healthcare, limiting the effectiveness of EHRs and AI solutions.3 This will lead to increasing variance in healthcare systems and could contribute to healthcare inequalities.
3. Lack of Clinician Input: Both of these companies do not seek intense clinician input or scrutiny for development. When creating an AI tool for healthcare, clinicians need to be intimately involved with development. Although Epic has long been the leader in EHR development, this is a new frontier of medicine. To create the best product, clinicians must be leading AI development in healthcare rather than a software company like Microsoft.
4. Potential for biased algorithms: The use of incomplete or biased datasets in training generative AI models can lead to biased algorithms, negatively impacting patient care and outcomes. Undoubtedly, the first renditions of this partnership will be biased and incomplete. This will call for broad collaboration between these two software giants, clinicians, regulators and other parties as to eliminate bias and broaden the applicability of the technology.
5. Data privacy and security concerns: Collaborations between these two corporations should raise concerns about data privacy and security, given their history of high-profile breaches in the last calendar year.4,5
6. Vendor lock-in: This partnership provides only one collaboration and solution for the future of AI in healthcare. Leaving healthcare providers to rely on a single solution, they may be locked into using specific EHR platforms, which will reduce competition and stifle innovation.
With these drawbacks to the new partnership between Microsoft Azure and Epic Healthcare Systems, we propose a novel approach to generative AI incorporation into EHRs. Our specific example involves DICOM images and their recognition and evaluation.
A New Approach: ChatGPT-4, EHRs, and DICOM Images:
To address concerns about this proposed introduction of AI into healthcare, we propose a novel approach using generative AI with ChatGPT-4 for EHRs and reading DICOM images. This approach allows for a more agile implementation of AI technologies, reduces the risk of biased algorithms, and encourages interoperability, data privacy, and security.
Ten-Step Process and Timeline:
1. Identify stakeholders (1-2 months):
Engage with key stakeholders, including healthcare providers, patients, researchers, and regulatory bodies, to ensure their needs and concerns are addressed.
2. Develop data standards (2-4 months):
Establish standardized data formats for EHRs and DICOM images to promote interoperability and streamline data sharing.
3. Assemble diverse, unbiased datasets (4-6 months):
Gather comprehensive and diverse datasets for training AI algorithms, ensuring they are representative of the target population.
4. Train ChatGPT-4 models (6-12 months):
Train generative AI models using these unbiased datasets to ensure accurate and reliable predictions.
5. Implement rigorous validation (12-18 months):
Validate the AI models through rigorous testing and evaluation, ensuring they meet performance benchmarks and comply with ethical guidelines.
6. Develop data privacy and security protocols (12-18 months):
Establish robust data privacy and security protocols to protect sensitive patient information.
7. Integrate ChatGPT-4 into EHR systems (18-24 months):
Seamlessly integrate ChatGPT-4 with existing EHR systems, allowing healthcare providers to access AI-generated insights.
8. Implement ChatGPT-4 for DICOM image analysis (24-30 months):
Leverage ChatGPT-4 for automated analysis of DICOM images, improving diagnostic accuracy and reducing radiologist workload.
9. Monitor and refine AI models (30-36 months):
Continuously monitor the performance of AI models, incorporating feedback from healthcare providers and patients, and refining the models as needed.
10. Foster a collaborative ecosystem (36-48 months):
Encourage collaboration between healthcare providers, researchers, and other stakeholders to share insights, best practices, and promote continuous innovation in the application of generative AI to EHRs and DICOM images.
Discussion:
This process would create a secure and useful AI tool that integrates effectively into EHRs. If AI models are created prior to EHR integration, it could be integrated into multiple different EHR systems and improve interoperability. There are additional benefits, challenges, and potential implications of using ChatGPT-4 with EHRs and DICOM images in healthcare and they are outlined below.
Benefits:
1. Improved accuracy and efficiency: By leveraging ChatGPT-4’s generative capabilities, the proposed approach can help improve the accuracy of EHRs and DICOM image analysis, reducing errors and enabling healthcare providers to make more informed decisions.
2. Reduced clinician burnout: Automating time-consuming tasks, such as data entry and image analysis, can help alleviate clinician workload, ultimately reducing burnout and allowing healthcare providers to focus on patient care.
3. Enhanced interoperability: The establishment of standardized data formats for EHRs and DICOM images can promote seamless data sharing between healthcare providers, fostering collaboration and improving patient outcomes.
4. Encouragement of innovation: By reducing vendor lock-in and fostering a collaborative ecosystem, the proposed approach can stimulate innovation, driving continuous advancements in AI-driven healthcare solutions.
Challenges:
1. Data privacy and security: Ensuring the privacy and security of sensitive patient data is a critical concern in the proposed approach. The development and implementation of robust data privacy and security protocols are necessary to address this challenge.
2. Ethical considerations: The use of AI in healthcare raises ethical questions, such as the potential for biased algorithms, transparency, and accountability. Ensuring that AI models adhere to ethical guidelines and are transparent in their decision-making processes is essential to mitigating these concerns.
3. Integration with existing systems: Integrating ChatGPT-4 with existing EHR systems may pose technical challenges, requiring collaboration between healthcare providers, technology vendors, and other stakeholders.
4. Adoption and acceptance: Convincing healthcare providers to adopt and trust AI-driven solutions may be a challenge. Ensuring the effectiveness, reliability, and interpretability of AI models is crucial for fostering acceptance among healthcare providers.
Potential Implications:
1. Personalized patient care: The integration of ChatGPT-4 with EHRs and DICOM images could enable more personalized patient care, as AI models can analyze vast amounts of data and generate tailored insights for individual patients.
2. Cost reduction: Automating various tasks and improving the efficiency of EHRs and DICOM image analysis could lead to cost reductions in healthcare, benefitting both providers and patients.
3. Global health impact: The proposed approach has the potential to benefit healthcare systems worldwide, particularly in low-resource settings, where access to specialized care and diagnostic tools may be limited.
Conclusion:
While the partnership between Microsoft and Epic offers potential benefits, it also raises concerns about the pace of innovation, biased algorithms, interoperability, data privacy, and security. By leveraging ChatGPT-4 with EHRs and DICOM images, we can address these concerns and realize the full potential of generative AI in healthcare. The proposed ten-step process and timeline offer a roadmap for the successful implementation of this novel approach, fostering a more agile and ethical AI-driven healthcare ecosystem.
References:
Microsoft News Center. Microsoft and epic expand strategic collaboration with integration of Azure Openai Service. Microsoft. https://news.microsoft.com/2023/04/17/microsoft-and-epic-expand-strategic-collaboration-with-integration-of-azure-openai-service/. Published April 17, 2023. Accessed April 26, 2023.
Wright, George & Heijden, Kees & Bradfield, Ronald & Burt, George & Cairns, George. (2004). Why Organizations are Slow to Adapt and Change—and What Can Be Done About It. Journal of General Management. 29.
Reisman M. EHRs: The Challenge of Making Electronic Data Usable and Interoperable. P T. 2017 Sep;42(9):572-575
LLC EM. Epic management provides notice of data security incident. PR Newswire: press release distribution, targeting, monitoring and marketing. https://www.prnewswire.com/news-releases/epic-management-provides-notice-of-data-security-incident-301703362.html. Published December 14, 2022. Accessed April 26, 2023. Heiligenstein MX. Microsoft data breaches: Full timeline through 2023. Firewall Times. https://firewalltimes.com/microsoft-data-breach-timeline/. Published April 10, 2023. Accessed April 26, 2023.