AI and Machine Learning: The Evolving Reality of Predictive Healthcare

Written By: Dr David Owens, Specialist in Family Medicine

The advent of Artificial Intelligence (AI) in healthcare is bringing a shift towards more predictive, personalised, and efficient care. This is being driven by significant advances in technology but is also bringing challenges around implementation and data security. As AI reshapes patient care, diagnostic accuracy, and treatment planning, it also requires a critical examination of the issues surrounding data integration, algorithmic transparency, and the ethical use of AI. The healthcare sector stands at a crossroads, tasked with adopting AI innovations while fostering an environment that ensures patient safety, data privacy, and equitable access to care.

AI's Use in Healthcare

AI is poised to transform healthcare across various fields, including patient risk stratification, patient communication, personalised care, and disease outbreak prediction. These applications are not just theoretical; they are already making a significant impact.

The NHS, for example, leverages AI to sift through health records for early disease prediction¹, aiming to enhance patient outcomes with timely interventions. In a similar vein, a Vanderbilt University study² combined electronic health records (EHR) and genetic information to predict cardiovascular disease. 

AI's Role in Advancing Medical Care

The integration of AI into healthcare is already leading to improvement in both patient care and medical research. Unity Health Toronto³ has improved patient care by integrating over 50 AI tools into clinical practice, significantly improving outcomes through early sepsis detection and brain-bleed identification. 

Google's DeepMind project is making strides in medical diagnostics by accurately diagnosing diabetic retinopathy and macular oedema with over 94% accuracy⁴, showcasing AI's potential to enhance diagnostic precision. Insilico Medicine utilised AI machine learning in order to reduce⁵ timelines for drug discovery, accelerating the development of new treatments. 

Challenges in AI and Healthcare IT System Integration

Interoperability and data integration are critical issues in the marriage of AI with existing healthcare IT systems and data sources, notably in the widespread implementation of Electronic Health Records (EHRs). EHRs are increasingly adopted worldwide, with 47% of nations employing national EHR systems⁶ to enhance healthcare efficiency, efficacy, and to reduce costs. However, the diversity of EHR data formats poses a substantial challenge to interoperability. This challenge is exacerbated by the AI's need to process data across these varied formats, complicating the integration of AI technologies into healthcare systems. For instance, a significant obstacle to AI advancements in healthcare is the lack of transparency and interpretability of AI algorithms, especially when dealing with unstructured data. This complexity accentuates the necessity for robust interoperability standards that facilitate AI models to effectively communicate with and utilise data from different EHRs.

The deployment phase of AI in healthcare highlights the importance of establishing and adhering to interoperability standards. Without these standards, AI models may struggle to connect with and interpret data across the myriad of electronic health records, limiting their effectiveness and application in real-world healthcare settings.. As such, healthcare professionals and IT experts will need to collaborate in order to develop and implement interoperability standards to facilitate the usefulness of AI integration.

Navigating Ethical and Regulatory Landscapes in AI Integration

The integration of AI-driven solutions in healthcare brings complex ethical and regulatory challenges. Bias in AI algorithms is an important issue, as evidenced by a Nature article⁷ on an AI algorithm for breast cancer screening that disproportionately classified black patients as "low risk." This disparity highlights the need for diverse and inclusive training datasets to prevent bias, which can emerge at any stage of AI development, from data collection to deployment

Regulatory challenges further complicate the landscape, particularly due to the rapid evolution of AI in healthcare and the consequent lag in regulatory frameworks. Although the EU Parliament has adopted the AI ACT⁸, aiming to regulate AI systems across the EU, specific regulations tailored to healthcare AI are still lacking. This gap is critical given the sensitivity of health data and the potential for significant adverse effects on individuals and communities. Existing frameworks like the GDPR offer some protection for personal data, but healthcare data require additional safeguards to prevent breaches of patient privacy. As an example, regulations could demand stringent encryption, regular audits, and strict penalties for data breaches, akin to the standards set by the Health Insurance Portability and Accountability Act (HIPAA) in the United States. Such measures would ensure that AI applications in healthcare remain transparent and secure in their handling of patient data, addressing both ethical and regulatory concerns to foster a safe and equitable integration of AI into healthcare systems.

We are currently sitting at the threshold of a new area in healthcare which will be driven by advancements in Artificial Intelligence (AI) . The future offers enhanced diagnostic accuracy, personalised treatment, and improved patient outcomes. However, the road to fully realising the potential of AI in healthcare is not without challenge. We must overcome issues around data interoperability, ethical application, and regulatory compliance in addition to currently unknown future challenges. As we move forward, it is essential that the healthcare community collaborates in order to overcome these hurdles, ensuring that the integration into healthcare delivers on its promise of improved care for patients.

References

1. Artificial intelligence (AI) and machine learning (2023) NHS England. Available at: https://www.england.nhs.uk/long-read/artificial-intelligence-ai-and-machine-learning/ (Accessed: 15 March 2024).
2.  Zhao, J. et al. (2019) ‘Learning from longitudinal data in electronic health record and genetic data to improve cardiovascular event prediction’, Scientific Reports, 9(1). doi:10.1038/s41598-018-36745-x.
3.  Bresge, A. (2024) How ai will – and won’t – change health care in 2024, University of Toronto. Available at: https://www.utoronto.ca/news/how-ai-will-and-won-t-change-health-care-2024 (Accessed: 15 March 2024).
4.  Gulshan, V. et al. (2016a) ‘Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs’, JAMA, 316(22), p. 2402. doi:10.1001/jama.2016.17216.
5. Vora, L.K. et al. (2023) ‘Artificial Intelligence in pharmaceutical technology and Drug Delivery Design’, Pharmaceutics, 15(7), p. 1916. doi:10.3390/pharmaceutics15071916.
6.  Borna, S. et al. (2023) ‘Artificial Intelligence Models in Health Information Exchange: A systematic review of clinical implications’, Healthcare, 11(18), p. 2584. doi:10.3390/healthcare11182584.
7.  Vokinger, K.N., Feuerriegel, S. and Kesselheim, A.S. (2021b) ‘Mitigating bias in machine learning for medicine’, Communications Medicine, 1(1). doi:10.1038/s43856-021-00028-w.
8.  Hainsdorf, C. et al. (2023) Dawn of the EU’s AI Act: Political Agreement reached on world’s first comprehensive horizontal AI regulation, White & Case LLP. Available at: https://www.whitecase.com/insight-alert/dawn-eus-ai-act-political-agreement-reached-worlds-first-comprehensive-horizontal-ai (Accessed: 15 March 2024).