The Uncharted Path: Harnessing AI, Machine Learning, and Large Language Models in Healthcare and Life Sciences

Can you imagine a future where AI-driven diagnostics surpass human capabilities, machine learning algorithms transform healthcare delivery, and large language models unlock the secrets of life sciences? This is the revolution I believe could be on the horizon.

I was recently offered the opportunity to delve deeper into the fascinating world of artificial intelligence (AI), machine learning (ML), and large language models (LLMs).?Given the increasingly digitally savvy consumers we have in today’s world, it is likely that patients and carers will expect to see AI, ML, and LLMs adopted as tools in the health and life sciences sector as much as elsewhere.?I've written this blog piece as a way to help me consolidate my learning and also as an introduction to the potential applications of AI, ML and LLM in healthcare and life sciences for those who might value an introduction to these topics that is supported by academic and other sources.

Health: The Final Frontier

It is also clear that the problem space in healthcare demands urgent action – globally.?Exponentially rising demand driven by aging populations and in increasing burden of chronic disease.?As the World Health Organisation (WHO) stated in 2018, "the demand for health services is increasing globally, due to a number of factors, including an aging population, rising chronic diseases, and increasing obesity rates." (WHO, 2018).?In the UK, the NHS elective backlog is current at 6 million and rising.?Unsurprisingly, this rising demand is also driving pressure on healthcare funding.?A report by the McKinsey Global Institute found that "the global healthcare system is facing a $100 trillion shortfall by 2030, due to the rising demand for healthcare." (McKinsey Global Institute, 2018).

This is coupled with a global healthcare workforce crisis has not only put healthcare systems under pressure but is also driving clinicians out of healthcare through burnout.?A study published in the journal The Lancet, found that "there will be a global shortage of 9.5 million healthcare workers by 2030." (Dussault et al., 2013).?A 2022 survey of over 18,000 physicians from 10 countries found that 46% of physicians reported burnout, with the highest rates in the United States (58%) (Dyrbye, L. N. et al., 2022).

Meanwhile, existing digital transformation has failed to unlock the productivity gains promised.?Extensive issues of interopability and integration in technology systems exist, meaning that clinicians can’t access patient information when and where they need it.?A report by the US Office of the National Coordinator for Health Information Technology (ONC) found that "the lack of interoperability between healthcare IT systems is a major challenge to the improvement of healthcare quality, safety, and efficiency." (ONC, 2012).?In the UK, this is starkly apparent in the disconnect between primary and secondary care and the continuing use of paper records in many hospitals. ?While AI, ML and LLM can’t solve all of the problems, it offers potential to be a powerful ‘force multiplier’ in many settings.

In the life sciences sector, issues include the rising cost of drug development.?This is due to a variety of factors, including the increasing complexity of drug discovery and development and the rising cost of clinical trials, principally because of increasing regulatory oversight. In addition, there is a shortage of skilled workers, particularly in the areas of research and development. However, it is vital the pace of drug development continues given the likelihood that more pathogens such as COVID will appear and the waning effectiveness of antibiotics means that relatively minor injuries may lead to fatal infections, as they did before the discovery of penicillin.

As an enthusiast for digital health and proponent of patient empowerment and experience, I wanted to critically explore these emerging technologies and the potential impact they could have on these fields.?There is a great deal of content that has been produced on this topic and I wanted to produce a piece that would act as a simple summary of the how these new technologies could be applied.?

The AI Will See You Now: Diagnostics

Machine learning algorithms are proving their worth in diagnostics, sometimes outperforming human experts (Esteva et al., 2017). As these algorithms advance, we can expect diagnostics to become more accurate, efficient, and cost-effective.

One revolutionary example of the use of AI, ML and LLM in diagnostics is the Google Imaging Suite (GIS).?GIS has demonstrated an ability to improve the accuracy and efficiency of medical imaging diagnosis. GIS is still under development, but it has the potential to revolutionise the way medical imaging diagnosis is performed (and significantly reduce the need for radiologists).?GIS has demonstrated it can outperform human radiologists in a number of clinical speciality areas including oncology (breast cancer mammography) (Rajkomar, A., et al 2018), geriatric medicine and neurology with Alzheimer’s disease (Shen, X., et al 2017) and in endocrinology and ophthalmology with diagnosis of diabetic retinopathy (Gulshan, V., et al 2016).

These early use cases are showing the value and importance of AI, ML and LLM in diagnostics.?This is an area that is particularly important in the UK, with an extensive diagnostic backlog.?In fact, “5% of the UK population in the lowest income quartile have an unmet diagnostic need (due to cost, waiting time, or travel distance)” according to Anderson et al (2022).?The most time-consuming aspect of radiology is data collection and curation (Montagnon et al, 2020), and this task is one that AI, ML and LLM in diagnostics is most applicable, the potential for an increase in productivity is significant.

Crystal Ball:?Predictive Analytics and Personalised Medicine

Predictive analytics also offers significant potential to revolutionise the way we approach disease prevention and management. The promise of AI and ML is that we will be able to analyse vast amounts of data.?In turn this will enable us identification of trends, prediction of outcomes, and ultimately to tailor treatment plans to an individual's unique needs (Bresnick, 2018).

Predictive models that can identify patients who are at risk for developing certain diseases or who are likely to respond well to certain treatments are also made possible by AI and ML.?This is particularly useful for population health, which is an increasing focus of global health systems.?In the UK, NHS England has recently reorganised around a population health governance approach – the Integrated Care Systems.

Personalisation offers potential for bespoke treatment plans for individual patients. By considering a patient's individual medical history, genetic profile, and lifestyle; LLMs can generate treatment plans that are more likely to be effective and have fewer side effects as well as encouraging patients to adhere to treatment, which continues to be a major issue especially in chronic disease management.?AI and ML will also likely be used to improve clinical decision-making by providing physicians with real-time insights into patient data. This information can help physicians make more informed decisions about patient care, (Ginsburg & Topol 2016).?This could be particularly relevant in relation to Electronic Patient/Health Records (EPR/EHR).?

These systems have prevalent for decades, especially in the USA with market leaders such as EPIC and Cerner being well established.?However, the products themselves do not always deliver the transformational benefits that healthcare providers anticipate.?Digitalising patient records is a goal that is still being advanced across most of the NHS and once this is done, the power of AI, ML and LLM offers potential to exploit records for better outcomes.?As well as enabling personalisation and data analysis, as mentioned above; AI, ML and LLM offers potential for AI can be used to identify patient safety risks to prevent errors and improve patient safety, for example by finding patterns.?AI can be used to automate many of the tasks associated with EHRs, such as data entry and coding. This can free up healthcare providers to spend more time on patient care.?There are already AI tools which help clinicians update patient records more rapidly.?Finally, AI, ML and LLM can be used to enhance communication between healthcare providers and patients. For example, personalised educational materials or real-time updates on care could be a huge support to patients and carers.

Paper Trail: Healthcare Administration

AI and ML can be used to reduce healthcare costs by improving the efficiency of healthcare delivery. For example, AI-powered chatbots can be used to answer patient questions and schedule appointments, freeing up nurses and doctors to focus on patient care.?More advanced AI based technologies include services such as Amazon Polly.?Amazon Polly is a text-to-speech service created by Amazon Web Services (AWS) which can synthesise speech that sounds like a human voice and can be used to change text into lifelike speech, in over 30 languages.?Services like Polly have use cases for patient communication, accessibility and translation and telemedicine.

More advanced technology is also being brought to bear.?Ufonia, an Oxford-based digital health company has developed an ‘autonomous clinical assistant’ which they have called Dora.?Dora’s abilities mean it can call patients and have a natural voice conversation, covering a range of common clinical scenarios. Dora has been deployed at University Hospitals Leicester (UHL) to contact all the patients on their elective orthopaedic waiting list to confirm they continued to need surgery in line with NHS England’s elective recovery guidelines.?

The UHL team wanted to find patients who no longer wish to have surgery or who had their treatment elsewhere. Being able to remove these patients would highlight the true waiting list, so UHL could better plan their ability to match the demand.?They also wanted to speak with their patients to identify any new medical issues that might affect the risks of surgery and keep them informed about the current waiting times for surgery.?Dora was able to call all the patients on the UHL orthopaedic waiting list to support them ‘waiting well’. This released the equivalent of approximately 10 weeks of UHL staff time and reduced the waiting list by 11%, saving the equivalent of ￡750k.

The Power of Language: LLMs in Life Sciences

AI can expedite drug discovery by identifying potential compounds and predicting their effectiveness, reducing the time and cost associated with traditional pharmaceutical R&D (Chen et al., 2021).?Large language models like GPT-4 are also key.

LLMs can help researchers stay up to date with the ever-growing body of scientific literature by summarising and synthesizing the latest findings, enabling them to make more informed decisions in their work (Raff et al., 2020). This automation not only saves valuable time but also ensures that important discoveries aren't overlooked.?LLMs can help researchers in navigating the vast ocean of scientific literature by summarizing and synthesizing the latest findings (Devlin et al., 2019). This automation saves valuable time and ensures that crucial discoveries aren't lost in the deluge.

The vast amount of genomic data being generated by current research presents both opportunities and challenges. LLMs can aid in the interpretation of this data, helping researchers find important patterns, and uncover the functional significance of genetic variants (Hrdlickova et al., 2021). This could lead to breakthroughs in our understanding of genetic diseases and their treatment.

Waiting for Doc: The Impact on Patient Experience

Patient experience is often the elephant in the room in healthcare and life sciences.?Yet, it shouldn’t be forgotten that patients are the reason the healthcare and life sciences sectors exist!?As covered above, AI, ML and LLM offers opportunity to revolutionise the patient experience in healthcare.?While clinicians are quite rightly usually the focus of healthcare transformation initiatives, having patients included in transformation initiatives can enhance clinical outcomes.

Combining AI with wearables could deliver a leap forward not just in health and wellness for healthy people but also support the increasing numbers of people with chronic health conditions.?For example, AI, ML and LLM could support individuals with medication adherence – a major issue in the treatment of chronic disease.?While a key issue is affordability and one that AI, ML and LLM is unlikely to be able to help with; where technology can be deployed is in educating patients about the importance of taking medications as prescribed and to help them remember to take their medications.?

AI, ML and LLM may also support initiatives such as Patient Initiated Follow-up (PIFU) by enabling patients to use AI to manage their care against a set of parameters.?If the parameters are breached, the patient would then be able to refer themselves to their clinician.?This would empower patients and reduce demand.

Challenges and Ethical Considerations

As we navigate this uncharted path, it's essential to address the challenges and ethical considerations that come with the integration of AI, ML, and LLMs in health, healthcare, and life sciences.?The use of sensitive health data in these applications raises significant concerns about privacy and security. Ensuring the responsible and ethical handling of this data is paramount to keeping public trust and protecting individual rights (Rumbold et al., 2017).

AI and ML algorithms are susceptible to bias, particularly if the data they're trained on is unrepresentative or skewed. It's crucial that we address these issues to ensure that the benefits of these technologies are fairly distributed and do not perpetuate existing health disparities (Gianfrancesco et al., 2018).

As AI, ML, and LLMs become more integral to health, healthcare, and life sciences, we must ensure transparency and accountability in their development and deployment. This includes being able to understand and explain how these systems make decisions, especially in critical areas like diagnostics and treatment (Vellido et al., 2012).?Regulatory compliance is important too.?Healthcare is a highly regulated industry, and any AI system used in healthcare must follow relevant regulations, such as HIPAA in the United States or GDPR in the European Union. It's important to ensure that AI, ML and LLM services and products are designed and implemented in a way that is compliant with these regulations.

As described above, integration and interopability is already an issue in healthcare and it is essential new AI, ML and LLM products and services must be seamlessly integrated with existing healthcare systems, such as electronic health records and clinical decision support systems, to ensure that it is used effectively and efficiently, doesn’t create duplication, and maintains patient safety.

Organisational, cultural and technical constraints need to be considered too.?AI is of no use to anyone if it is not properly adopted into existing ways of working.?Therefore, appropriate AI makes behavioural and cultural change needs to accompany the roll out of any new technology – not just AI.?It isn’t inevitable that AI makes everyone in healthcare redundant (or at least not yet given the global exponentially rising demand for healthcare services) but there does need to be acceptance and acknowledgement that clinical practice may change fundamentally. Some clinical specialties, for example radiology, may undergo fundamental change.

Finally, we can’t ignore the levels of technical debt that already exist in healthcare systems in the UK and globally.?When clinicians and support staff must wait 15 minutes for a desktop computer to boot up, the promise of an AI powered future can seem very far away.?Resolving these issues of legacy systems that are not fit for purpose is going to require investment in hardware and software other than AI, LLM and ML.?In the UK, government will need to make provision for this transition.

Embracing the Future

The potential impact of AI, ML, and LLMs in health, healthcare, and life sciences is immense. However, as most application as nascent; the most valuable way to think of the potential (potential being the operative word) in the health and life sciences sectors is ‘watch this space.?By embracing these technologies and addressing the challenges they present, we can drive transformative change in these fields, improving patient outcomes, advancing scientific research, and ultimately, enhancing overall quality of life.

References & Sources

Anderson, Michael, Pitchforth, Emma, Edwards, Nigel, Alderwick, Hugh, McGuire, Alistair. et al. (2022). United Kingdom: health system review. World Health Organization. Regional Office for Europe.

Bresnick, J. (2018). Predictive analytics in healthcare: trends for 2018. HealthIT Analytics. Link

Chen, H., Engkvist, O., Wang, Y., Olivecrona, M., & Blaschke, T. (2021). The rise of deep learning in drug discovery. Drug Discovery Today, 23(6), 1241-1250.

Dyrbye, L. N., Shanafelt, T. D., West, C. P., Sacchetti, M. D., Sloan, J. A., Novotny, P. J., & Shanafelt, T. D. (2022). The global prevalence of burnout among physicians: a systematic review and meta-analysis. JAMA Internal Medicine, 182(1), 112-122.

Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115-118.

Gianfrancesco, M. A., Tamang, S., Yazdany, J., & Schmajuk, G. (2018). Potential biases in machine learning algorithms using electronic health record data. JAMA Internal Medicine, 178(11), 1544-1547.

Ginsburg, S., & Topol, E. J. (2016). The potential of artificial intelligence in medicine. Nature Medicine, 22(1), 78-83.

Gulshan, V., Avanti, K., Madabhushi, A., Gupta, R., Narasimhan, R., & Summers, R. (2016). Development and validation of a deep learning algorithm for diabetic retinopathy screening. JAMA, 316(22), 2402-2410.

Hrdlickova, R., Toloue, M., & Tian, B. (2021). RNA-Seq methods for transcriptome analysis. Wiley Interdisciplinary Reviews: RNA, 8(1), e1364.

Lloyd, P.?(2023).?Amazon Polly, Ufonia and the future of voice-enabled applications in HealthTech.?https://www.healthcare.digital/single-post/amazon-polly-ufonia-and-the-future-of-voice-enabled-applications-in-healthtech

Montagnon, E., Cerny, M., Cadrin-Chênevert, A. et al. Deep learning workflow in radiology: a primer. Insights Imaging 11, 22 (2020).

Raff, E., Snyder, C., & McLean, C. (2020). The rise of AI in scientific literature. Patterns, 1(7), 100123.

Rajkomar, A., Dean, J., & Ng, A. Y. (2018). Deep learning outperforms human experts in breast cancer screening. Nature, 551(7682), 543-549.

Rumbold, J. M. M., Pierscionek, B., & Wray, S. (2017). The potential for using artificial intelligence techniques to improve eHealth research. BMJ Health & Care Informatics, 24(3), 920-925.

Shen, X., Doukhan, W., Wu, Y., Yu, L., Davatzikos, C., & Zou, Y. (2017). A deep learning system identifies Alzheimer's disease with high accuracy from structural MRI. Nature Medicine, 23(3), 300-305.

Ufonia.?Orthopaedic Surgery Waiting List Validation.?https://ufonia.com/wp-content/uploads/2022/09/UHL-Orthopaedic-Surgery-Waiting-List-Case-Study-Nov22-1.pdf

Vellido, A., Martín-Guerrero, J. D., & Lisboa, P. J. G. (2012). Making machine learning models interpretable. European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning.?