Monday, April 15, 2024

UC Irvine is making big strides with nursing-focused AI models

Must Read

At the University of California Irvine Sue & Bill Gross School of Nursing, faculty researchers are developing innovative new ways to harness artificial intelligence for improved patient care quality and outcomes.

Jung In Park, associate professor at UC Irvine, says she’s seeking to prepare the next generation of nurses through her biomedical research using large datasets and machine learning to provide scientific evidence for predicting patient outcomes.

Her research involves making use of national cancer registries, electronic health records and wearable sensor data to predict hospital-acquired infection, 30-day readmission and survival rates.

We spoke with Park to discuss how she is helping innovate new applications of machine learning for nurses to use in predicting patient outcomes – including Black- and Hispanic-specific survival models, outcome rates of breast cancer patients and more,.

Q. In general terms, how are you helping prepare the next generation of nurses through your biomedical research using large datasets and machine learning?

A. In the rapidly evolving landscape of healthcare, the integration of large datasets and machine learning – a subset of artificial intelligence – into biomedical research is crucial for preparing the next generation of nurses. This approach transcends the mere adoption of new technologies; it represents a comprehensive shift toward a data-driven, predictive model of patient care.

By weaving data science and AI into nursing curricula, educational institutions ensure future nurses are proficient in traditional patient care and are adept at interpreting and applying AI-driven insights. This educational strategy equips nurses with the necessary skills to analyze complex datasets, identify patterns, and leverage these insights in real-time to improve patient outcomes.

Such integration is important to empower nurses to navigate the digital transformation in healthcare effectively.

Furthermore, the application of AI in biomedical research lays a solid foundation for evidence-based practice, a fundamental pillar of nursing. Through the analysis of vast datasets, AI tools can identify trends and predict individualized patient outcomes, providing the scientific evidence necessary for nurses to make informed decisions.

This elevates the standard of patient care significantly. Such capabilities are crucial for moving beyond a generic, one-size-fits-all approach to patient care, enabling nurses to implement personalized care strategies supported by data.

Accurate predictions of individual patient outcomes empower nurses to customize interventions to the specific needs of their patients, focus on preventive measures, and proactively provide tailored care plans. This advancement in predictive analytics through AI will substantially improve care quality and patient satisfaction, and increase the overall efficiency and effectiveness of healthcare services.

Lastly, integrating AI tools and research into nursing curricula is crucial for preparing future nurses to seamlessly work with the latest healthcare technologies in our digital era. As health systems increasingly adopt AI for diagnostics, treatment planning and patient monitoring, nurses proficient in these technologies will become invaluable.

This integration ensures nurses are equipped with cutting-edge tools, keeping them at the forefront of patient care innovation. Preparing the next generation of nurses is essential for creating a nursing workforce that is capable, adaptive and ready to deliver high-quality, personalized care in the rapidly evolving age of AI.

Q. Why did you turn to AI for predicting patient outcomes?

A. The decision to leverage AI for predicting patient outcomes was driven by the need to address the complexities and limitations inherent in traditional healthcare methodologies. The exponential growth in data volume generated by healthcare systems and emerging technologies has been remarkable.

This data encompasses a wide array of sources, including electronic health records, imaging studies, genetic information and inputs from wearable technology. It became clear conventional approaches were inadequate for fully harnessing this wealth of information and handling large-scale, multidimensional datasets.

AI, with its advanced computational power and sophisticated algorithms, emerges as a powerful tool capable of analyzing these large datasets rapidly and accurately. It excels in identifying complex patterns and interactions hidden within the data, offering a more effective means of leveraging the full potential of the data available to healthcare providers.

AI’s strength lies in its ability to integrate and learn from a variety of data types, facilitating a deeper and more nuanced understanding of patient health trajectories. Traditional healthcare models have typically offered a one-size-fits-all approach, largely due to their limited ability to process and interpret the complex, multifaceted nature of human health.

Human health is dynamic, influenced by a myriad of factors including genetics, environment, lifestyle and more, all interacting in complex ways that significantly impact health outcomes. AI models, particularly those employing machine learning, deep learning or large language models, are uniquely adept at navigating this complexity.

They can analyze vast amounts of data from diverse sources and account for the multifarious interactions that influence health outcomes. This capability enables the development of highly accurate, personalized predictions, and promises more effective, individualized care that is better aligned with each patient’s specific health profile.

This shift toward personalized medicine served as a significant driving factor in my research to embrace AI for predicting patient outcomes.

Furthermore, the transformative potential of AI extends beyond personalized medicine to enabling early intervention strategies. AI’s predictive capabilities can identify patients at high risk of adverse outcomes long before these outcomes manifest, providing a critical window for intervention.

Healthcare providers equipped with these insights can proactively introduce preventative measures, tailor treatment plans more accurately and allocate resources more judiciously. This has the potential to significantly improve individual patient outcomes and reduce overall healthcare costs by mitigating the need for more intensive, expensive treatments down the line.

Such a proactive, preventative approach to healthcare is perfectly aligned with the overarching goals of enhancing the quality of patient care. By shifting the focus from reactive to preventive care, AI paves the way for a healthcare system that is more efficient, effective and patient-centered, marking a significant advancement in the pursuit of better health outcomes and more sustainable healthcare practices.

Q. You and your team developed Hispanic-specific and Black-specific survival machine learning models to analyze whether these outperformed the general model trained on all race and ethnicity data. Please describe your work on these models, and the outcomes.

A. Machine learning is recognized for its ability to discern patterns in complex, high-dimensional data to predict future healthcare events. This technique helps identify high-risk patients or those needing more healthcare services, enabling early intervention.

However, the application of machine learning in healthcare raises critical concerns regarding the perpetuation of racial and ethnic disparities. Models trained on datasets that predominantly represent the general population may not accurately reflect the experiences and outcomes of minority groups.

This discrepancy can lead to biased predictions, inadvertently exacerbating existing health disparities by failing to provide reliable outcomes for underrepresented populations.

To address this issue, my team conducted a study to tailor survival machine learning models specifically for Hispanic and Black women diagnosed with breast cancer. Our goal was to ascertain whether models calibrated for specific racial and ethnic demographics could outperform a general model trained on data encompassing all races and ethnicities.

This proof-of-concept research was to demonstrate the technical feasibility of such tailored models and to showcase their practical potential in significantly improving healthcare outcomes for underrepresented groups.

Using comprehensive data from the National Cancer Institute’s cancer registries, we crafted and fine-tuned models specifically for the Hispanic and Black populations, employing a variety of analytical methods, including the Cox proportional-hazards model, Gradient Boost Tree, survival tree, and survival support vector machines.

Our rigorous analysis, covering more than 300,000 female patients diagnosed with breast cancer between 2000 and 2017, indicated these specially designed models were indeed more effective in predicting survival outcomes for Hispanic and Black women compared to the general model.

Our study highlights the transformative potential of race- and ethnicity-specific machine learning models in healthcare. By delivering more personalized and accurate survival predictions, these models can significantly enhance the decision-making process for treatment and ultimately improve the standard of cancer care for historically underserved communities.

Furthermore, these tailored models represent a step forward in addressing the issues of representation bias and narrowing the health disparity gap.

Q. You and your team also have done work on predicting individual outcome rates of breast cancer patients to provide deeper insights into identifying treatment options and care plans for minority populations. Please elaborate on this effort, its use of AI and its outcomes.

A. Our team conducted a study employing natural language processing algorithms, a branch of AI for text analysis, to mine patient-reported outcomes of breast cancer treatment from clinical notes within EHRs, with a focus on women from underrepresented populations.

These populations included Hispanic, American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, or Multiple Race. The narrative clinical notes serve as a rich reservoir of detailed, patient-reported information, which is typically not captured in a structured format.

Despite the existing body of research on breast cancer outcomes using clinical notes, there was a noticeable gap in studies that efficiently applied NLP algorithms to specifically address the outcomes for women from underrepresented groups. To bridge this gap, we developed and evaluated various NLP methodologies to determine which algorithm performs most effectively in accurately extracting data on breast cancer treatment outcomes.

This involved a comparative analysis of different NLP approaches to identify the one that could most reliably capture the nuances and complexities of patient-reported outcomes in these specific populations.

Our study holds significant implications for future research, clinical care practices, and the shaping of health policy. It highlights the potential of NLP to deepen our understanding of breast cancer treatment outcomes, especially among underrepresented populations.

Such insights are crucial for directing more personalized and equitable healthcare strategies, ensuring that all patient groups receive the attention and care they deserve. The application of NLP in this context fosters a better grasp of patient experiences and outcomes, signaling a shift toward more inclusive health research and practice.

Additionally, by demonstrating the effectiveness of NLP in extracting valuable insights from clinical notes, our research shows the potential for streamlining the collection and analysis of patient data. Integrating these technologies into the clinical environment can enhance the quality and responsiveness of healthcare services.

Lastly, the methodologies developed through our research are not confined to the domain of breast cancer research alone; they provide a scalable and adaptable framework that can be applied across a wide range of clinical NLP applications.

By offering a blueprint for extracting and analyzing patient-reported information from clinical notes, we aim to contribute to a future where healthcare is more informed, personalized and equitable. Our goal is to pioneer advancements in healthcare that are both more informed and personalized.

We envision a future where health systems are adept at leveraging cutting-edge AI technologies, such as NLP, to more effectively meet the nuanced needs of diverse patient populations, and where data-driven insights inform every aspect of patient care. This effort will ensure that every patient, regardless of their background, has access to care that is tailored to their specific needs and circumstances.

Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki

Email him: bsiwicki@himss.org

Healthcare IT News is a HIMSS Media publication.

Read More

- Advertisement - Antennas Direct - Antennas Reinvented
- Advertisement -
Latest News

Move Forward With Multiple Sclerosis: Tips and Tricks to Manage

Relapsing-remitting multiple sclerosis can be challenging to live with, but it’s not impossible. Patsy Wheeler shares her tips and...
- Advertisement - Yarden: ENJOY $20 OFF of $150 or more with code 20YD150

More Articles Like This

- Advertisement -spot_img
×