Augmented intelligence (AI) tools are becoming mainstream, but experts caution healthcare leaders that not all AI applications are equal. Data scientists explore the pros and cons of using auto-generated responses by ChatGPT, an AI natural language model, to analyze hospital length of stay.
Download
Physicians are experimenting with augmented intelligence (AI) tools like ChatGPT to help explain medical treatments or procedures in plain, conversational language.
ChatGPT, developed by OpenAI, is one of the most advanced publicly available language models, partly due to how it captures the nuances and intricacies of human language and then generates appropriate and contextually relevant responses. Based on the generative pre-trained transformer (GPT) architecture, ChatGPT uses deep learning techniques to produce human-like responses to natural language inputs.
Healthcare leaders are increasingly recognizing the potential benefits of integrating AI technologies like ChatGPT into their operations. One area where AI has shown promise is in improving patient care through personalized medicine. By using language appropriate for different age groups or adjusting their tone, for instance, doctors can effectively address patients’ concerns and may alleviate their fears, which improves patient engagement and compliance with treatment plans. Although the AI-generated response can complement a healthcare professional’s knowledge and expertise, experts caution providers to review and validate the response for clinical accuracy.
Despite the promise of augmented intelligence in healthcare, its limitations have hindered widespread adoption. Therefore, what is the hype surrounding AI in healthcare? And are health systems missing opportunities to leverage AI applications to solve their most pressing challenges? Data Scientists Jason Jones, PhD and Jessica Curran of Health Catalyst recently explored these important and timely questions in a webinar entitled, AI in Healthcare: Hype, Hope & Missed Opportunities.
In addition to patient education, experts envision AI solutions assisting clinicians by automating routine, administrative tasks, such as managing electronic health records or identifying patterns and trends within large datasets. Streamlining these tasks is increasingly important given the current strain on hospitals and healthcare provider organizations due to financial and labor hardships.
For example, the average length of stay (LOS) for patients in a hospital can contribute to escalating medical costs and unpredictable insurance reimbursements. If not addressed, these issues could delay timely post-acute medical care and hamper a patient’s recovery. Delays in discharge are also known to strain an already overworked hospital staff. To add to the urgency in addressing LOS, the American Hospital Association released a report indicating that the average LOS in hospitals has increased by 19 percent in 2022 compared to 2019. AI tools like ChatGPT pique the interest of healthcare leaders looking for high-tech and ready-made solutions to assist their clinicians and data analysts in addressing these known challenges.
Jones and Curran tested LOS analysis and scenarios through ChatGPT during the July 19 webinar. After inputting summary LOS data into the tool, Curran, who serves as Vice President of Data Science and Analytics at Health Catalyst, then asked ChatGPT what the healthcare data could reveal about a hospital’s LOS. Specifically, she asked ChatGPT the following question: What is our baseline performance, and has anything changed over time? Curran discussed the benefits and drawbacks of the resulting AI-generated analysis:
Pros:
Cons:
Despite its strong ability to summarize and explain information, the pair of data scientists deemed the application inadequate, failing to meet to the level of expertise needed to inform clinical care and effectively assess program or organizational performance.
In the example of tracking LOS, ChatGPT was able to produce a sophisticated response through R code, a universal programming language, and plot the data over time; however, the data and graph lacked the insight to help answer the question at hand: Where are we with LOS in our organization? Unfortunately, the output of ChatGPT proved insufficient to address complex operational challenges. The question remains: Are there any AI tools available that can generate reliable answers for healthcare organizations seeking to improve their LOS?
To answer that question, Curran and Jones, Chief Analytics Officer and Chief Data Science Officer, turned to Healthcare.AI, which integrates into business intelligence tools and uses well-established and cutting-edge statistical and machine learning techniques to solve healthcare problems at the clinical and organizational level.
During the webinar, they applied the same LOS data to Healthcare.AI. The solution instantly produced clear and readable data interpretations, including:
Healthcare leaders and providers often struggle with seeing beyond the present situation, especially when confronted with urgent circumstances or attending to patients in real time. But experts contend that the integration of AI tools can help health system executives overcome these limitations and gain insights into potential future scenarios at the organizational level. What’s more, focusing on the root cause of clinical or organizational challenges could enable organizations to enhance their healthcare data output and expedite AI-driven outcomes. Experts assert that AI creates opportunities to broaden the discussion and address what is not readily apparent from a traditional BI chart or analysis, such as:
“Missed opportunities arise when there isn’t adequate data analysis to support the discussion,” Curran said during the presentation.
Predictive modeling, forecasting, and anomaly detection are among the various applications of AI tools. In the LOS example, organizations can leverage Healthcare.AI to determine which facility is treating patients with more critical medical conditions and who are at a greater risk of mortality based on observed to expected (O/E) ratios. Additionally, AI end users can compare specific severities across patient populations and identify the facilities that are likely to improve their performance by implementing strategies to reduce LOS.
Beyond quality and financial concerns such as LOS, incorporating AI can drive better decisions and actions for the entire healthcare community. Improvement areas may also include addressing the high cost of prescription medications. By utilizing this technology to evaluate patient populations, health systems can save both time and costs associated with patient trials. Along those lines, AI functionality can easily draw comparisons between patients participating in a low-cost medication program to those not enrolled in the program, and begin to answer critical questions, such as: Does this method impact care quality? How do patient outcomes compare? How will program interventions change outcomes?
Experts agree that ChatGPT stands out as a remarkable development in natural language processing. While its impact on healthcare remains limited in scope, the immense potential of augmented intelligence tools cannot be underestimated. Indeed, when it comes to leveraging AI in clinical settings, the goal for the entire organization should be to gain a shared understanding of the current state or problem, as well as the desired result, or where they want to be. A shared focus will guide the decision-making process so that healthcare leaders can choose a solution that best fits their needs and delivers maximum benefit. When evaluating AI tools, Jones and Curran urged health systems to prioritize the following features:
When it comes to adopting augmented intelligence tools like Healthcare.AI and natural language models like ChatGPT, it’s still in the early stages and requires careful consideration. Yet, as new technological capabilities emerge and patients expect more from healthcare systems and providers, data scientists believe AI will be instrumental in facilitating greater collaboration, streamlined processes, and improved diagnostic accuracy.
Would you like to learn more about this topic? Here are some articles we suggest:
