As tempting as it is to jump on the AI bandwagon, healthcare organizations need to cautiously evaluate how and where to implement these emerging technologies. Forward-thinking healthcare organizations can start small with AI, selectively applying it to lower-risk scenarios. For example, chatbots and virtual assistants can help patients conveniently schedule appointments, freeing up call center resources. Robotic process automation can take on administrative tasks like billing and claims processing. The communications team could use AI tools to revamp website content. However, each use case requires matching the right underlying AI technology to achieve optimal results.
For example, large language models (LLMs) like ChatGPT are foundational machine learning models trained on vast amounts of information. LLMs can intake text and prompts to turn them into enriched content, excelling at natural language tasks like summarization. While LLMs are a powerful tool for simple AI tasks, they still lack the reliability needed for data-driven analytics and automation. LLMs use a limited kind of logic, reasoning, and even common sense to produce text, but they can’t yet perform complex calculations or analysis. Sometimes they “hallucinate,” producing plausible-sounding answers that are incorrect, and sometimes they cannot adequately explain their conclusions. Various tricks have been tried to improve these shortcomings, but today’s best approaches still involve passing analytics tasks to some other (non-LLM) process.
Hospitals seeking to leverage AI for data-driven decision-making should consider leveraging an AI-driven algorithmic solution. Hospitals have troves of data—from financial, to clinical, to operational. Integrating these disparate data sources to extract meaning has long challenged healthcare. Now, algorithm-based analytics solutions can help hospitals to connect these data dots.
Unlocking Healthcare Data with AI
Algorithms are nothing new, but combining them with the power of AI is taking healthcare analytics to the next level. Here’s how algorithms are driving AI-based advancement with healthcare data:
- Augmented analytics help to automate analysis tasks to reduce the administrative burden of manual analysis. For example, automated clinical outlier detection or clinical alerts for at-risk patients can help free up valuable time for clinicians while improving patient care.
- AI-enabled predictive analytics is a powerful tool, especially when it comes to operational decision-making. This technology can help clinicians to identify clinical trends and make data-driven decisions.
- Benchmarking analytics can help hospitals measure performance against industry standards and identify risk indicators for patients using scoring methodologies. For example, LACE scores can be used to predict the likelihood of readmission.
- Statistical analysis approaches, like regression analysis and control charts, using tools like R and Python, can be integrated with open AI libraries to enhance AI capabilities. There are many R and Python libraries for statistical analysis and machine learning, and these can be used as powerful back-end tools to support analysis and decision-making. Many analytics vendors also provide statistical tools for a lightweight “no code” approach.
AI-driven algorithmic approaches are the next frontier for healthcare analytics. These novel techniques based on the best of machine intelligence, paired with human expertise, can empower hospitals to extract maximum value from data while keeping patients safe.
Questions to Ask Before Adopting AI
The key is for hospitals to understand each technology’s strength and the best use case for applying it. Furthermore, hospitals must consider how to thoughtfully combine the forces of AI-based tools and human intelligence. As expert AI computer scientist Oren Etzioni aptly states, “AI is a tool. The choice about how it gets deployed is ours.”
When evaluating new AI solutions, healthcare organizations should ask critical questions, such as:
- Is the technology transparent in how it reaches decisions? Black-box models can harbor risks.
- How reliable and stable is the solution based on rigorous testing?
- Can the system explain its reasoning and actions to human auditors?
- Does the vendor offer robust governance practices around security, privacy, and ethics?
- If the AI makes a mistake, how is the system retrained and errors corrected?
AI should not run fully unleashed within healthcare; appropriate oversight is crucial. But if avoided entirely, organizations miss out on potential benefits. The measured approach is to run limited pilots with analyst supervision, empirically evaluating AI on metrics like improved patient outcomes and clinician productivity before expanding its reach.