Albert: Utilizing Large Language Models in Healthcare Solutions
Table of Contents
- Introduction
2. Albert in Health Management
3. Albert and Large Language Models
4. Generating a Prompt
5. Diabetes Website
6. Conclusion
1. Introduction
Digital transformation, especially after the pandemic, has gained great momentum in the healthcare sector. This evolution offers significant advantages in critical areas of healthcare processes, especially in fundamental processes such as patient history tracking, diagnosis, research, and treatment planning based on theoretical knowledge. For example, if a patient receives services from different healthcare institutions, they can quickly and securely share all their information with healthcare professionals thanks to digital health records. This digitalization provides healthcare professionals with the opportunity to manage data management, patient monitoring and treatment processes more effectively.
Within the scope of digitalization, artificial intelligence technologies that have entered our lives are increasingly demonstrating their benefits in every sector, and they have the potential to revolutionize the healthcare sector as well. One of the potential benefits offered by artificial intelligence is its ability to communicate and interact with users. For example, thanks to this communication ability, a practical tool can be provided to patients to easily access the information they need, thereby providing support in better understanding and managing diseases.
One of the most important components for effective health management is that patients are aware of their health. In this context, health literacy is of great importance. As Albert Health, we have developed disease-specific AI-based websites where users can ask their health questions and get information about everything they are curious about about their health, in order to support individuals to better understand their health, manage it effectively and live a healthier life.
2. Albert in Healthcare Management
Albert is a clinical outcomes-driven and AI based chronic disease management platform. Albert to users; It offers customized disease management programs for various chronic diseases, including diabetes, asthma, hypertension, MS, and cancer, through a mobile device. Users can set medication reminders, access drug package insert information, and receive warnings about drug interactions. You can access healthy living, nutrition and exercise recommendations. Important health metrics such as blood sugar level, blood pressure values, weight changes, daily amount of water drunk, daily step count can be recorded in Albert and the measurements can be easily tracked. They can store their health data in one place and share it with their relatives and healthcare professionals.
On the other hand, Albert is an voice-based AI health assistant who is always with patients and whom they can consult. Users can control the application with voice commands and ask Albert their health questions by voice or text. Albert also finds the most appropriate answers by understanding the needs of the patients according to their diseases.
One of the most important components for effectively managing healthcare is for patients to be conscious about their health. In this context, health literacy holds great importance. With this aim in mind, at Albert Health, we have developed artificial intelligence-based websites where individuals can ask their health questions and obtain information on any aspect of their health to better understand, manage effectively, and lead a healthier life.
3. Albert and Large Language Models
Websites offer us the opportunity to address and support a wide audience by quickly reaching people in different geographies. However, it is inevitable that there will be incorrect, insufficient or incomplete information on the internet. Overcoming such problems, language models that can perform various functions on human language occupy a special place in this field of artificial intelligence.
Many applications are used and developed in the healthcare industry, such as virtual nurses that can communicate with patients and models that can keep clinical notes for doctors. Especially in this sector, these models, which enable patients to communicate interactively, have a critical role in patient communication and contribute to public health by increasing the accuracy and accessibility of health information.
As Albert Health, we eliminate the mobile application registration process and offer individuals with chronic diseases a tool to obtain reliable information anonymously, with a user-friendly interface and websites developed with an artificial intelligence language model. Then, we direct individuals who have gained awareness about their health to our mobile application so that they can manage their health effectively.
The main objective of these websites is to enable individuals to acquire information more effectively, reliably, and easily, and to provide them with an interactive experience in health-related topics. Developed by our AI engineers and validated by our medical team, these websites allow individuals to freely ask their questions anonymously at any time. Additionally, users are reminded in each session to consult a healthcare professional for the most accurate guidance when conversing with an artificial intelligence, thereby encouraging them to always access the most accurate information. Individuals who can obtain reliable information can make more informed decisions about their health and maintain more effective communication with healthcare professionals.
4. Generating a Prompt
Large Language Models (LLMs) is a general term given to machine learning models that understand complex language structures, handle various text tasks, and generate human-like text.
Generative Pre-trained Transformer (GPT) is an advanced language model utilized to achieve predetermined objectives within its scope. This model is based on the Transformer architecture, which forms the foundation of language processing models, and it has been pre-trained on a large dataset.
Prompt is a term used in LLMs. It is typically a text provided as an initial input to the model, containing instructions or guidelines for it to perform a specific task. This text enables the model to generate the desired output and accomplish a particular task.
Users convey their questions to GPT via a prompt specifically developed by Albert Health. The instructions contained within this prompt are designed using a technique called prompt engineering. This technique ensures that the responses meet a certain quality and standard. The developed system was initially tested using a dataset containing frequently asked questions from within the mobile application and questions about diseases. The responses provided by GPT to the test dataset were recorded. Subsequently, our medical team examined the responses, and it was observed that the developed GPT prompt effectively provided responses to user needs, demonstrating high performance in real-life scenarios.
Following the validation conducted by our medical team, the quality of responses has been evaluated in terms of reliability, comprehensibility, medical accuracy, and risk. It has been observed that the system meets these criteria adequately. Subsequently, the website has been made available to users. After its launch, the quality and consistency of responses are monitored, and any anomalies detected are addressed promptly by revising the system if necessary.
Through the developed prompts, it is of great importance, especially in health-focused and artificial intelligence-supported website projects, to verify and evaluate the accuracy and clarity of every response generated by GPT. In this context, regular review of responses aims to ensure users’ continuous access to reliable information. Throughout the project development process, the use of real user data and its validation by our medical team, prioritizing data security and privacy principles in the information gathering and processing processes, maintaining user anonymity, and continuously observing response performance for improvements are crucial.
5. Diabetes Website
The Republic of Turkey Ministry of Health is meticulously working to control the financial burden of diabetes on the healthcare system in our country. According to the “Sixth Diabetes Atlas” report published by the International Diabetes Federation (IDF) in 2013, Turkey’s diabetic population will rank among the top 10 countries with the highest diabetic population by the year 2035. Globally, it is expected that by 2035, there will be 592 million people with diabetes [1]. Therefore, it has been decided to create a website to increase access to information about diabetes and raise awareness among the public. This website aims to enhance the public’s awareness of health by providing information obtained from reliable sources on the prevention, treatment, and management of diabetes.
Individuals with diabetes often struggle to access fast, accurate, and reliable information. Therefore, we aim to support those living with diabetes by providing access to a reliable source of information through the first website we have developed specifically for diabetes. Users can access the Albert Diabetes website through an interface created to submit their questions. They can receive responses generated by GPT using an enhanced prompt. This way, a platform is provided for individuals with diabetes to quickly access reliable information about their condition and effectively manage their diabetes.
The website can be accessed through this link. It supports both Turkish and English languages.
6. Conclusion
Albert Diabetes website aims to provide fast and reliable information without adding layers such as signing up for a mobile application. This approach allows users to easily find answers to their diabetes-related questions and access health information more quickly. GPT is quite successful in providing answers to questions specific to diabetes.
Analyses conducted to evaluate the effectiveness of the Albert Diabetes website indicate that the platform has been successfully embraced by users. According to the data obtained by examining the responses provided by GPT, it has been determined that a total of 347 different sessions occurred, with 795 requests made. On average, 2.29 interactions took place in each session. Thus, we can observe that the platform is actively used.
However, while the Albert Diabetes website aims to provide a fast and effective user experience, it may encounter some challenges. For instance, when addressing sensitive health-related issues, it is crucial to maintain red lines such as user privacy and security. Even if no information is collected from the user, care should be taken not to share the user’s personal information when posing questions to the model.
However, the Albert platform plans to expand in the near future with a new website providing information on sexually transmitted diseases. This site aims to prioritize privacy, allowing users to find answers to their health-related questions while protecting their privacy. Additionally, it directs users to the Albert mobile application to effectively manage their health.
Albert makes a significant contribution to the digital health sector with its innovative and technology-driven approach. Observing how AI-powered websites transform interaction and information sharing in the healthcare field is quite exciting for the future of the healthcare system.
Hope to meet again in our next writings…
References
[1] IDF Diabetes Atlas. (n.d.). Copyright © IDF Diabetes Atlas 2024. All Rights Reserved. https://diabetesatlas.org/
