Objective
The main goal of this study is to measure where gaps currently exist in the care of people with Type 1 Diabetes (T1D) both before and after they are diagnosed. We also want to expand the utility of an existing AI tool so that it can spot undiagnosed T1D in both children and adults. The specific objectives are:
Objective 1: Identifying Gaps in Care and Inequities for T1D Patients Using Electronic Medical Record (EMR) Data
This aim will examine how well clinical guidelines for T1D care are adhered to for different groups of people. We will consider factors such as:
• Where people live (e.g., rural vs. urban areas)
• Racial and ethnic backgrounds
• Gender
• Types of healthcare organizations (e.g., small community health centers vs. large hospitals, availability of pediatric endocrinologists, etc.)
We will evaluate how well the care of these groups follows the American Diabetes Association (ADA) guidelines for managing T1D, along with other recent clinical measures. We will also examine social determinants of health (SDoH) such as income, insurance status, and access to healthcare to understand how these contribute to differences in care and outcomes. This will help us identify not only where care is lacking, but also opportunities to improve early detection, intervention, and ongoing management, especially for historically underserved populations.
Objective 2: Expanding the Adult T1D Detection Algorithm
Breakthrough T1D and IQVIA have already developed an AI tool that uses EMR data to identify the risk of adult-onset T1D in patients initially diagnosed with Type 2 diabetes (T2D). This work will expand the use of this tool to see if it can identify undiagnosed T1D in people of all ages, regardless of their diabetes history. We will test the tool’s ability to detect early-stage T1D (before symptoms appear) as well as fully developed T1D in both children and adults. Our objective will be to update the tool and measure how well it works for different age groups and stages of T1D.
Background Rationale
The urgency to address Type 1 diabetes (T1D) as a public health issue has grown significantly in recent years due to its increasing prevalence and the consequences of delayed diagnosis. According to CDC data, about 352,000 children and teens in the U.S. have diabetes, and 304,000 of these cases are T1D. The increase in T1D among this population is worrying because early symptoms are often subtle and can be mistaken for less serious conditions. If children are diagnosed late, they are more likely to develop severe complications such as diabetic ketoacidosis (DKA), which can lead to emergency room visits and intensive care. This not only poses immediate health risks but can also lead to long-term complications like cognitive impairments and higher healthcare costs.
Getting diagnosed early in T1D development can lead to better outcomes for patients, including better preservation of insulin-producing cells and lower mortality rates. However, early detection is a challenge in underserved communities where access to specialized care and screening programs is limited. These areas often face systemic issues like poor healthcare infrastructure, low health literacy, and limited access to endocrinologists, which all contribute to delays in diagnosis.
AI tools can significantly improve early detection. Traditional diagnosis methods rely heavily on doctors’ interpretations of symptoms and test results, which can vary and be subjective. AI models apply the same criteria consistently, ensuring that subtle signs of T1D are not missed. As more patients are diagnosed and treated, these models can become more accurate by adapting to new clinical patterns and incorporating new diagnostic markers.
Health disparities in T1D care are significant for racial and ethnic minorities and people from low-income backgrounds. African Americans and Hispanic individuals often have less access to specialized diabetes care and face systemic barriers like inadequate insurance, limited primary care access, and geographic isolation. Socioeconomic status also plays a big role, as people from lower-income households are less likely to have regular follow-ups, education on self-management, or access to advanced diabetes technologies like continuous glucose monitors (CGMs) and insulin pumps. African American and Hispanic youth with T1D often have higher HbA1c levels (a measure of blood sugar control) compared to their white peers. This is partly because they have less access to advanced diabetes management tools, which are recommended by the ADA. Hispanic children and those with public insurance are also less likely to use CGMs compared to non-Hispanic, privately insured peers.
Managing T1D involves many aspects, including monitoring blood sugar, insulin therapy, regular screening for complications, and preventive care. Traditional methods of evaluating how well patients follow their care plans involve manual review of medical records which can be time-consuming, prone to errors, and biased. This study automates the process of tracking clinical practices and provides insights into how closely care aligns with established standards like those from the ADA. By identifying patients who have missed necessary tests or follow-up appointments, the algorithm can help healthcare organizations ensure that these patients receive the care they need.
Description of Project
Despite advances in the scientific community’s understanding of Type 1 diabetes (T1D), early and accurate diagnosis remains a significant challenge. This is especially true when trying to tell the difference between T1D and Type 2 diabetes (T2D) in adults, as well as recognizing T1D in people of different ages and stages of the disease. Breakthrough T1D and IQVIA have created an AI/ML (Artificial Intelligence/Machine Learning) tool that uses data from electronic medical records (EMR) to spot the risk of adult-onset T1D in patients who were first thought to have T2D. This algorithm has shown promising initial results in detecting T1D and is now being prospectively validated in three healthcare organizations.
The American Diabetes Association (ADA) has guidelines for managing T1D, but they aren’t always followed, especially in underserved areas and among vulnerable groups. Specifically, there are still big differences in how T1D is diagnosed and managed, particularly across different races, locations, and social conditions. These gaps are even more noticeable in children and underserved communities.
Our research aims to improve how we detect T1D using AI across all ages and stages of the disease, and to address gaps in care both before and after diagnosis. The project has two main goals: (1) Finding gaps and inequalities in care using EMR data to see how well T1D management guidelines are being followed, and (2) Expanding the existing AI/ML tool to include children, and testing its ability to detect early stages of the disease. This means refining the tool to find undiagnosed T1D in both children and adults, even in complex cases where the T1D patient’s symptoms look like T2D.
This project aims to shorten the time it takes to diagnose T1D and ensure that once diagnosed, patients get care that follows established guidelines. This will improve the short and long-term health of T1D patients and reduce their healthcare costs.
Anticipated Outcome
Using AI tools to detect Type 1 diabetes (T1D) early and ensure patient care is in line with clinical guidelines has two main benefits: (1) improving patient health and (2) reducing healthcare costs for both patients and healthcare organizations.
Early diagnosis of T1D leads to improved health outcomes and quality of life. Patients diagnosed early can manage their blood sugar levels more effectively, which means fewer health emergencies and better long-term health. This reduces their need for expensive medical interventions and helps them maintain their daily lifestyle, lowering the indirect costs of diabetes care. For example, using continuous glucose monitors (CGMs) early on can significantly lower HbA1C levels and average blood glucose levels.
Early detection of T1D also cuts down on healthcare costs. Delayed diagnosis often leads to poor blood sugar control and serious complications like kidney failure, nerve damage, and heart disease, which are very expensive to treat. As an example, the average cost for a hospital stay due to diabetic ketoacidosis is $29,981, with some cases costing up to $284,357. By diagnosing T1D earlier, we can prevent these costly hospital visits and save millions of dollars each year.
The goal of this project is to shorten the time it takes to diagnose T1D and ensure that once diagnosed, patients receive care that follows established guidelines. This will improve both short-term and long-term health outcomes for T1D patients and reduce their healthcare costs.
Relevance to T1D
Many studies have shown that AI and machine learning can help predict diabetes. Breakthrough T1D and IQVIA have created an Adult T1D Detection algorithm that uses electronic medical records (EMR) to identify the risk of adult-onset Type 1 diabetes (T1D) in patients who were diagnosed with Type 2 diabetes (T2D). This tool has performed well both at the initial T2D diagnosis and at any point after and is currently being prospectively validated in several US health systems.
While there are AI models focused on T2D, there are fewer AI tools for detecting T1D across all ages and regardless of previous diabetes history. This study aims to expand the Adult T1D Detection algorithm to identify T1D in people of all ages and stages of the disease. We will explore the feasibility of early-stage detection (Stages 1 and 2), even though these stages might have fewer data points and smaller sample sizes.
Research has shown that screening for autoantibodies can predict future T1D diagnosis in children, but routine screening is still uncommon. Between 14% to 34% of children with T1D experience delays in diagnosis, with the time from symptom onset to diagnosis ranging from 2 to 315 days. A major reason for these delays is a lack of awareness and knowledge about T1D symptoms. Many parents mistake early signs of T1D, like excessive thirst and frequent urination, for less serious conditions. Similarly, healthcare providers often do not recognize these symptoms quickly or attribute them to other illnesses.
Misdiagnosis is a common problem. Studies have shown that conditions like respiratory infections, gastrointestinal issues, and urinary tract infections are often mistaken for T1D symptoms. In some cases, children were not tested for blood glucose levels during their initial visits to primary care providers, which significantly delayed the correct diagnosis. Delays in getting appointments and inadequate follow-up procedures also contribute to late diagnoses. Many parents face challenges in securing timely appointments with doctors, leading to further delays. Addressing these diagnostic delays is crucial, especially in underserved populations. When evaluating adherence to clinical guidelines, we will consider social determinants of health (SDoH) to identify disparities in care. When expanding the Adult T1D Detection algorithm, we will focus on detecting biases related to age, gender, and race/ethnicity.