Objective
The objective of this proposal is to use a machine learning (ML) model deployed against electronic medical record (EMR) data to identify type 1 diabetes (T1D) patients potentially misdiagnosed as type 2 diabetes (T2D) patients and raise clinician awareness where antibody testing may be appropriate when clinical risk factors exist. A clinical decision support tool will be implemented within clinical workflows to be actioned when reviewed by a clinician (physician, nurse, pharmacist, etc.). The identification of these patients at risk for T1D should raise awareness for clinicians to review patient history and consider the antibody testing.
Background Rationale
JDRF and IQVIA previously partnered to develop a machine learning algorithm to identify adult-onset Type 1 Diabetes (T1D) patients who have been misdiagnosed with Type 2 Diabetes (T2D), referred to as Adult T1D Detection. A product of this collaboration is a predictive model that identifies a T2D patient clinical profile suggestive of having T1D. Specifically, a patient with a diagnosis of T2D is more likely to be a true T1D patient, i.e., misdiagnosed with type 2, if they are younger, have lower weight and body mass index (BMI), and have fewer visits for hypertension (among hundreds of other predictors).
JDRF and IQVIA are looking to partner with HealthShare Exchange (HSX), an aggregator of health system data, to validate the predictive model.
The validation will have multiple steps including conducting a retrospective validation, localizing the algorithm, designing a process for clinical workflow integration, piloting the solution as a part of a prospective interventional validation, and deploying the solution into clinical care.
Description of Project
Type 1 Diabetes (T1D) was once thought to be a childhood onset disease with strong association to family history. It is now estimated that 90% of T1D patients do not have a family history of the disease, and almost half of newly diagnosed cases occur in adults. Nearly 40% of those adults with T1D are initially misdiagnosed with Type 2 Diabetes (T2D), leading to delays in treatment and risk for complications such as diabetic ketoacidosis.
Advances in data collection through electronic medical records (EMRs) have made it possible to apply machine learning (ML) statistical analysis to large anonymized clinical data sets and identify patterns and predictors of patients likely to have a specific disease.
In collaboration with the Juvenile Diabetes Research Foundation (JDRF), IQVIA has created a ML predictive model to identify T2D patients who should be evaluated for T1D based on their clinical history. The recommended testing for T1D is a blood test that looks for antibodies against insulin. The presence of these antibodies in the blood is associated to patients with T1D.
IQVIA and HealthShare Exchange (HSX) propose a collaboration to identify three healthcare organizations within the HSX footprint to validate this predictive model in clinical workflows. The anticipated outcome of this deployment is to identify T2D patients who should be evaluated for T1D and have the antibody blood test.
IQVIA and HSX will work with HSX healthcare partners to integrate the predictive model into their respective EMRs and engage clinical team members (physicians, nurses, pharmacists, etc.) to access the results of the model and review patient history for appropriateness of the antibody testing. When ordered, antibody results will be reviewed by the physician and appropriate treatment plan changes will be addressed.
Anticipated Outcome
We anticipate increased clinician awareness through identification of patients who should be evaluated for T1D and have the antibody blood test.
Through this pilot, we hope to identify best practices for patient screening and EMR integration. These learnings will be shared with JDRF and used to establish a blueprint for deploying this screening tool at scale into additional health systems in the future.
We anticipate the learnings from this project will also be used to inform JDRF and other T1D clinical practice standard authors to assess the clinical benefit of this predictive model and inform review/revision of current guidelines for T1D screening.
Relevance to T1D
Adult-onset T1D can be misdiagnosed as T2D, adversely impacting patient outcomes and potentially delaying necessary treatment. This project seeks to identify T2D patients to be considered for re-evaluation of T1D and have bloodwork for the detection of insulin antibodies.