Diagnosing diabetes based on traditional criteria has increasingly become unreliable.
A recent study used a systematic review to identify specific algorithms that can be used to type diabetes based on patient health information. The research, published in Diabetes Research and Clinical Practice, identified the best algorithms to type diabetes using electronic health records (EHR) and clinical and administrative data.1
According to researchers, over the past 20 years, extensive diabetes registries and comprehensive administrative and clinical databases have been a gold mine for population-based research.1 This data has the potential to improve care and health service planning. “This is particularly crucial for conditions like young-onset diabetes, where the low prevalence makes it challenging to gather sufficient data for meaningful outcomes,” the authors wrote.1
Diagnosing diabetes based on traditional criteria has increasingly become unreliable. More than half of new type 1 cases are being diagnosed in adults and type 2 is rising among young people, the authors noted. Information from EHR, administrative, and clinical databases could improve diabetes diagnosis and treatment. EHR can enhance surveillance and help accurately type diabetes in patients, improving health outcomes.1 While there have been several efforts to obtain this data, no systematic review has been completed until now, according to the authors.
Researchers performed a systematic review of the literature and selected 19 qualifying studies with low bias from EMBASE and MEDLINE published between January 2000 and January 2023. The studies examined algorithm performance in identifying type 1 and type 2 diabetes. Researchers evaluated studies based on their ability to define diabetes types accurately by using diagnostic metrics against a variety of reference standards. They assessed study quality using the Quality Assessment of Diagnostic Accuracy Studies.
Insulin use data was highly accurate in identifying both diabetes types.1 However, researchers noted that algorithms relying on oral hypoglycemic agent (OHAs) prescriptions were less likely to accurately type diabetes. The authors said that algorithms worked better for younger patients at the time of diagnosis and that typing accuracy decreased with increasing age. However, “single-criterion algorithms based on age at diagnosis or prescription of OHAs, generally perform poorly,” researchers said.1
Algorithms based on multiple diagnostic codes worked best to predict type 1 versus type 2 diabetes using health data.1 This included calculating code ratios to enhance typing. Encounter diabetes diagnostic codes, such as codes from ICD-10-CM and ICD-9-CM, were essential in this model’s success rate. However, diagnostic code accuracy depended on variables such as medical record documentation quality and specific coding guidelines.1
In addition to the use of multiple codes, “Approaches with more than one criterion may also increase sensitivity in distinguishing diabetes type,” researchers said.1 Most of the top 10 best-performing algorithms had used multiple criteria.
When diabetes diagnostic codes weren’t available, self-reported diabetes type, either alone or with other predictors, outperformed alternative approaches.1 However, researchers said self-reported data is not typically available in EHRs or administrative databases.
Machine learning (ML) algorithms played a small role in the study, but demonstrated potential in reducing false positives and false negatives. The authors noted that combining rule-based algorithms, clinical guidelines, and ML could be a future direction for improved diabetes type classification.
For study limitations, most data came from high-income countries, so the results may not apply to low-income countries. In addition, the authors noted that there were concerns about the reference standards used, highlighting the necessity of more standardized criteria in diabetes typing. Another limitation was that “a critical shortcoming in the literature is the scarcity of studies attempting to externally validate previously-published algorithms.”1
The authors emphasized that their research lays the foundation for more accurate diabetes typing and treatment strategies.
“The results of this review demonstrate that with use of EHR-based data, the presence of multiple diabetes diagnostic codes is one of the most readily-available and accurate predictors for identifying diabetes type,” the researchers concluded.1 “As technology and our ability to analyze big data continue to evolve and improve, ML algorithms with or without human expertise will be expected to become an important approach for accurately distinguishing diabetes type.”