AI system may accurately diagnose Sjögren’s using routine lab tests: Study
Publicly available tool could be practical for clinical use, researchers say
Written by |
A new publicly available artificial intelligence (AI) tool that analyzes results from routine laboratory blood tests may improve the diagnosis of Sjögren’s disease, according to a multicenter study in China.
The AI system, called Sjogren Multi-criterion Feature Integration Framework (SMFIF), distinguished people with Sjögren’s from those with Sjögren’s-like manifestations but no Sjögren’s with an accuracy of up to 96%. The tool also effectively distinguished Sjögren’s patients from people with other autoimmune disorders.
“SMFIF is publicly available and provides probabilities of [Sjögren’s] based on laboratory data, offering a practical diagnostic tool for clinical use,” researchers wrote.
Details of the AI tool’s development and predictive performance were described in the study “A multi-criterion feature integration framework for accurate diagnosis of Sjögren’s disease using routine laboratory tests,” published in npj Digital Medicine.
Diagnosing Sjögren’s can be challenging
Sjögren’s is an autoimmune disease marked by damage to moisture-producing glands, including those that produce tears and saliva. Although dryness of the eyes and/or the mouth is the most common symptom of the disorder, it can also affect other tissues, leading to a persistent dry cough, prolonged fatigue, and chronic pain.
Because the exact cause of Sjögren’s is unknown and its symptoms mimic those of other conditions, diagnosis can be challenging. Moreover, not all symptoms appear at the same time, so clinicians may treat them individually without recognizing a more widespread condition.
A common feature among most people with Sjögren’s is the presence of self-reactive antibodies, specifically anti-Ro (SS-A) and anti-La (SS-B). Still, 1 in 4 Sjögren’s patients lack these antibodies, and they are not specific to Sjögren’s, as they have been found in other autoimmune conditions.
Therefore, a definitive diagnosis currently relies on several tests, including invasive biopsy of the tear or salivary glands to look for patterns of inflammatory cells and the severity of inflammation.
“The lack of a single, highly specific and sensitive biomarker … contributes significantly to the challenges in diagnosis, leading to underdiagnosis and misdiagnosis,” the researchers wrote.
Researchers developed machine learning tool
In this study, a team of researchers in China conducted a study (NCT06982482) to develop a machine learning tool, a type of AI that learns from data and identifies patterns, which could identify Sjögren’s-specific patterns in routine laboratory blood tests.
The team collected lab data from 34,958 adults across three hospitals in China. Participants included Sjögren’s patients, as well as a control group of sex- and age-matched adults who showed Sjögren-like manifestations but didn’t have Sjögren’s.
Laboratory data were collected from one month before diagnosis to any treatment. The new model, SMFIF, was developed based on 16 blood markers identified as the most relevant for Sjögren’s. These included markers of inflammation, kidney function, liver function, blood sugar (glucose), cardiovascular disease, and bone health.
To train the model, the team used results from nearly 10,000 Sjögren’s patients and 8,000 controls on those 16 core features. They then tested and validated the tool across three distinct groups of participants.
AI system outperformed traditional biomarkers
SMFIF’s ability to distinguish Sjögren’s from Sjögren’s-like manifestations (without Sjögren’s) or other autoimmune diseases was assessed using the area under the curve, or AUC, with values closer to 1.0 indicating better performance.
The tool’s AUC for the test set, which included nearly 3,500 Sjögren’s patients and 4,000 controls, was 0.923. A similar AUC was obtained for an internal validation set (from one hospital) of about 4,200 patients and 5,100 controls.
When using data from an external validation set (from two hospitals) comprising around 140 Sjögren’s patients and 400 controls, SMFIF’s had an AUC of 0.964. However, results showed a lower degree of consistency relative to the other two sets, suggesting that “further refinement may be needed to enhance sensitivity when applied to different external conditions,” the team wrote.
This model provides a low-cost, accessible and accurate diagnostic tool for [Sjögren’s].
Notably, SMFIF outperformed the results of traditional biomarkers, including self-reactive antibodies like SS-A, SS-B, and ANA, as well as other autoimmune biomarkers such as rheumatoid factor, immunoglobulin G, C4, C3, and globulin.
“SMFIF based on 16 key features of routine laboratory items achieved satisfactory and stable performance and was significantly better than the traditional markers SSA/Ro, SSB/La and ANA,” the researchers wrote.
Moreover, SMFIF distinguished Sjögren’s with elevated accuracy from other autoimmune conditions, including rheumatoid arthritis (AUC: 0.834), lupus (AUC: 0.890), scleroderma (AUC: 0.842), and osteoarthritis (AUC: 0.915).
“The clinical utility of SMFIF depends on its ability to distinguish [Sjögren’s] not only from healthy individuals but, more importantly, from patients with other immune-mediated and chronic conditions,” the team wrote. “This model provides a low-cost, accessible and accurate diagnostic tool for [Sjögren’s],” however, studies following patients over time and “data sets from other regions are needed to further confirm the feasibility and generalizability of SMFIF.”
