A Boruta-LASSO machine learning approach identified age, RDW, and HbA1c as independent predictors of CAD in LTBI patients, with a significant RDW-LTBI interaction associated with a 1.737-fold increase in CAD risk, suggesting LTBI-associated chronic inflammation may contribute to CAD risk.
Key Findings
Results
A nomogram incorporating age, RDW, and HbA1c predicted CAD in LTBI-positive individuals with an AUC of 0.783 in the derivation cohort and 0.738 in the external validation cohort.
616 participants were enrolled from a dual-center retrospective case-control study after screening 862 records.
Derivation cohort: n = 453; external validation cohort: n = 163.
Mean absolute error (MAE) was 0.285 in the derivation cohort and 0.303 in the external validation cohort.
The model demonstrated favorable net clinical benefit on decision curve analysis.
Candidate predictors were selected using a two-step Boruta-LASSO machine learning approach before incorporation into multivariable logistic regression.
Results
A significant multiplicative interaction between RDW and LTBI status was associated with a 1.737-fold increase in CAD risk.
The RDW-LTBI interaction term had an odds ratio of 1.737 (95% CI 1.092–2.877, P = 0.025).
This interaction remained significant after statistical testing.
The finding suggests that LTBI modifies the relationship between RDW and CAD risk.
Patients underwent both coronary CT angiography (CCTA) for CAD diagnosis and interferon-γ release assay (IGRA) for LTBI identification.
Results
RDW was positively correlated with multiple pro-inflammatory markers specifically in LTBI-positive patients.
RDW was positively correlated with interleukin-6 (IL-6; r = 0.345).
RDW was positively correlated with C-reactive protein (CRP; r = 0.251).
RDW was positively correlated with serum amyloid A (SAA; r = 0.238).
RDW was positively correlated with monocyte-to-lymphocyte ratio (MLR; r = 0.214).
All correlations survived Benjamini-Hochberg correction for multiple comparisons at FDR < 0.05.
Results
Age, RDW, and HbA1c were identified as independent predictors of CAD in individuals with LTBI.
These three variables were selected through the Boruta feature selection algorithm followed by LASSO regression.
All three predictors were incorporated into the final multivariable logistic regression nomogram.
Traditional lipid-based risk factors were not identified as independent predictors in this LTBI-affected population.
The authors note these are routine clinical variables that could facilitate risk stratification in resource-limited settings.
Methods
The study used a dual-center retrospective case-control design enrolling patients who underwent both CCTA and IGRA testing.
862 records were screened, with 616 participants ultimately enrolled.
Cases were patients with CAD diagnosed by CCTA; controls were those without CAD.
The derivation cohort (n = 453) came from one center and the validation cohort (n = 163) from a second center.
Both coronary CT angiography and interferon-γ release assay were required for enrollment.
What This Means
This research suggests that a simple three-variable tool using age, a routine blood test result called red blood cell distribution width (RDW), and a measure of blood sugar control (HbA1c) can help predict which people with latent tuberculosis infection (LTBI) — meaning they carry the TB bacteria but are not actively sick — are at higher risk for coronary artery disease (CAD), which involves narrowing of the heart's arteries. The researchers used advanced machine learning methods to identify these predictors from medical records of 616 patients at two hospitals in China, and the prediction tool performed reasonably well both in the group used to build it and in a separate group used to test it.
A particularly notable finding is that RDW — a standard measure of variation in red blood cell size that is included in routine blood counts — appears to interact specifically with LTBI status to raise CAD risk by approximately 1.7-fold. In LTBI-positive patients, higher RDW was also linked to higher levels of several inflammation markers in the blood, including IL-6, CRP, and serum amyloid A. This pattern is consistent with the idea that the low-grade, persistent inflammation caused by latent TB infection may promote heart disease through mechanisms that are not well captured by traditional heart disease risk factors like cholesterol levels.
This research suggests that people with latent tuberculosis — a condition affecting roughly a quarter of the world's population — may benefit from cardiovascular risk screening using inexpensive, widely available blood tests. The findings could be especially relevant in regions where TB is common and resources for advanced cardiac testing are limited, potentially allowing earlier identification of those at higher risk for heart disease. However, because this was a retrospective study at two centers, larger prospective studies would be needed to confirm these findings before clinical implementation.
Sun F, Zhou Y, Ma H, Ye M, Yang Y. (2026). Red blood cell distribution width is associated with coronary artery disease in latent tuberculosis infection: a machine learning-based study.. Frontiers in cellular and infection microbiology. https://doi.org/10.3389/fcimb.2026.1836144