Development and External Validation of Logistic Regression Models Integrating Conventional Complete Blood Count and...

Differentiating dengue from other acute febrile illnesses during initial clinical evaluation remains challenging because of the often-nonspecific nature of presenting manifestations. In this study, logistic regression models were developed and externally validated to predict reverse transcription polymerase chain reaction (RT-PCR)-confirmed dengue by integrating conventional complete blood count (CBC) parameters and cell population data from automated hematology analyzers.

This multi-stage retrospective diagnostic accuracy study analyzed conventional CBC parameters and leukocyte-related cell population data (CPD) features for predicting RT-PCR-confirmed dengue. Models were developed in clinically suspected cases based on RT-PCR results from a single hospital and subsequently evaluated using internal validation and three external settings. Regression coefficients and classification cutoff values were fixed based on the training cohort.

In total, 1689 RT-PCR-tested clinical specimens were analyzed (426 dengue-positive and 1263 dengue-negative cases). Compared with RT-PCR-negative febrile controls, dengue-positive cases exhibited lower white blood cell and platelet counts, along with a higher monocyte distribution width (all p < 0.001). A parsimonious primary model based on four predictors achieved an area under the curve (AUC) of 0.86 in the training cohort and 0.85 in internal validation. In the independent hospital cohort, the model maintained strong discrimination (AUC = 0.85), with high sensitivity and negative predictive value using the prespecified training-derived cutoff. Predicted-positive rates remained low in the healthy cohort.

Logistic regression models integrating conventional CBC and leukocyte-related CPD features demonstrated stable performance across validation settings and may facilitate dengue risk stratification using routinely available laboratory data.