Prospective multicenter study finds 46.4% eight-week hemoglobin response to Zaizao Shengxue capsule

In a prospective multicenter real-world study of adults receiving Zaizao Shengxue capsule for anemia, 204 of 440 patients in the analysis cohort achieved the predefined hemoglobin response after eight weeks. The overall response rate was 46.4%, while an Extra Trees model using anemia etiology, age and estimated glomerular filtration rate achieved an area under the receiver operating characteristic curve of 0.769 (95% CI, 0.670–0.860) in four external validation centers.

The observational study was conducted at 14 centers in China from May 2023 through December 2025. Investigators prospectively approached eligible patients treated in routine clinical practice and recorded baseline characteristics, follow-up findings, concomitant treatments, adherence and adverse events in a standardized electronic data capture system. Participants were 18 to 80 years old and met the study definition of anemia: hemoglobin below 120 g/L in men or below 110 g/L in non-pregnant women.

Patients received a commercially manufactured preparation at five capsules orally three times daily for eight consecutive weeks. Assessments were scheduled at weeks 2, 4, 6 and 8. Response was defined as an increase in hemoglobin of at least 20 g/L, equivalent to 2 g/dL, from baseline to week 8. Of 589 enrolled patients, 77 lacked an evaluable week-8 hemoglobin measurement. A further 72 patients received additional interventions expected to affect hemoglobin, such as antitumor therapy, iron or other anemia-directed treatment, transfusion or erythropoiesis-directed therapy, and were excluded. The resulting analysis cohort included 440 patients.

For prediction-model development, 338 patients from 10 centers were divided by stratified random sampling into a training cohort of 238 and an internal validation cohort of 100. Another 102 patients from four independent centers formed the external validation cohort. Those external patients were not used for feature selection, model fitting, hyperparameter tuning or algorithm selection. Investigators initially reviewed 32 pretreatment variables covering demographics, anemia characteristics, comorbidities, prior anemia treatment, transfusion history and routine laboratory measurements; correlation and collinearity screening reduced the field to 24 candidate predictors.

Five feature-selection methods were applied, and variables selected by at least four were retained. Anemia etiology was chosen by all five methods, while age and estimated glomerular filtration rate, or eGFR, were each selected by four. Ten algorithms were then trained and compared, including decision trees, XGBoost, support vector machines, artificial neural networks, linear discriminant analysis, Gaussian naive Bayes, logistic regression, LightGBM, CatBoost and Extra Trees. Missing continuous and categorical predictor values were imputed with median and mode values derived from the training cohort, but week-8 hemoglobin values were not imputed.

Extra Trees was selected after internal validation and locked before evaluation in the external cohort. Its internal-validation AUC was 0.821 (95% CI, 0.729–0.902), compared with 0.769 (95% CI, 0.670–0.860) in external validation. At a probability threshold of 0.50, external-validation sensitivity was 0.551 and specificity was 0.811, while average precision was 0.720. Calibration was closer to the ideal intercept of 0 and slope of 1 internally than externally; the external calibration intercept was 0.435 and the slope was 0.850, indicating differences between predicted and observed response probabilities. Exploratory pairwise DeLong comparisons in the internal cohort found no statistically significant AUC differences between Extra Trees and the other algorithms, and the comparisons were not adjusted for multiple testing.

Observed eight-week response rates varied by the physician-assigned cause of anemia: 24.0% in cancer-related anemia, 74.4% in iron-deficiency anemia and 77.0% in anemia of chronic disease. Grouped SHAP analysis identified anemia etiology as the largest contributor to Extra Trees predictions, followed by age and eGFR. SHAP dependence plots showed separation among the three etiologies and nonlinear contribution patterns for age and eGFR. These analyses described how the model generated predictions; they did not establish causal effects of the predictors, Zaizao Shengxue capsule or other treatments.

The study has several limitations. Excluding patients without an evaluable week-8 hemoglobin measurement may have introduced selection bias, while excluding those receiving other interventions that could affect hemoglobin limits applicability to patients who need concurrent care. Anemia etiology was based on routine clinical diagnoses and may not capture overlapping mechanisms; detailed iron indices, inflammatory markers and nutritional measures were not uniformly available. The response endpoint also depended partly on baseline hemoglobin. In addition, the external cohort was relatively small, between-center heterogeneity may have affected performance estimates, and calibration was imperfect.

All analyzed patients received Zaizao Shengxue capsule, and the study included neither an untreated group nor an active comparator. The results therefore describe observed response under real-world care and support prediction, not comparative effectiveness or an individualized estimate of the capsule’s causal treatment benefit. The external-validation findings provide evidence that the model can predict response among similar patients using three routinely available baseline variables. The authors concluded that larger prospective evaluation is needed to establish its clinical utility.

Subject of Research: Prediction of eight-week hemoglobin response to Zaizao Shengxue capsule in adults with anemia

Article Title: Zaizao Shengxue capsule for anemia: a prospective multicenter real-world study of hemoglobin response and predictive modeling

Article References: Xing, F., Wu, J., Cheng, C., Chen, S., Qin, Z., Wang, Y., Zhang, Y., Tang, J., & Li, X. (2026). Zaizao Shengxue capsule for anemia: a prospective multicenter real-world study of hemoglobin response and predictive modeling. Frontiers in Pharmacology, 17, Article 1957386. https://doi.org/10.3389/fphar.2026.1957386

Image Credits: AI Generated

DOI: 10.3389/fphar.2026.1957386

Keywords: Zaizao Shengxue capsule, anemia, hemoglobin response, prospective multicenter study, real-world evidence, Extra Trees, prediction model, external validation, traditional Chinese medicine, Zaizao, Shengxue, capsule