Clinical Decision Support System (CDSS)

We set out to build something that a patient in rural Maharashtra could use to understand their blood report without waiting days for a doctor's appointment. What we ended up with is a three-layer ML system that classifies infection risk, distinguishes five types of anemia, flags organ stress across eleven body systems, and delivers the results in both English and Hindi — all from a single CBC report. The numbers hold up: 82% accuracy on real patient infection data, 99.5% on anemia classification, and organ stress accuracy ranging from 78% to 98%. But the result we are most proud of is simpler than any of those metrics — the system never once confused a Thalassemia patient for an Iron Deficiency patient. In a clinical context, that distinction matters more than overall accuracy. The system is live at cdss.symbianslab.com. The code is open. Future work will focus on collecting Indian patient data to improve Layer 1's regional validity, adding medication effect modeling, and building a more locally relevant nutritional database for the prescription engine.

[1] B. Bhasuran, Q. Jin, A. Deville, Y. Wu, K. Hanna, Z. Lu, and Z. He, "LabQAR: A Manually Curated Dataset for Question Answering on Laboratory Test Reference Ranges and Interpretation," medRxiv, 2025, doi: 10.1101/2025.06.03.25328882. [2] F. Cabitza et al., "Development, evaluation, and validation of machine learning models for COVID-19 detection based on routine blood tests," Clin. Chem. Lab. Med., vol. 59, no. 2, pp. 421–431, 2020, doi: 10.1515/cclm-2020-1294. [3] M. Zuin, G. Rigatelli, C. Bilato, M. Correale, and G. Zuliani, "Prediction of SARS-CoV-2-positivity from million-scale complete blood counts using machine learning," Commun. Med., vol. 2, no. 1, p. 60, 2022, doi: 10.1038/s43856-022-00129-0. [4] J. Parraga-Alava, M. Dorn, and J. Meza, "CBCovid19EC: A dataset Complete Blood Count and PCR test for COVID-19 detection in Ecuadorian population," Data in Brief, vol. 47, p. 109016, 2023, doi: 10.1016/j.dib.2023.109016. [5] W. C. Mentzer, "Differentiation of iron deficiency from thalassaemia trait," Lancet, vol. 1, no. 7808, p. 882, 1973. [6] J. E. May, M. B. Marques, V. V. B. Reddy, and R. Gangaraju, "Three neglected numbers in the CBC: the RDW, MPV, and NRBC count," Cleve. Clin. J. Med., vol. 86, no. 3, pp. 167–172, 2019, doi: 10.3949/ccjm.86a.18072. [7] World Health Organization, "Haemoglobin concentrations for the diagnosis of anaemia and assessment of severity," WHO Vitamin and Mineral Nutrition Information System, Geneva, Switzerland, 2011. [8] A. L. Goldberger et al., "PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals," Circulation, vol. 101, no. 23, pp. e215–e220, 2000. [9] A. E. W. Johnson et al., "MIMIC-III, a freely accessible critical care database," Sci. Data, vol. 3, p. 160035, 2016. [10] M. A. Reyna et al., "Early Prediction of Sepsis from Clinical Data: The PhysioNet/Computing in Cardiology Challenge 2019," Crit. Care Med., vol. 48, no. 2, pp. 210–217, 2020. [11] United States Department of Agriculture, "FoodData Central," Agricultural Research Service. [Accessed: Apr. 8, 2026]. [Online]. Available: https://fdc.nal.usda.gov. [12] S. Tiangolo, "FastAPI: Modern, fast, web framework for building APIs with Python." [Accessed: Apr. 8, 2026]. [Online]. Available: https://fastapi.tiangolo.com. [13] F. Pedregosa et al., "Scikit-learn: Machine learning in Python," J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011. [14] Thalassemia International Federation, Guidelines for the Management of Transfusion Dependent Thalassaemia, 3rd ed. Nicosia, Cyprus: TIF, 2014. [15] T. Singhal, "A Review of Coronavirus Disease-2019 (COVID-19)," Indian J. Pediatr., vol. 87, no. 4, pp. 281–286, 2020.