HealthSurge AI: An Agentic Deep Learning Framework for Predictive Patient Influx Management in Indian Hospitals

We began this project when an administrator in Pune truthfully told us how staffing decisions were made: experience, she said. HealthSurge AI is a second tool for her — one that links public-domain data to a supervisor's mobile phone in a way that makes timely response possible. After two years of real data from three hospitals in Maharashtra, it achieved a 48-hour MAPE of 8.3 percent and detected every real surge in the test period before it peaked. An operational trial gave early signs that the forecast leads to earlier staff deployment and significantly shorter patient queues. The H1 nursing supervisors said they preferred the recommendations panel to the forecast chart. There is still work to do: more sites, more states, another season of controlled testing. We believe it is a direction worth exploring and hope this work can help those who follow.

[1] World Health Organization, "Health workforce requirements for universal health coverage," WHO Tech. Rep., Geneva, 2016. [2] M. Jain, A. Sharma, and R. Gupta, "Managing mass casualty events at the Kumbh Mela," J. Public Health, vol. 42, no. 3, pp. 612-621, 2020. [3] S. Nair, P. Tripathi, and V. Kumar, "Air quality and emergency room admissions for respiratory illness in Delhi," Indian J. Med. Res., vol. 152, no. 4, pp. 378-387, 2020. [4] World Health Organization, "Global Health Estimates 2020," WHO, Geneva, 2020. [5] S. L. Bernstein et al., "The effect of emergency department crowding on clinically oriented outcomes," Acad. Emerg. Med., vol. 16, no. 1, pp. 1-10, 2009. [6] S. Hochreiter and J. Schmidhuber, "Long short-term memory," Neural Comput., vol. 9, no. 8, pp. 1735-1780, 1997. [7] N. R. Hoot et al., "Forecasting emergency department crowding: A discrete event simulation," Ann. Emerg. Med., vol. 52, no. 2, pp. 116-125, 2008. [8] S. Halachmi et al., "Machine learning-based model for bed management in a general hospital," J. Med. Syst., vol. 42, no. 12, p. 255, 2018. [9] E. Choi et al., "Doctor AI: Predicting clinical events via recurrent neural networks," Proc. MLHC, PMLR vol. 56, pp. 301-318, 2016. [10] Z. C. Lipton, D. C. Kale, C. Elkan, and R. Wetzel, "Learning to diagnose with LSTM recurrent neural networks," Proc. ICLR, 2016.