Project Description

In the age of big data analytics, one must consider the continuum from predictive to prescriptive analytics to help managers to improve their day-to-day operations in large-scale healthcare systems. These systems often run under uncertainties and in rapidly changing environments. Good prescriptive management solutions require building high-fidelity models that are adaptive to the changing environment. Consequently, a framework for learning stochastic models from data in this setting is imperative. These learnt models need to be seamlessly integrated with data-driven prescriptive methods to optimize system operations.

In this research project, we work closely with the largest hospital systems in the state of Indiana and propose a methodological framework that collaboratively leverages deep learning and stochastic process theory to revolutionize workload prediction and resource planning, such as capacity and staffing. These developments are expected to enable fundamental improvements in short-term and long-term operations for healthcare delivery. Our research agenda, in support of this broader goal, includes (i) a novel framework for inferring stochastic models of time-varying, large-scale healthcare systems; (ii) a robust resource allocation framework that accounts for model uncertainty and natural stochastic variation; (iii) integration and deployment of our algorithms into all 16 hospitals belonging to the collaborating healthcare system.

Start Date

05/01/2021

Postdoc Qualifications

Applicants hold (or are about to complete) a PhD in Operations Research, Industrial Engineering, Applied Mathematics, Electrical Engineering or a related discipline. A strong background in stochastic modeling and optimization methods is required. Research experience in statistics/machine learning/deep learning would be a great advantage. A willingness to learn the fundamental theory and methods of statistics/machine learning/deep learning is necessary. 
Programming skills, fluency in English and excellent communication and presentation skills are essential. 

Co-Advisors

Harsha Honnappa, honnappa@purdue.edu, School of Industrial Engineering, engineering.purdue.edu/SSL

Pengyi Shi, shi178@purdue.edu, Krannert School of Business, https://web.ics.purdue.edu/~shi178

References

“Estimating Stochastic Poisson Intensities Using Deep Latent Models”, R. Wang, P.
Jaiswal and H. Honnappa, Proceedings of the Winter Simulation Conference (2020).

“Timing it Right: Balancing Inpatient Congestion versus Readmission Risk at Discharge," P. Shi, J. E. Helm, J. Deglise-Hawkinson, and J. Pan. Operations Research, forthcoming

“The ∆(i)/GI/1 Queueing Model, and its Fluid and Diffusion Approximations”, H. Honnappa , R. Jain and A. R. Ward, Queueing Systems: Theory and Applications, 80.1-2 (2015): 71-103.

"Inpatient Bed Overflow: An Approximate Dynamic Programming Approach," J. G. Dai, P. Shi
I
Manufacturing and Service Operations Management. 2019; 21(4):894-911.