Machine learning and optimization can help solve a major problem with hospital-at-home care

The Block Center for Technology and Society recently hosted Professors Holly Wiberg and Peter Zhang to discuss distributed healthcare systems and supply chain optimization. Their emerging research, supported by a Block Seed Fund Award, explores the opportunities, operational challenges, and viability of scaling hospital-at-home (HaH) care in rural and underserved communities.

HaH offers patients the opportunity to receive acute-level care at home rather than in an inpatient bed, whether through early hospital discharge or direct admission from an emergency department. This care model has been shown to be clinically safe and desirable to patients, and it fits within a broader push to shift care “closer to home.” HaH programs have existed since 1999 and expanded significantly following COVID-19 pandemic investments by the Centers for Medicare & Medicaid Services, with roughly 400 approved hospital programs. However, widespread adoption and operationalization remain severely limited: delivering care across widely dispersed geographic areas creates complex logistical challenges around supply chain management, staffing constraints, cost, and fairness for both patients and providers.

“The clinical evidence for hospital at home has been strong for a long time, but operations are a major barrier,” says Zhang. “When everyone is in one facility, resources are already there to meet demand, and you benefit from pooling across patients. Once patients go home, you have to build delivery infrastructure to reach them, and you absorb much more uncertainty into your planning, which generally comes at a cost in efficiency. Our goal is to use optimization to design delivery models that provide safe care while keeping costs manageable.”

To address these challenges, Professors Wiberg and Zhang developed a machine learning and optimization framework that jointly determines patient suitability and supply fulfillment strategies. The model dynamically monitors patient trajectories to predict clinical and operational needs while identifying patients with the lowest medical risk who are most feasible to treat at home. By integrating predictive patient modeling with supply chain optimization, hospitals can budget and allocate resources to meet patient needs responsively when they arise. Their joint patient-supply chain distribution model accounts for in-patient hospital costs, HaH staff visit time and expenses, procurement, transportation, and inventory holding costs while maintaining strict service guarantees. A key takeaway of the research is the value of taking an uncertainty-aware approach to supply chain management. Incorporating uncertainty quantification allows healthcare providers to maintain adaptive decision-making when allocating resources to at-home patients.

This research reveals significant differences in operational viability based on geographic density. Initial findings show the model is most effective in densely populated urban areas near distribution centers. The viability is also sensitive to the exact setup – an apartment with tight spaces may not be suitable, even in a dense city. Conversely, in rural regions where residents live far apart, high supply chain and operational transportation costs continue to pose a barrier to making HaH care financially viable.

“Rural areas face two challenges at once: lower volume to spread the fixed costs over, and longer travel distances that make staffing much harder,” Wiberg says. “Operational strategies like sharing infrastructure across hospitals help, but in many of these places they may not be enough on their own. That is a policy question as much as an operations one.” Policymakers must grapple with who is going to bear the home care cost, and whether reimbursement policy, which was negotiated primarily for in-hospital inpatient care, should be adjusted based on the delivery environment.

Ultimately, Wiberg and Zhang’s research offers a roadmap for healthcare administrators to evaluate where distributed care models can succeed. By balancing patient safety, predictive analytics, and supply chain optimization, their framework provides actionable insights into improving patient outcomes and minimizing total care delivery costs.