Optimizing Patient Placement in Normal Care Units: An Instrumental Causal Forest Approach Minimizing Mortality

TL;DR

This paper presents a data-driven method using instrumental variable causal forests to optimize patient placement in hospital units, reducing mortality by balancing specialization and utilization without expanding capacity.

Contribution

It introduces a novel application of instrumental variable causal forests for personalized patient placement and designs a minimax regret policy to improve health outcomes.

Findings

Clear trade-off between specialization and utilization affecting mortality.

The proposed policy reduces mortality without capacity expansion.

Personalized placement improves resource efficiency.

Abstract

Normal care units (NCU) placement affects health outcomes. NCUs in a hospital have different specialisations. There are patients that can potentially stay in multiple different NCUs. On a given day the NCUs are on different utilisation levels, which also affects health outcomes. Our approach uses instrumental variable causal forests, with emergency admission as an instrument, to estimate how the effect of NCU placement varies across patients and utilisation levels. The results show a clear trade-off between specialisation and utilization. Based on these findings, we design a minimax regret placement policy, using frequentist, Balke-Pearl and Manski bounds, that lowers mortality without capacity expansion. The policy reallocates patients according to their individualized average treatment effects, showing that data-driven patient placement can improve outcomes by using existing resources more efficiently.