Prescriptive analytics for inventory management in health care

Abstract

In this paper we propose an innovative optimization method for drug replenishment in a hospital ward. Although hospital logistics has in general shifted towards a zero local inventory policy, still most wards keep an inventory of drugs necessary for short time periods (between one and three days). The model we propose makes use of machine learning coupled with stochastic optimization in order to take into consideration both the historical usage patterns of drugs and the ward’s current situation to minimize inventory levels as well as the necessity for emergency replenishments. Differently from generic inventory models, drugs are associated with patients who tend to consume a similar amount of drugs every day during their stay. Our aim in this paper was to show that, with suitably defined ward features, a machine learning method is capable of selecting a set of scenarios over which a stochastic optimization approach is capable of delivering good decisions, both in terms of small order quantities and of service quality (avoiding emergency replenishment orders). An ad hoc feature engineering procedure has been designed to exploit ward features, easily collected every day in the ward. Moreover, we introduce a new validation metric tailored for selecting hyperparameters to balance between optimality and robustness. Our results, based on a dataset composed of one year of drug administrations in an Italian ward, support the superiority of the proposed approach versus traditional ones.

Keywords:

• Prescriptive analytics
• machine learning
• inventory optimization
• stochastic optimization
• K-nearest neighbors
• decision trees
• random forest
• XGBoost

Disclosure statement

No potential conflict of interest was reported by the authors.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

1 To be more precise, for the last test example the training set consists of all the other examples in the dataset.

Additional information

Funding

The research presented in this paper was partially supported by “Project LINFA - Logistica Intelligente del Farmaco” - Tuscany Region - PAR FAS 2007-2013, Linea d’Azione 1.1 - Bando FAR FAS 2014.