Logo image
Back
Identifying De Facto Prescription Norms in a Hospital Setting: A study with antibiotics
Conference proceeding

Identifying De Facto Prescription Norms in a Hospital Setting: A study with antibiotics

Syeda Momina Tabish, Philip Polgreen, Alberto Maria Segre and Padmini Srinivasan
2019 IEEE International Conference on Healthcare Informatics (ICHI), pp.1-11
06/2019
DOI: 10.1109/ICHI.2019.8904497

View Online

Abstract

The improper use of antimicrobials can cause serious public health problems because of mutations of bacteria into resistant strains. Given growing concerns over microbial resistance, antibiotic stewardship programs are being established in various health care settings. This paper proposes the notion of a data-driven community "norm" that can assist in the stewardship process. A norm represents the collective prescription patterns of a hospital community with regards to a class of drugs. The paper also proposes a set of desired properties of a norm and an approach to extract such norms from patient prescription records. The norm is framed as a weighted list of ICD-9 codes with higher weights indicating a greater likelihood of antibiotic prescription. We demonstrate our ideas by generating norms for antibiotic (vs non-antibiotic) prescriptions and for Fluoroquinolone, a particular class of antibiotics (vs other antibiotics). The norms, extracted from a 9-year dataset of prescription records from a large Midwestern medical center, are sensible and useful as demonstrated in three validation efforts: (1) ability to predict prescriptions for new visits, (2) consistency when norms are generated from different data subsets and (3) connections with de jure norms. For example, when predicting antibiotic prescriptions for new visits we achieve nDCG scores of around 0.95 and P@100 scores greater than 0.7. Additionally, we show the benefit of including norm based features in a traditional SVM based classifier trained to make prescription predictions. Beyond assisting with antibiotic stewardship programs the underlying data-driven algorithms may be used to extract community norms for any class of prescription and other treatment strategies. Most importantly, our algorithms allow us to address a new and important class of problems in health care informatics, that of comparing prescription norms across regions, communities, population subgroups and so on with expert guidelines.
 Hospitals Antibiotics Antimicrobial Stewardship Prescription Norms Electronic Health Records Medical diagnostic imaging Antibiotic Prescription Guidelines Diseases Immune system

Details

Metrics

20 Record Views
Logo image