Clinical Data Intelligence & Advanced Analytics to Reduce Drug Diversion across the Care Delivery Cycle and Drug Supply Chain in Health Systems

点击翻译按钮获取中文摘要

This SBIR project will research mechanisms to detect when Health Care Workers (HCWs) in hospitals steal or “divert” legal drugs either to abuse themselves or to illegally sell to others. We focus on HCWs in hospitals because of the alarming rates of substance abuse and diversion in hospitals, with multiple studies finding roughly 10% of our nation’s nurses, anesthesiologists, and pharmacists are currently diverting drugs in their workplaces. HCWs are becoming addicted, destroying their careers, jeopardizing their patients’ safety, and increasingly dying from drug diversion overdoses. Diversion continues even though most hospitals already lock­p addictive drugs in Automated Dispensing Machines (ADMs), and run monthly “anomalous usage” computer reports to try to detect diversion. Hospitals broadly agree these current methods have two main weaknesses: 1. Data in the ADM only show part of the equation: the dispensing of the drug from the locked cabinet, ignoring drug administration data in the Electronic Medical Record (EMR), as well as other data available in other existing hospital computer systems. 2. Motivated diverters can game the system with falsified data entries to avoid detection. This SBIR project will conduct research to address these two problems by building a computer system with (a) automated data feeds from multiple existing hospital computer systems and (b) advanced analytics to flag potential diversion for investigation. We will test the following four hypotheses: • Data Consolidation hypotheses and experimentation plan: Phase 1: If we consolidate data from two systems (EMR & ADM), then we can detect diversion that would have been undetected using data only from the ADM (Hypothesis 1) Phase 2: If we consolidate data from five systems (EMR, ADM, Purchasing Systems, Internal Inventory System(s), and Employee Time Clocks) then we can detect diversion that would have been undetected using only EMR & ADM data (Hypothesis 3) • Data Analytics hypotheses and experimentation plan: Phase 1: If we create and test algorithms on blinded, consolidated, historical data from EMR/ADM, then we can detect known cases of drug diversion that that current methods do not detect, with fewer Type II errors (“false negatives”). (Hypothesis 2) Phase 2: If we refine and test additional algorithms using near­real­time, consolidated data from the five computer systems above, then we can detect drug diversion that current methods do not detect, faster, with fewer Type I errors (“false positives”) and fewer Type II errors. (Hypothesis 4)