Doctor shopping: Insights from two-mode social network analysis

The proposed project will analyze patterns of doctor shopping for prescription opioids and benzodiazepines using two-mode (i.e. affiliation) social network analysis (SNA) and machine learning techniques. Abuse of prescription opioids and benzodiazepines in the U.S. has risen rapidly, creating a public health crisis. Mortality from drug overdose is higher than motor vehicle accidents and is among the nation’s leading preventable causes of death. Doctor shopping is defined as obtaining controlled substances from multiple health care practitioners simultaneously, exceeding the recommended dosage. Doctor shopping is a principle method of obtaining controlled substances for misuse, and an indicator of escalating drug abuse that is associated with a two-fold risk for fatal overdose. Lack of consensus about criteria for classifying doctor shopping has led to wide variation in estimated rates of questionable prescribing activity. This ambiguity poses barriers to understanding factors underlying doctor shopping, and impedes the evaluation of prescription drug policies. Currently, numerical thresholds or multiple provider episodes (MPEs; i.e. overlapping prescriptions) are most often used to identify doctor shoppers, resulting in false positives and negatives. A goal of the proposed study is to determine whether incorporating information about actors’ connectedness in a two-mode social network of doctor shoppers and prescribers can produce more valid indictors of illicit behavior. Two-mode network analysis focuses on ties between two different classes or sets of entities. In the proposed project, data form a two-mode network with prescribers in one class and patients in the other, and ties are present only between prescribers and patients. Two-mode SNA is ideal for examining structural patterns in social interaction between two sets of actors, and for determining the most active and central actors in a network. Consistently targeting prescribers that are key players in prescription drug networks may be indicative of strategic doctor shopping behavior or information sharing among patients. Insights from SNA will be used to fine-tune doctor-shopping indicators, improving our ability to detect early signs of abuse, as well as behavior that is intermittent, ambiguous, or less intense, but still problematic. The specific aims of the proposed study are to: 1) Examine the utility of network analysis and machine learning for improving doctor shopping indicators; and 2) Identify characteristics of doctor shopping patients, targeted prescribers, and the point of service using two-mode SNA. To our knowledge, this is the first study of prescription drug abuse to use two-mode SNA as its methodological approach. This study is significant in applying two-mode SNA – a complex, systems science methodology – to drug abuse research and has the potential to facilitate cost-effective information extraction from extant datasets for translation to policy and practice. The long-term goal of this research is to leverage insights from SNA and machine learning to improve detection and prevention of doctor shopping and related fraudulent activities, and ultimately reduce the prevalence and public health impact of prescription drug abuse.