physicians or nurse practitioners in case of an alarming
situation. The engine also lets the provider know if the intervention is not working as intended.
The prevention engine’s objective is to generate insights
that can aid policy design and collaboration among various stakeholders within the healthcare landscape to reduce the incidents of prescription abuse.
Quartet of Engines Decoded
The design of the detection engine rests on syntactic and semantic analysis of physician notes from the EHR system using
context free grammars and medical-specific data dictionaries by
the Unified Medical Language System (UMLS) and SNOMED.
At the core of this analysis is the mining of physician notes
based on ontologies created along these three dimensions: prescription opioids typically abused by patients; behaviors and
symptoms that prescription opioid seekers complain about;
and diseases and conditions that such patients typically report.
These ontologies can be either created through systematic text
mining of physician notes of opioid-use disorder patients or can
be developed in collaboration with clinicians utilizing the information available in SNOMED, research journals published in
PubMed, and the UMLS. 3, 4
In either case, validation by clinical experts will help the engine become more accurate relatively faster. Mining of physician
notes will reveal a list of patients who are currently suffering from
opioid abuse. Superimposing this analysis with the creation of a
lexicon specific to problems at hand then helps in the identification of patients who are at risk of opioid abuse. While there is no
shortcut to ensure that this engine will flag minimal false positives, with continuous feedback from clinicians and recalibration
by data scientists, it is fair to assume that the engine will become
reasonably reliable in a short amount of time.
Coupling the above results with analysis of laboratory data,
The engine can also consider prescription events from state
prescription drug monitoring program (PDMP) data so that ad-
verse interactions among medications are avoided, as well in-
clude rules corresponding to guidance from the Centers for Dis-
ease Control and Prevention (CDC) and the Substance Abuse
and Mental Health Administration (SAMHSA), on handling
The need for the monitoring engine is clear—human lives are
at stake. The monitoring engine takes a two-pronged approach
to track and measure the effectiveness of interventions, including
patient adherence to the treatment and efficacy of the treatment,
respectively. Wearables, Internet of Things (Io T) data, and remote
heath monitoring will have to come together to make this engine
a reality. There are niche companies who are doing work in the
direction of making remote patient monitoring a reality, and collaboration with them might be the best way forward. The data from
wearables and remote health monitoring, when coupled with EHR
data, can help in tracking adherence to, and measuring the effectiveness of, interventions.
The prevention engine is the most elusive in the data science
world. It is intended to help policymakers and lobbyists gather
real-world evidence on how prescription opioid abuse is being
treated, what is working, and what is not. This engine generates
self-service reports on efficacy of various interventions, under
various conditions. These reports to executive management can
aid policy directions and decisions. Output from the other three
engines can help generate these reports.
In an ideal scenario a collaboration, created through colla-
Quartet of Prescription Opioid Abuse Engines
Continued on page 54