ISDS Annual Conference Proceedings 2017. This is an Open Access article distributed under the terms of the Creative Commons Attribution-
Noncommercial 3.0 Unported License (http://creativecommons.org/licenses/by-nc/3.0/), permitting all non-commercial use, distribution,  
and reproduction in any medium, provided the original work is properly cited.

ISDS 2016 Conference Abstracts

Multidimensional Tensor Scan for Drug Overdose 
Surveillance
Daniel B. Neill*
H.J. Heinz III College, Carnegie Mellon University, Pittsburgh, PA, USA

Objective
We present the multidimensional tensor scan (MDTS), a new 

method for identifying emerging patterns in multidimensional 
spatio-temporal data, and demonstrate the utility of this approach 
for discovering emerging geographic, demographic, and behavioral 
trends in fatal drug overdoses.

Introduction
Drug overdoses are an increasingly serious problem in the United 

States and worldwide. The CDC estimates that 47,055 drug overdose 
deaths occurred in the United States in 2014, 61% of which involved 
opioids (including heroin, pain relievers such as oxycodone, and 
synthetics).1 Overdose deaths involving opioids increased 3-fold 
from 2000 to 2014.1 These statistics motivate public health to identify 
emerging trends in overdoses, including geographic, demographic, 
and behavioral patterns (e.g., which combinations of drugs are 
involved). Early detection can inform prevention and response efforts, 
as well as quantifying the effects of drug legislation and other policy 
changes.

The fast subset scan2 detects significant spatial patterns of disease 
by efficiently maximizing a log-likelihood ratio statistic over subsets 
of data points, and has recently been extended to multidimensional 
data (MD-Scan).3 While MD-Scan is a potentially useful tool for drug 
overdose surveillance, the high dimensionality and sparsity of the data 
requires a new approach to estimate and represent baselines (expected 
counts), maintaining both accuracy and efficient computation when 
searching over subsets.

Methods
The multidimensional tensor scan (MDTS) is a new approach to 

subset scanning in multidimensional data. In addition to detecting 
the spatial area (subset of locations) and time window affected by 
an emerging outbreak, MDTS can also identify the affected subset 
of values for each observed attribute. For example, given the drug 
overdose surveillance data described below, MDTS can identify the 
affected genders, races, age ranges, and which drugs were involved. 
MDTS finds subsets of the attribute space with higher than expected 
case counts, first using a novel tensor decomposition approach 
to estimate the expected counts. MDTS then iteratively applies a 
conditional optimization step, optimizing over all subsets of values 
for each attribute conditional on the current subsets of values for all 
other attributes3, and using the linear-time subset scanning property2 
to make each conditional optimization step computationally efficient. 
The resulting approach has high power to detect and characterize 
emerging trends which may only affect a subset of the monitored 
population (e.g., specific ages, genders, neighborhoods, or users of 
particular combinations of drugs).

Results
We used MDTS to analyze publicly available data from the 

Allegheny County, PA medical examiner’s office and to detect 
emerging overdose patterns and trends. The dataset consists of 
~2000 fatal accidental drug overdoses between 2008 and 2015.  
For each overdose victim, we have date, location (zip code), age 
decile, gender, race, and the presence/absence of 27 commonly 
abused drugs in their system. The highest-scoring clusters discovered 
by MDTS were shared with Allegheny County’s Dept. of Human 
Services and their feedback obtained.

One set of potentially relevant findings from our analysis 
involved fentanyl, a dangerous and potent opioid which has been a 
serious problem in western PA. In addition to identifying two well-
known, large clusters of overdoses—14 deaths in January 2014 and  
26 deaths in March-April 2015—MDTS was able to provide additional 
information about each cluster. For example, the first cluster was 
likely due to fentanyl-laced heroin, while the second was more likely 
due to fentanyl disguised as heroin (only 11 victims had heroin in 
their system). Moreover, the second cluster was initially confined 
to the Pittsburgh suburb of McKeesport and a typical demographic 
(white males ages 20-49), before spreading across the county. Our 
analysis demonstrated that prospective surveillance using MDTS 
would have identified the cluster as early as March 29th, enabling 
targeted prevention efforts. MDTS also discovered a previously 
unidentified, highly localized cluster of fentanyl-related overdoses 
affecting an unusual and underserved demographic (elderly black 
males near downtown Pittsburgh). This cluster occurred in January-
February 2015, and may have been related to the larger cluster of 
fentanyl-related overdoses that occurred two months later. Finally, 
we identified multiple overdose clusters involving combinations 
of methadone and Xanax between 2008 and 2012, and observed 
dramatic reductions in these clusters corresponding to the passage 
of the Methadone Death and Incident Review Act (October 2012), 
which increased state oversight of methadone clinics and prescribing 
physicians.

Conclusions
Retrospective analysis of Allegheny County overdose data 

suggests high potential utility for a prospective overdose surveillance 
system, which would enable public health users to identify emerging 
patterns of overdoses in their early stages and facilitate targeted and 
effective health interventions. The MDTS approach can also be used 
for other multidimensional public health surveillance tasks, such as 
STI surveillance, where the patterns or outbreaks of interest may have 
demographic, geographic, and behavioral components.

Keywords
event detection; outbreak detection; subset scan; drug overdose 
surveillance

Acknowledgments
This work was partially supported by NSF grant IIS-0953330. The author 
wishes to thank Eric Hulsey (Allegheny County DHS) for feedback on the 
discovered overdose clusters.

References
[1] Rudd RA, Aleshire N, Zibbell JE, Gladden RM. Increases in drug 

and opioid overdose deaths: United States, 2000–2014. MMWR 2016; 
64(40): 1378-1382.

[2] Neill DB. Fast subset scan for spatial pattern detection. J Royal Stat 
Soc B 2012; 74(2): 337-360.

[3] Neill DB, Kumar T. Fast multidimensional subset scan for outbreak 
detection and characterization. Online J Pub Health Inform 2013; 5(1): 
e156.

*Daniel B. Neill
E-mail: neill@cs.cmu.edu

Online Journal of Public Health Informatics * ISSN 1947-2579 * http://ojphi.org * 9(1):e20, 2017