Social-Network Analysis for Pain Medications:
Influential physicians may not be high-volume
Abhinav Choudhury, Shruti Kaushik, Varun Dutt
School of Computing and Electrical Engineering
Indian Institute of Technology Mandi
Himachal Pradesh, India -175005
Abstract—According to the Institute of Medicine of the
National Academies, more than 100 million Americans suffer
from chronic pain related to diabetes, heart disease, and cancer
combined. Adoption of pain medications and safe healthcare
practices is a major global policy concern. This adoption process
is highly influenced by the interpersonal network of physicians
prescribing medications to treat pain. However, existing research
into physician networks have been hospital-specific, applied to a
smaller number of physicians, and dependent upon physicians’
self-reports. In this paper, using big-data and data-mining, we
overcome these limitations: By using a case of 30+ hospitals
spanning across 2000+ physicians, we create a social network
containing physicians’ prescription data and adoption behavior
of pain medications. The social network assumes that connected
physicians work in the same hospital and belong to the same
specialty or specialty group. Then, using the centrality measures,
degree and eigenvector centrality, we analyze prescription
volumes and proportion of adopters of pain medications. We also
analyze gender effects. Results revealed that the most influential
physicians were not the high-volume prescribers. Males’
physicians were more influential compared to female physicians;
however, females prescribed more volume compared to males.
Our results help us identify critical physicians from certain core
specialties and specialty groups who may be approached by
patients seeking pain relief.
Keywords—Social network analysis; eigenvector centrality;
gender; pain medications; big-data.
According to the Institute of Medicine of the National
Academies, more than 100 million Americans suffer from
chronic pain related to cases of diabetes, heart disease, and
cancer combined . Pain medications are medicines
prescribed by multitude of physicians from different specialties
. Adoption of pain medication is a lively process highly
dependent on the interpersonal interactions between members
of a social network.
Social Network Analysis (SNA) , which views the
structure of social interactions as networks composed of nodes
(physicians) interconnected by edges (social relations,
friendship, advice), is an ideal approach for describing
interaction patterns to study how social inﬂuence is transmitted
among physicians and how it affects their contingent behavior
. A social network/system is the pattern of friendship,
advice, communication or support which exists among the
members of a social system [3,5-7]. Social networks can
promote innovation processes and expand opportunities for
learning . Such networks can be used to find key-opinion
leaders inside a social system.
Whenever a new pain medication is released, the adoption
process in a social network is unlikely to be simultaneous, i.e.,
not all physicians adopt the medication at the same time .
For many new medications, an individual’s decision to adopt
the technology depends on whether his contacts have adopted it
already . The adoption process may differ from one
individual to another. While some may adopt the innovation
early; others may adopt it late or may not adopt it at all . In
fact, many times new medicine innovations may require long
period of time, mostly some years, from the time when they
become available in the market to the time when they maybe
widely adopted by physicians . Thus, the spread or
diffusion of innovations (e.g., of pain medications) could be
thought of as a “social process” that is highly dependent on the
members of the social system. Social inﬂuence, often
crystallized as opinion exchange and behavior imitation, is
mostly conveyed through physicians’ interpersonal social
interactions with members of his social network .
Existing applications of social-network analyses for
investigation adoption of an innovation among physicians has
been hospital-specific, and were created using physicians’ self-
reports [13-14]. Thus, social-network analysis research that
considers many physicians over several hospitals in a
geographic area and that is independent of physician’s self-
reports is much needed, but lacking in literature.
In this paper, using big-data and data-mining, we create a
social network of several physicians over a number of hospitals
in the US by considering assumptions on physician’s specialty
and affiliation to the same hospital. In what follows, we first
cover background literature related to applications of social
network analyses to adoption of innovations among physicians.
Next, we build a social network and use it to draw conclusions
on critical physicians, prescription volume, and adoption
II. RELATED WORK
The rate of adoption of a new medication (innovation)
varies from physician-to-physician . Most importantly, the
medication that a physician prescribes is likely to be highly
influenced by the interpersonal communication of the
physician with members of his personal network . While
some physicians may adopt a new medication early, most of
them may want to test the waters first. Thus, most physicians
may want to wait for others in his social system to have tried
the innovation first. It has been shown that physicians who are
‘socially proximate’ in a social environment often use one
another as information sources to manage the uncertainty of
adopting new antibiotic drugs .
Previous work has focused on questionnaire-based or
survey-based approaches to create the social network. For
example, Creswick and Westbrook  conducted a study in
the renal ward of an Australian metropolitan teaching hospital.
They designed a social network questionnaire containing ﬁve
social network questions about each person in the ward. The
advice-seeking interactions of doctors, nurses, allied health
professionals (including a pharmacist) and a ward clerk were
examined using the data from questionnaires administered to
staff. Keating et al.  have also used a survey-based
approach for creating a social network. Their study population
included an alphabetized list of 38 primary-care physicians at a
major Boston teaching hospital. Participants were asked to
“circle the number of conversations that they have had with
each of the following primary-care physicians in the past 6-
months, which have influenced their thinking on women’s
health issues” (response options were 0, 1–3, or ≥4). Physicians
were also asked to report one individual, inside or outside of
the practice, who most likely influences their practice regarding
women’s health issues.
These approaches to social network creation suffer from
sampling bias, i.e., the responses and outcomes of the survey
are highly dependent on the chosen sample. In this paper, we
create a social network by mining big-data from 30+ hospitals
and 2000+ physicians of US. Then, using the prescription data
of these physicians, we create a directed influence graph for
identifying the critical physicians using degree (in-degree and
out-degree)  and eigenvector centrality. Degree centrality
refers to the number of ties a physician has with other
physicians . Eigenvector centrality  is a measure of the
influence of a physician in a network of physicians. It assigns
relative scores to nodes representing physicians in the network
based on the concept that connections to high-scoring nodes
contribute more to the score of the node in question than equal
connections to low-scoring nodes . In this paper, we focus
on the diffusion of pain medications in a large network of
physicians prescribing pain medications.
We used a medical-prescriptions dataset and a physician-
affiliation dataset to create the social network of physicians
prescribing pain medications.1 These datasets are proprietary
and could be purchased from entities collecting such data
anonymously. In US, each physician is affiliated to multiple
hospitals and each hospital is affiliated to an Integrated
Delivery Network (IDN). An IDN is a formal system of
providers and hospitals that provides both healthcare services
and a health insurance plan to patients in a geographic area
. We have considered an IDN located in Boston, MA,
consisting of 30+ hospitals affiliated under it and 2000+
physicians. The prescription dataset consisted of billions of
physicians prescribing certain pain medications between years
2011 and 2016.
B. Social Network Creation
The following assumption was used to create a social network:
Physicians affiliated or working in the same hospital within the
same specialty or specialty group and prescribing the same
pain medication were assumed to influence each other.
Specialty groups2 are like clusters that contain specialties from
similar fields that are related to one another. Once the social
network was created, we used the adoption pattern for a
popular pain-medication (called M) to create a directed
influence graph. The time of first prescription of the pain
medication by a physician was taken as the time of adoption of
the innovation by the physician. Physicians adopting M at an
earlier date were assumed to influence other physicians’
adoption of M in his/her social network. In the directed graph,
an arrow (A, B) is directed from A to B, where ‘A’ signifies
the physician that has been influenced by physician ‘B’ for
medication’s adoption. After the creation of the social network,
we visualized the sociogram using gephi  and used graph
centrality measures like degree  and eigenvector centrality
 to pinpoint the critical physicians’ that have influenced the
In this section, we consider the results of how physicians relate
to other physicians in their network. Fig. 1 shows the directed
influence graph between various physicians in the social
network. There are 22 nodes (a node represents a physician)
and 158 edges in the social network in Figure 1. There exist
two isolated clusters in the graph belonging to different
specialty group: Primary-care physician (PCP) and NRPPHA
(Nurse-practitioner). The average path length is 1, which
indicates that, on average, there is only one tie between each
pair of nodes in the influence network. After the creation of the
social network, we applied graph centrality measures to
identify critical physicians within the network.
1 Data shown may be altered to protect privacy and uphold non-
2 These specialty groups were provided with the data.
We found that physicians from specialty internal medicine and
family medicine had the highest in-degree and eigen-vector
centralities (see Fig. 2 and 3). High in-degree centrality for a
physician signify that the physician has influenced more
number of physicians in his/her social network; while, a high
eigenvector centrality for a physician signify that the physician
is connected to highly influential nodes and is in turn also
highly influential. As such, they play a significant role in the
diffusion of the pain medication within the network.
Fig. 2. Average In-degree for different specialties.
Next, we analyzed the total number of scripts of medicine M
prescribed by each physician. We found that even though
physicians from specialties internal medicine and family
medicine were highly influential (high eigenvector centrality
and in-degree, see Fig. 2 and 3); neurology and nurse
practitioners were the ones prescribing the highest volumes of
medication M within the IDN (see Fig. 4). Thus, the most
influential physicians may not be the ones prescribing the most
Furthermore, Fig. 5 shows the proportion of physicians under
different specialty groups who had adopted the medicine M.
We observed that more number of physicians belonging to PCP
specialty were adopting the pain medication compared to
physicians from other specialty groups.
Fig. 3. Average Eigenvector centrality for different specialties.
Fig. 4. Average number of scripts prescribed for different specialty.
Fig. 5. Proportion of physicians adopting medicine M from different
Next, we analyzed the demographic data (gender) of the
physicians. We found that number of males (16) and female
(15) physicians prescribing the medication M were similar;
however, females (mostly nurse-practitioners) were prescribing
higher volume of scripts compared to male physicians (males:
132,004; females: 159,361). Furthermore, males had a higher
influence (high eigenvector centrality) compared to females
(males: 0.23; females: 0.05).
V. DISCUSSION AND CONCLUSION
Pain medications are the highest prescribed medication in
the US ; and, in this paper, using big-data and machine-
learning, we show interesting contrasting patterns between
Fig. 1. Directed Influence graph between different prescribers of
pain medication M. The numbers associated with each node are
the identification number of each physician. Two isolated clusters
of specialty group (PCP and Nurse Practitioner) are visible.
influencers, volume prescribers, and adopters of pain
medications. In this paper, we used the following assumption
to create the social network: Physicians affiliated or working in
the same hospital within the same specialty or specialty group
and prescribing the same pain medication were assumed to
influence each other. Using many physicians across several
hospitals, our results reveal that Primary-Care Physicians
(PCPs) with specialties like family medicine and internal
medicine were the most influential doctors prescribing pain
medications. However, these PCPs were not among the highest
volume prescribers of the pain medications.
Firstly, we found that physicians with specialties family
Medicine and internal Medicine were highly influential (high
eigenvector centrality) in the diffusion of pain medication
within their social network. One likely reason for this result
could be that physicians with these specialties are primary-care
providers, who are the first point-of-contact for any
undiagnosed pain issues reported by patients. These PCPs
know many other specialties as they are in the process of
regularly referring patients to these specialties.
Furthermore, the average path length was 1 in our social
network, which indicates that, on average, there is only one tie
between each pair of physician nodes in the influence network.
Thus, on average, each physician is only one step away from
each other physicians in their network and the network is very
cohesive. This unit path length is also indicative of the referral
pattern between PCPs and other medical specialties in the
Second, our results showed that nurse practitioners and
neurologists were prescribing more scripts of pain medication
than other specialties. One likely reason for this observation
could be that nurse practitioners are trained to manage acute
and chronic medical conditions . Thus, they come in
contact with more number of patients suffering from pain as
compared to the physicians from other specialties.
Finally, our results showed that females physicians were
prescribing more scripts of pain medication as compared to
males. Upon investigation we found that this can be attributed
to the fact that Nurse Practitioners in the IDN are mostly
females and since they are prescribing high number of scripts
of the medicine, the prescription volume is skewed towards
female physicians as compared to male physicians.
In our paper we have considered a single Integrated delivery
network (IDN) located in Massachusetts for the creating the
social network. Also we have considered only a single pain
medication for creating the directed influence graph. Even
though we started with 2000+ physicians from 30+ hospitals,
due to considering a single medication we got a smaller graph.
By using multiple pain medications we can overcome this
limitation. Also in our future work, when we will consider
multitude of IDNs’ situated all over the US and multiple pain
medications for creating the directed influence graph and also
bolster our assumptions by using data from social network
VII. FUTURE SCOPE
Social network analysis can be an efficient and powerful tool
to find key opinion leaders. In the future we will consider
referral patterns, author-co-author relationship and nomination
studies to strengthen the reliability of the connections between
the physicians. We will also take into consideration different
pain medications for creation of the directed influence
The project was supported from grant (awards:
#IITM/CONS/PPLP/VD/03) to Varun Dutt.
 Institute of Medicine Report from the Committee on Advancing
Pain Research, Care, and Education. 2011. Relieving Pain in
America, A Blueprint for Transforming Prevention, Care, Education
and Research. The National Academies Press
 “Pain management" Wikipedia: The Free Encyclopedia. Wikimedia
Foundation, Inc. 21 March 2017. Web. 24 March. 2017
 Scott, J.: Network Analysis: A Handbook. Sage. Newbury Park, CA
 Zheng, K., Padman, R., Krackhardt, D., Johnson, M. P., &
Diamond, H. S, “Social networks and physician adoption of
electronic health records: insights from an empirical study”, Journal
of the American Medical Informatics Association, 17(3), 2010: 328-
 Knoke, D., Kuklinski, J.H., “Network analysis”. Sage. Newbury
Park, CA, 1982
 Burt, R.S., Minor, M.J., “Applied Network Analysis”. Sage.
Newbury Park, CA, 1983
 Wellman, B., “Structural analysis: From method and metaphor to
theory and substance” Contemporary Studies in Sociology. 15, 19-
 Kolleck, Nina., "Social network analysis in innovation research:
using a mixed methods approach to analyze social innovations."
European Journal of Futures Research 1.1 2013: 1-9.6.
 Kaushik, Shruti, Kaushik, S., Choudhury, A., Mallik, K., Moid, A.,
& Dutt, V. "Applying Data Mining to Healthcare: A Study of Social
Network of Physicians and Patient Journeys." Machine Learning
and Data Mining in Pattern Recognition. Springer International
Publishing, 2016. 599-613.
 Young, H. Peyton. "The diffusion of innovations in social
networks." The economy as an evolving complex system III:
Current perspectives and future directions 267 (2006).
 Choudhury, A., Kaushik, S., and Dutt, V., “Data-mining
Medications for Pain: Analyzing adoption behavior of physicians
and switching behavior of patients”, Proceedings of the 2nd
International Conference on Internet of things, Data and Cloud
Computing. ACM, 2017, In press
 Everett Roger, “Diffusion of Innovation”, Free Press. 1983
 Creswick, Nerida, and Johanna I. Westbrook. "Social network
analysis of medication advice-seeking interactions among staff in an
Australian hospital." International journal of medical informatics
79.6 (2010): e116-e125.
 Keating, Nancy L., et al, “Factors affecting influential discussions
among physicians: a social network analysis of a primary care
practice." Journal of general internal medicine 22.6 (2007): 794-
 Valente, T.W., “Social network thresholds in the diffusion of
innovations”, Social networks. 18.1, 69-89 (1996)
 Coleman J, Katz E, Menzel H., “Medical innovation: a diffusion
study”, 2nd edn. New York, NY: Bobbs-Merrill, 1966.
 Freeman, Linton C. "Centrality in social networks conceptual
clarification." Social networks 1.3 (1978): 215-239.
 Ruhnau, Britta. "Eigenvector-centrality—a node-centrality?" Social
networks 22.4 (2000): 357-365.
 “Eigenvector centrality." Wikipedia: The Free Encyclopedia.
Wikimedia Foundation, Inc. 12 November 2016. Web. 24 March.
 Jared Landis. “Post-Acute Care Cheat Sheet: Integrated Delivery
Networks”, Advisory Board 28 April, 2014.Web. 24 March. 2017.
 Bastian, Mathieu, Sebastien Heymann, and Mathieu Jacomy.
"Gephi: open source software for exploring and manipulating
networks." ICWSM 8, 2009, 361-362.
 Leonard J. Paulozzi., Karin A. Mack., and Jason M. Hockenberry.
Vital Signs: Variation Among States in Prescribing of Opioid Pain
Relievers and Benzodiazepines — United States, 2012. Morbidity
and Mortality Weekly Report, July 4, 2014, US Centers for Disease
Control, p. 564.
 “Nurse_practitioner." Wikipedia: The Free Encyclopedia.
Wikimedia Foundation, Inc. 18 November 2016. Web. 24 March.