Understanding the Primary-Specialty Referral Mechanism using Network Science
Published in The International Conference on Complex Networks and Their Applications, 2019
Recommended citation: Joao Casal da Veiga, Qiwei Han, Claudia Soares, "Understanding the Primary-Specialty Referral Mechanism using Network Science." The International Conference on Complex Networks and Their Applications, 2019. https://shirshendu.ccny.cuny.edu/PDF%20Files/CPD_BookOfAbs.pdf#page=57
Extended abstract
The medical referral between Primary Care Physicians (PCP) and specialists represents the formal mechanism in the health system to address the need of patients for specialty care [5]. Typically, for a given patient with clinical needs, PCP can make a choice of several specialists to whom they may refer and their choice would have important downstream effects. As such, primary-specialty referral may affect many aspects of patient care, such as quality of care, patient satisfaction and health care costs, etc. [3]. Researchers recently leveraged the patient consultation history extracted from insurance claims data to construct the patient sharing network between physicians based on the shared patients [3, 6]. Essentially, patient sharing network operationalizes an informal information-sharing network in which physicians provide care to shared patients. This network does not necessarily conform to the formal organizational structure that physicians are affiliated with, but may provide valuable insights in explaining the referral mechanism. For example, both [3] and [7] performed social network analysis on Medicare administrative data and showed that structure of patient sharing networks and the position of physicians in the network has a significant relationship with the overall cost and intensity of care. [2] further discovered small-world structure and strong correlations between certain network statistics with health system statistics. These metrics derived from network science can serve as informative features to boost predictive model performance and optimize health system for improved medical outcomes [1]. In this paper, we aim to add to the literature of understanding the primary-specialty referral mechanism. We obtain a large-scale patient consultation dataset from a private European health provider with over 9 million consultations between 1.3 million patients and 2,308 physicians (515 PCP and 1793 specialists) in 7 hospitals between 2012-2017. The primary-specialty referral is defined as when a patient consults a PCP and then a specialist within 30 days. In other words, there only exist links between two distinct set of physician nodes, namely PCP and specialists. As such, we develop a weighted bipartite network where 460 PCP are connected with 1,542 specialists through 78,593 edges. The edge weight of referral network represents the number of patients that PCP refer to the specialists. Importantly, 306 physicians do not have any edge to the referral 39 network, which raises the potential inefficiency concerns for their lack of involvement in referral process. Besides, we obtain additional physician registration data from the Human Resource department of the provider, including gender, age, education and internship institution, specialty, working hospital, etc. As we augment the referral network with such information, summary statistics show PCP and specialists indeed have different background. The gender composition for PCP is 69% of female and 31% male, and 43% of female and 57% of male for specialists. Also, we observe certain level of homophily. For example, more than half of the referrals are made between the physicians with the same gender and more than 70% of the referrals are made between those with the age difference less than 10 years. More interestingly, 67% of the referrals are made between physicians that work at the same hospital, implying that referral network may exhibit strong clustering in terms of physicians’ background. Following [8], we compute macro-level structural metrics from the resulting referral network and compare them against those of a synthetic Erdős-Rényi random network. We find that degree distributions (the number of specialists that PCP refers to, and the number of PCPs that specialist receives from) for the referral network do not follow Poisson distribution. Meanwhile, we obtain average clustering coefficient for the referral network (0.17) to measure the fraction of the number of observed squares to the total number of possible squares in the network. It is about 2.5 times higher than that of random network (0.07). This represents an essential precondition for referral network to exhibit small-world structure and suggests that physicians in the referral network have higher tendency to cluster together. We also quantify both betweenness and closeness centrality for physicians in the referral network. The former describes the number of shortest paths that pass through a physician while the latter describes the reciprocal of the sum of distance to all other physicians in the network. Top 25% and bottom 25% of PCP in terms of betweenness centrality initiate 58.4% and 0.07% of referrals, respectively. Top 25% and bottom 25% of specialists in terms of betweenness centrality receive 55.9% and 3.3% of referrals, respectively. Again, this raises the inefficiency concern as referral process occurs high skewed towards a small number of physicians. We adopt the popular modularity-based optimization algorithm Louvain to extract communities from the referral network [4]. In total, we identify 7 distinct communities, which happens to correspond to the number of hospitals of the provider. Figure 1 shows the visualization of community structure. In general, PCP tend to refer patients to specialists belonging to the same community, which indicates that physicians may form a “referral clique” wherein referral process occurs more likely than to physicians from different communities. Meanwhile, there is one community (in purple) that is located distantly from other communities, which contains physicians mostly working at the hospital in a different region. Moreover, we demonstrate that physicians within the same community share more similarity in terms of their background, namely, they are at the similar age, have similar number of years of experience, used to study and intern at the same institution and now work at the same hospital. Our results show that referral network may highly overlap with the social network of physicians and in the future work we plan to explore the correlation between them.
BibTeX
@inproceedings{Veiga2019,
author = "da Veiga, Joao Casal and Han, Qiwei and Soares, Claudia",
title = "Understanding the Primary-Specialty Referral Mechanism using Network Science",
booktitle = "The International Conference on Complex Networks and Their Applications",
year = "2019",
abstract = "The medical referral between Primary Care Physicians (PCP) and specialists represents the formal mechanism in the health system to address the need of patients for specialty care [5]. Typically, for a given patient with clinical needs, PCP can make a choice of several specialists to whom they may refer and their choice would have important downstream effects. As such, primary-specialty referral may affect many aspects of patient care, such as quality of care, patient satisfaction and health care costs, etc. [3]. Researchers recently leveraged the patient consultation history extracted from insurance claims data to construct the patient sharing network between physicians based on the shared patients [3, 6]. Essentially, patient sharing network operationalizes an informal information-sharing network in which physicians provide care to shared patients. This network does not necessarily conform to the formal organizational structure that physicians are affiliated with, but may provide valuable insights in explaining the referral mechanism. For example, both [3] and [7] performed social network analysis on Medicare administrative data and showed that structure of patient sharing networks and the position of physicians in the network has a significant relationship with the overall cost and intensity of care. [2] further discovered small-world structure and strong correlations between certain network statistics with health system statistics. These metrics derived from network science can serve as informative features to boost predictive model performance and optimize health system for improved medical outcomes [1]. In this paper, we aim to add to the literature of understanding the primary-specialty referral mechanism. We obtain a large-scale patient consultation dataset from a private European health provider with over 9 million consultations between 1.3 million patients and 2,308 physicians (515 PCP and 1793 specialists) in 7 hospitals between 2012-2017. The primary-specialty referral is defined as when a patient consults a PCP and then a specialist within 30 days. In other words, there only exist links between two distinct set of physician nodes, namely PCP and specialists. As such, we develop a weighted bipartite network where 460 PCP are connected with 1,542 specialists through 78,593 edges. The edge weight of referral network represents the number of patients that PCP refer to the specialists. Importantly, 306 physicians do not have any edge to the referral 39 network, which raises the potential inefficiency concerns for their lack of involvement in referral process. Besides, we obtain additional physician registration data from the Human Resource department of the provider, including gender, age, education and internship institution, specialty, working hospital, etc. As we augment the referral network with such information, summary statistics show PCP and specialists indeed have different background. The gender composition for PCP is 69\% of female and 31\% male, and 43\% of female and 57\% of male for specialists. Also, we observe certain level of homophily. For example, more than half of the referrals are made between the physicians with the same gender and more than 70\% of the referrals are made between those with the age difference less than 10 years. More interestingly, 67\% of the referrals are made between physicians that work at the same hospital, implying that referral network may exhibit strong clustering in terms of physicians’ background. Following [8], we compute macro-level structural metrics from the resulting referral network and compare them against those of a synthetic Erdős-Rényi random network. We find that degree distributions (the number of specialists that PCP refers to, and the number of PCPs that specialist receives from) for the referral network do not follow Poisson distribution. Meanwhile, we obtain average clustering coefficient for the referral network (0.17) to measure the fraction of the number of observed squares to the total number of possible squares in the network. It is about 2.5 times higher than that of random network (0.07). This represents an essential precondition for referral network to exhibit small-world structure and suggests that physicians in the referral network have higher tendency to cluster together. We also quantify both betweenness and closeness centrality for physicians in the referral network. The former describes the number of shortest paths that pass through a physician while the latter describes the reciprocal of the sum of distance to all other physicians in the network. Top 25\% and bottom 25\% of PCP in terms of betweenness centrality initiate 58.4\% and 0.07\% of referrals, respectively. Top 25\% and bottom 25\% of specialists in terms of betweenness centrality receive 55.9\% and 3.3\% of referrals, respectively. Again, this raises the inefficiency concern as referral process occurs high skewed towards a small number of physicians. We adopt the popular modularity-based optimization algorithm Louvain to extract communities from the referral network [4]. In total, we identify 7 distinct communities, which happens to correspond to the number of hospitals of the provider. Figure 1 shows the visualization of community structure. In general, PCP tend to refer patients to specialists belonging to the same community, which indicates that physicians may form a “referral clique” wherein referral process occurs more likely than to physicians from different communities. Meanwhile, there is one community (in purple) that is located distantly from other communities, which contains physicians mostly working at the hospital in a different region. Moreover, we demonstrate that physicians within the same community share more similarity in terms of their background, namely, they are at the similar age, have similar number of years of experience, used to study and intern at the same institution and now work at the same hospital. Our results show that referral network may highly overlap with the social network of physicians and in the future work we plan to explore the correlation between them.",
abstractsource = "https://shirshendu.ccny.cuny.edu/PDF\%20Files/CPD\_BookOfAbs.pdf\#page=57",
abstractsourcelabel = "Conference extended abstract",
url = "https://shirshendu.ccny.cuny.edu/PDF\%20Files/CPD\_BookOfAbs.pdf\#page=57",
abstractheading = "Extended abstract"
}