Variation in Provider Connectedness Associates With Outcomes of Inflammatory Bowel Diseases in an Analysis of Data From a National Health System

Abstract

BACKGROUND & AIMS:

Inflammatory bowel diseases (IBD) often require multidisciplinary care with tight coordination among providers. Provider connectedness, a measure of the relationship among providers, is an important aspect of care coordination that has been linked to higher quality care. We aimed to assess variation in provider connectedness among medical centers, and to understand the association between this established measure of care coordination and outcomes of patients with IBD.

METHODS:

We conducted a national cohort study of 32,949 IBD patients with IBD from 2005 to 2014. We used network analysis to examine provider connectedness, defined using network properties that measure the strength of the collaborative relationship, team cohesiveness, and between-facility collaborations. We used multilevel modeling to examine variations in provider connectedness and association with patient outcomes.

RESULTS:

There was wide variation in provider connectedness among facilities in complexity, rural designation, and volume of patients with IBD. In a multivariable model, patients followed in a facility with team cohesiveness (odds ratio, 0.38; 95% CI, 0.16–0.88) and where providers often collaborated with providers outside their facility (odds ratio, 0.48; 95% CI, 0.31–0.75) were less likely to have clinically active disease, defined by a composite of outpatient flare, inpatient flare, and IBD-related surgery.

CONCLUSIONS:

A national study found evidence for heterogeneity in patient-sharing among IBD care teams. Patients with IBD seen at health centers with higher provider connectedness appear to have better outcomes. Understanding provider connectedness is a step toward designing network-based interventions to improve coordination and quality of care.

Patients with chronic conditions, particularly those with high-intensity specialty care needs, often interact with multiple providers, including primary care providers (PCP) and specialists.¹ These coordination “triads” among patients, PCPs, and specialists are common in the management of inflammatory bowel disease (IBD), a chronic inflammatory condition of the gastrointestinal tract. However, poor coordination between primary and specialty care providers can leave patients with IBD vulnerable to fragmented care and potentially poorer outcomes.^2–6

Provider connectedness, a network analysis measure of the relationship between providers, has been linked to higher quality care and represents a conceptually important aspect of care coordination.^7–9 Network analysis is a method of studying connections among entities and in health care, connections among providers, and how the structure of these connections impacts care quality. The most commonly used way of examining interactions among providers is to measure the number of patients they share, defined as provider connectedness. Provider connectedness is important to understand, particularly in IBD care where coordinated multidisciplinary care is required for management of the complexities of IBD treatment, health care maintenance, and associated health needs. Indeed, prior work noted regional and racial variation in provider connectedness as compared with provider isolation for coronary artery bypass graft surgery and total hip arthroplasty referrals^7,10 and an association between increased provider connectedness and perceived teamwork.¹⁰ Our primary objective was to characterize facilities with a higher level of provider connectedness and to understand the relationship between provider connectedness and IBD patient outcomes using a national cohort of Veterans as a case-model.

Methods

We conducted a retrospective cohort study of patients with IBD managed within the Veterans Health Administration (VHA) to examine provider connectedness. The VHA provides a unique opportunity to study care patterns as a system that has heavily invested in care coordination, while confounders, such as payer mix and institutional incentives, are well-controlled. Although the VHA, as an integrated health system within the United States, has a distinct structure, the problem of coordinating IBD care is universal and impacts other health networks within and outside the United States.¹¹ Therefore, the VHA is an ideal place to start this investigation. Physician connectedness is estimated using network analysis methods, which are increasingly used as an analytic tool in the health industry to characterize social structures.

Data Source and Study Population

The VHA is an integrated system with 130 medical centers that has invested substantial resources in care coordination and has ongoing efforts aimed at improving quality of care. We used the VHA Corporate Data Warehouse to identify patients with IBD who had an outpatient encounter between 2005 and 2014. Patients with IBD were identified with a previously validated algorithm using a combination of inpatient and outpatient International Classification of Diseases, Ninth Revision, Clinical Modification codes for Crohn’s disease (555.x) and ulcerative colitis (556.x).¹² All patients with IBD with at least 2 visit encounters and at least 1 visit with a PCP during the study period were included. Patients were followed for the first 3 years from their initial IBD visit and related data were analyzed.

Network Analysis Methods

A patient-sharing relationship refers to a collaborative relationship between 2 providers, defined as sharing at least 1 common patient. In the language of network analyses describing health care systems, a node often represents individual providers and edges, the patient-sharing relationship between 2 providers. Examining nodes and edges, or providers and patients they share, allows for visualization and quantification of patterns of care, whether they be surgical referral patterns for placement of automatic implantable cardioverter-defibrillators or hospital transfer patterns for critically ill patients.^7,13,14 Visualizing the providers and the patients they share gives the network its structure and that structure is quantified using a variety of network measures. To do so, the first step is to summarize the number of physicians (nodes), physician types (node color), the collaborations (edges) between physicians, the volume of patients that physicians care for (node size), and the number of collaborations (edge thickness) (Figure 1). Physicians with more collaborations are positioned more centrally in the network structure; those who share a higher number of patients are presumed to have a stronger tie, and are positioned closer to each other in a network graph, usually represented by a thicker line. The network structure can visually differentiate communities or clusters of providers, which may represent teams created intentionally or through serendipity.¹⁰ The role and collaboration pattern of each physician and the facility network as a whole can be further described using quantitative measures.

Figure 1.

Provider connectedness. (A) An example of a facility-specific network structure, where nodes are providers (PCP, gastroenterologist, surgeon and edges represent patient-sharing collaborations. (B) Conceptual model of the relationship between providers using network properties.

Mapping a Physician Network

To create a network structure, all gastroenterologists, PCPs, and surgeons were identified and considered as distinct nodes. Each provider was then assigned to a distinct facility based on affiliation and location of a majority of office visits. We then created a physician-level network structure by facility. A patient-provider relationship required at least 1 office visit. Each pair of providers who shared at least 1 common patient were connected using an edge. These edges were weighted-based on the number of shared patients. Trainees and midlevel providers were considered as unique providers in the network.

Characterizing Networks

Although visualization allows for a basic understanding of network organization, facility networks can be quantitatively described using (1) repeated tie fraction, (2) mean number of collaborations, (3) the ratio of patients shared across facilities, and (4) the clustering coefficient:

• Repeated tie fraction is the ratio of the number of collaborations between 2 providers that involve sharing more than a single patient over the total number of collaborations between 2 providers, and ranges from 0 to 1, with 0 representing no repeated collaborations between providers and 1 representing repeated collaborations between all providers in a facility. Repeated tie fraction reflects repeated collaboration between 2 providers over multiple patients that may suggest greater ability to coordinate care effectively.

• The mean number of shared patients in a collaboration measures the strengths of collaborations.

• The ratio of patients shared across facilities to the total number of patients treated in a specific facility ranges from 0 to 1, with 1 representing all patients treated in a single facility are also treated in other facilities. This measure may suggest more effective care coordination at a facility level.

• Clustering coefficient is the extent to which 3 physicians form a clique with pairwise collaborations. It is a measure of the probability that 2 providers in a network, each of whom shared a patient with a common third provider, also shared a patient themselves.^7,15 This describes the multidisciplinary physician structure that promotes teamwork and coordination.

Our primary independent variable, provider connectedness, is a time-invariant facility-level measure and was defined based on previously published studies, using these 4 quantitative network properties that measure collaborative strength (mean number of collaborations, repeated tie fraction), between-facility collaboration (ratio of patients shared across facilities), and team cohesiveness (clustering coefficient).^7,8,10 Each of these 4 properties contribute to the understanding of the relationship between providers (Figure 1).

We explored outcomes as a patient-level variable. The primary outcome was clinically active IBD over the 3-year follow-up period, defined by a composite of the incidence of outpatient flare requiring corticosteroids, inpatient flare, or IBD-related surgery. Outpatient flares were defined using the VHA Decision Support Systems National Data Extract, to identify corticosteroid prescriptions.¹⁶ Patients who received corticosteroid prescriptions for a non-IBD indication in the 7 days before the prescription fill date or prescriptions for less than a 1-week supply were excluded.¹⁶ Episodes involving a Clostridium difficile diagnosis with an associated vancomycin or metronidazole prescription were also excluded.¹⁶ Inpatient flares were defined by a primary or secondary diagnosis of IBD, and receipt of corticosteroid treatment.

Statistical Analysis

First, we described the overall distribution of network properties by facility, and examined correlations between network properties using the correlation coefficients. Six facilities with different characteristics were chosen in a purposeful sampling method to gain a visual understanding of the variation in provider connectedness. Next, we examined the characteristics most strongly associated with different network structures, using a univariable analysis of variance. Finally, we examined the association between provider connectedness (4 facility-level variables) and clinically active IBD (a patient level-variable) using a single multivariable hierarchical regression model and the quality of a model including these measures was confirmed using the Akaike Information Criterion. We controlled a priori for the patient (age, gender, race, IBD subtype, Charlson comorbidity index, use of immunomodulators or biologics) and facility-level (IBD patient volume, region, facility complexity designation, rural/urban commuting area designation) variables.

A sensitivity analysis was performed considering each outcome (outpatient flare, inpatient flare, IBD-related surgery) individually. All analysis was performed using R version 3.6.0 (R Statistical Software, Vienna, Austria). All authors had access to the study data and reviewed and approved the final manuscript. The study protocol was approved by the VA Ann Arbor Health System Institutional Review Board (IRB-2012–179).

Results

The study cohort included 32,949 Veterans; descriptors, including patient demographics and facility characteristics, are reported in Table 1. In this cohort, 5908 (17.9%) were taking immunomodulators or biologic medications, and an outpatient flare requiring corticosteroids affected 311 (16.1%) patients, whereas inpatient flare and IBD-related surgery occurred among 2052 (6.2%) and 839 (2.6%) patients, respectively. These patients received care across the 130 VHA facilities with 3774 (11.5%) patients receiving care from providers in multiple different facilities.

Table 1.

Characteristics of the Study Cohort

Patient-level factors	n (%)
Age, y
<51	9854 (29.91)
51–65	13,121 (39.82)
>65	9974 (30.27)
Gender
Male	30,328 (92.06)
Female	2621 (7.95)
Race
White	25,413 (77.13)
Black	439 (10.44)
Other	843 (2.56)
IBD type
Ulcerative colitis	18,523 (56.22)
Crohn’s disease	13,152 (39.92)
Indeterminate colitis	1274 (3.87)
Charlson comorbidity index (mean, SD)
0	17,201 (52.20)
1	7147 (21.69)
2	3722 (11.30)
>2	4879 (14.81)
Region
Northeast	6812 (20.67)
Midwest	8298 (25.18)
West	1658 (35.38)
South	6181 (18.76)
Immunomodulator or biologic use	5908 (17.93)
No	27,041 (82.07)
Immunomodulator use	4695 (14.25)
Biologic use	2649 (8.04)
Outpatient flare	5311 (16.12)
No	27,638 (83.88)
Hospitalization	2052 (6.23)
No	30,897 (93.77)
Surgery	839 (2.55)
No	32,110 (97.45)
Facility-level factors	n (%)
Facility complexity
Highest	14,038 (42.61)
High	7046 (21.38)
Mid-high	6071 (18.43)
Medium	2621 (7.95)
Low	3160 (9.59)
Rural	1341 (34.42)
Not rural	19,993 (60.68)
Facility-IBD volume
Lowest 25th percentile	3274 (9.94)
25–50th percentile	5837 (17.72)
50–75th percentile	8353 (25.35)
Highest 25th percentile	15,485 (47.00)

IBD, inflammatory bowel disease; SD, standard deviation.

Overall Distribution of Facility-Level Network Properties

Figure 2 describes the wide variation evident across facilities for each property. Correlations among network properties are common. A positive correlation exists between number of patients and providers by type, whereas the number of patients and providers negatively correlate with ratio of patients shared across facilities and the clustering coefficient (Supplementary Figure 1). Repeated tie fractions positively correlate with the mean number of shared patients, and negatively correlate with the ratio of patients shared across facilities.

Figure 2.

Distribution of network properties by facility (y-axis represents unique facilities, and x-axis the quantitative measurement). The mean number of collaborations per provider dyad by facility was 1.40 (standard deviation [SD], 0.18). The repeated tie fraction by facility had a mean of 0.20 (SD, 0.05) with a range of 0 to 1 with 1 representing the case where each collaboration involves at least 2 patients. The mean ratio of patients who were shared across facilities was 0.24 (SD, 0.11). The mean clustering coefficient across facilities was 0.36 (SD, 0.07) on a range of 0 to 1 with 1 representing the case where whenever 2 providers have common collaborator, they themselves collaborate.

We selected facilities with a moderate number of physicians for a more in-depth examination, so that the visualization is easy to understand yet illustrates the variation in network structures (Figure 3). There are differences in the number of providers in each network and their composition, with some facilities composed solely of PCPs, whereas others have specialists, such as gastroenterologists and surgeons. Some facilities have a single gastroenterologist who has a central position among PCPs in the care of patients with IBD (facility B), whereas others have no gastroenterologist and a central surgeon instead (facility D). Some facilities have multiple gastroenterologists and/or surgeons who share patients between themselves, along with PCPs. (facilities C and F).

Figure 3.

A sample of 6 of the 130 facility networks with different patient-sharing structures.

Provider Connectedness and Associated Factors

Associations between provider connectedness and both patient and facility-level factors are common (Table 2). More complex facilities are associated with a lower repeated tie fraction, a higher mean number of shared patients, and a lower clustering coefficient (Table 2). Higher complexity facilities are also associated with a higher ratio of patients shared across facilities, except for the lowest complexity facilities, where sharing patients across facilities is most common. Facilities with a rural designation are more likely to have a lower repeated tie fraction, a lower mean number of shared patients, a higher ratio of patients shared across facilities, and a higher clustering coefficient. Similarly, facilities with the lowest IBD patient volume have a lower mean number of shared patients, a higher ratio of patients shared across facilities, and a higher clustering coefficient (Table 2).

Table 2.

Associations Between Factors and Provider Connectedness by Network Property

	Repeated tie fraction		Mean number of collaborations		Ratio of patients shared across facilities		Clustering coefficient
	Mean	P value	Mean	P value	Mean	P value	Mean	P value
Patient-level factors
Age, y
≤50	0.203	.00	1.425	.54	0.212	0.09	0.332	.00
51–65	0.205		1.427		0.210		0.333
>65	0.207		1.425		0.209		0.336
Gender
Male	0.205	.79	1.426	.07	0.210	.78	0.334	.21
Female	0.205		1.432		0.211		0.333
Race
White	0.207	.00	1.429	.00	0.210	.00	0.335	.00
Black	0.202		1.432		0.203		0.335
Other	0.196		1.409		0.225		0.327
IBD type
Crohn’s disease	0.205	.63	1.426	.16	0.208	.00	0.335	.03
Ulcerative colitis	0.205		1.425		0.212		0.334
Indeterminate colitis	0.204		1.435		0.214		0.331
Charlson comorbidity index
0	0.204	.00	1.424	.00	0.212	.00	0.333	.08
1	0.206		1.426		0.210		0.335
2	0.206		1.426		0.210		0.334
>2	0.207		1.434		0.206		0.335
Immunomodulator or biologic use	0.204	.01	1.428	.26	0.209	.20	0.332	.00
No	0.205		1.426		0.211		0.334
Residing in rural region	0.205	.62	1.403	.00	0.212	.00	0.343	.00
Residing in nonrural region	0.205		1.439		0.209		0.029
Facility-level factors
Region
Northeast	0.179	.00	1.335	.00	0.277	.10	0.354	.63
Midwest	0.182		1.336		0.225		0.346
West	0.224		1.468		0.227		0.355
South	0.220		1.426		0.218		0.369
Facility complexity
Highest	0.190	.03	1.433	.00	0.214	.00	0.318	.00
High	0.196		1.422		0.196		0.341
Mid-high	0.230		1.461		0.191		0.356
Medium	0.215		1.369		0.212		0.388
Low	0.197		1.290		0.355		0.403
Rural	0.197	.61	1.322	.03	0.290	.01	0.405	.00
Not rural	0.203		1.411		0.227		0.346
Facility-IBD volume
Lowest 25th percentile	0.186	.12	1.287	.00	0.311	.00	0.409	.00
25–50th percentile	0.217		1.424		0.214		0.376
50–75th percentile	0.208		1.444		0.216		0.335
Highest 25th percentile	0.200		1.437		0.205		0.302

IBD, inflammatory bowel disease.

In a multivariable model, patients followed in a facility with more between-facility collaborations as defined by higher ratio of shared patients across facilities (adjusted odds ratio [OR], 0.48; 95% confidence interval [CI], 0.31–0.75) and team cohesiveness as defined by a higher clustering coefficient (OR, 0.38; 95% CI, 0.16–0.88) were less likely to have clinically active disease, as defined by a composite of outpatient flare, inpatient flare, and IBD-related surgery (Table 3). In a sensitivity analysis, we see these associations are more likely related to outpatient flares (Table 3). However, a patient cared for in a facility with a higher repeated tie fraction (OR, 0.002; 95% CI, 0.00–0.11) is less likely to undergo an IBD-related surgery. However, facilities with a higher number of mean collaborations were associated with a higher likelihood of IBD-related surgery (OR, 3.46; 95% CI, 1.52–7.81).

Table 3.

Provider Connectedness Properties and Outcomes^a

	Adjusted odds ratio	P value	95% confidence interval
Composite outcome
Repeat tie fraction	0.20	.046	0.04–0.97
Mean number of collaborations	1.57	.009	1.12–2.21
Ratio of shared patients across facilities	0.48	.001	0.31–0.75
Clustering coefficient	0.38	.025	0.16–0.88
Outpatient flare
Repeat tie fraction	0.26	.112	0.05–1.37
Mean number of collaborations	1.34	.122	0.92–1.92
Ratio of shared patients across facilities	0.46	.001	0.29–0.74
Clustering coefficient	0.23	.001	0.09–0.56
Inpatient flare
Repeat tie fraction	0.55	.662	0.04–7.92
Mean number of collaborations	1.68	.071	0.95–2.93
Ratio of shared patients across facilities	0.68	.316	0.32–1.44
Clustering coefficient	1.75	.446	0.41–7.38
IBD-related surgery
Repeat tie fraction	0.002	.002	0.00–0.11
Mean number of collaborations	3.46	.003	1.52–7.81
Ratio of shared patients across facilities	2.36	.121	0.79–6.93
Clustering coefficient	1.58	.681	0.18–13.89

NOTE. Boldface indicates statistical significance.

IBD, inflammatory bowel disease.

^aControlling for patient-level factors (age, sex, race, IBD type, immunomodulator or biologic use, Charlson comorbidity index, and residing in a rural/urban setting) and facility-level factors (rural/urban distinction, number of providers, number of gastroenterologists, region, facility complexity, and IBD volume).

Discussion

Medical centers demonstrate variation in provider connectedness by complexity, rural designation, and IBD volume. Furthermore, patients seen at a facility with a higher provider connectedness are more likely to have good IBD control (free of flares and surgery). Heterogeneity in patient-sharing relationships exists among the major physicians in an IBD care team. This is the first study to examine these patterns of chronic disease multidisciplinary management using network analysis methodology.

Although high-complexity, high-volume facilities play an essential role in tertiary care, these facilities are also characterized by low repeated tie fraction and low clustering coefficients. It is well known that IBD care provided at a specialized center with a high IBD patient volume and subspecialty trained physicians leads to increased adherence to evidence-based IBD practice.¹⁷ Facilities with a higher complexity have the resources to better manage patients with high-intensity specialty care needs, such as endoscopy units and infusion centers, which likely contribute to better clinical outcomes. However, these more complex facilities, with a higher patient volume and greater number of physicians, also are vulnerable to greater difficulty communicating with other providers and dispersion of patient care across several physicians. Low-complexity facilities, on the contrary, are more likely to carry a rural designation, have lower IBD volume, and may not have the resources to manage complex IBD. However, given a lower number of providers, they tend to have higher provider connectedness, in the form of higher repeated tie fraction and clustering coefficient, which may make care coordination an easier task.

Our study suggests that beyond patient and center-specific factors, higher provider connectedness was associated with better IBD outcomes. This is particularly true for patients who received care in facilities with a higher number of between-facility collaborations, which was associated with a higher likelihood of well-controlled disease, and may relate to patients receiving care in more specialized referral centers, although these centers are also more likely to care for the more complex and severe IBD cases. This is also true when considering team cohesiveness (clustering coefficient), a measure of a multidisciplinary physician structure that promotes teamwork and coordination. Providers who are more closely connected, and have a higher clustering coefficient, likely have a stronger comanagement relationship with more frequent communication, trust, and shared goals, leading to better outcomes.¹⁸

IBD requires multidisciplinary care; however, even when patients receive care by key provider, how care is coordinated among these providers is important to consider. Providers who see patients in relative isolation provide less optimal care, even when controlling for facility factors that generally predict isolation. However, an association between a higher mean number of collaborations and a worse disease control is also evident, and seems to be driven by IBD-related surgeries. This may be related to disparities in accessibility to IBD surgeons, who predominantly practice in facilities with a higher complexity.

It is important to consider how best to structure multidisciplinary care teams to facilitate high-quality care. Using this study’s findings to begin to identify which facilities possess certain network structures would allow us to target those where care coordination efforts may have the most benefit. In the VHA, established interventions, such as teleconferencing and electronic consultation, may be adapted to this role. Furthermore, in-depth exploration at the centers with high provider connectedness and optimal IBD patient outcomes may be warranted. This could provide deeper insight as to what facilities with effective care coordination are doing to support complex IBD patient management while also achieving beneficial patient outcomes. Understanding the processes at high provider connectedness facilities could uncover potential interventions to be translated to all VHA for patients with IBD and other chronic care conditions.^19,20 These network-based interventions support change by promoting effective team configurations (eg, building well-connected care teams) or targeting specific members in a network (eg, increasing access to teleconferencing and e-consultation for facilities with low provider connectedness).²¹ Outside the VHA, this could inform the configuration of team-based care, including the team structure of specialty medical home models.

Study Limitations

This study has its limitations including that which comes with using administrative data, including patient-specific details (eg, endoscopic activity, tobacco use), and the use of International Classification of Diseases, Ninth Revision, Clinical Modification codes to identify patients, although our IBD identification algorithm has been well-validated in the VHA.¹² Furthermore, given our patient population, our results may not be generalizable to patients with IBD outside the VHA; although non-VHA networks both within and outside the United States likely possess similar or higher heterogeneity and fragmentation, so the VHA is an ideal place to start this investigation. We are also limited to quantitative methods of describing these static relationships, and both dynamic changes in networks over time and deeper details of the patient-sharing relationships (eg, form of communication or referral agreements) cannot be elucidated using these methods. Additionally, to consider the influence of disease severity and complexity, we control a priori for IBD type and use of immunomodulators and biologics. In regards to our outcome measures, we focus on clinically active IBD rather than other surrogates for low-quality care (eg, inappropriate treatment) that are challenging to define. We also restrict the definition of outpatient flares to those requiring corticosteroids, which may lead to misclassification of biologic treated flares, although this would bias toward the null. Finally, administrative data are limited to the examination of shared patients, the most common way to measure relationships in health care networks; however, it does not necessarily translate to substantial provider interactions.^{7,8,10,21–23} These interactions should be explored in future mixed-methods studies.

Conclusions

Understanding provider connectedness is 1 step toward better understanding care coordination, in efforts to improve quality of care. IBD care varies by facility and provider connectedness within a network, even in an integrated system that promotes care coordination and quality. When considering how to improve care for patients with IBD, the relationship between providers may be a key target for both health systems and payers in the drive toward high-quality care. Future studies should focus on comparing the variability in network structure within the integrated VHA system with nonintegrated health systems both within and outside the United States, understanding the evolution of networks over time, and expanding investigation to other chronic conditions. Perhaps large centers need to consider a model of shared care, where certain physicians take on responsibility for high-intensity patients, with an emphasis on colocalization and standard forms of communication. However, one must better understand the current structures and processes that exist to facilitate care coordination in different facilities.

What You Need to Know.

Background

Inflammatory bowel diseases (IBD) often require multidisciplinary care with tight coordination among providers. Provider connectedness, a measure of the relationship among providers, is an important aspect of care coordination that has been linked with higher quality care.

Findings

A national cohort study found evidence for heterogeneity in patient-sharing among IBD care teams. Patients with IBD at health centers with higher provider connectedness seemed to have better outcomes.

Implications for patient care

Increasing our understanding of provider connectedness could lead to network-based interventions to improve coordination and quality of care for patients with IBD.

Abbreviations used in this paper:

CI

confidence interval

IBD

inflammatory bowel disease

OR

odds ratio

PCP

primary care provider

VHA

Veterans Health Administration