Assessment of Hospital Characteristics and Interhospital Transfer Patterns of Adults With Emergency General Surgery Conditions

Key Points Question Are interhospital transfers of patients with emergency general surgery (EGS) conditions consistently directed to hospitals with more resources and better outcomes? Findings In this cohort study using network analysis of interhospital transfers of 80 307 adults with EGS conditions, transfers were consistently directed to high-volume hospitals with more resources. However, transfers were not necessarily directed to hospitals with better outcomes (ie, lower risk-adjusted failure to rescue and lower in-hospital mortality). Meaning The results of this study suggest that opportunities exist to improve the selection of transfer destinations for patients with EGS conditions, with potential to improve clinical outcomes for transferred patients.


Introduction
More than 3 million admissions of patients with emergency general surgery (EGS) conditions occur annually in the US, accounting for 7.1% of all hospitalizations nationally. 1,2 Emergency general surgury conditions are also associated with high mortality and costs, placing substantial burden on the health care system. 1,[3][4][5][6][7][8][9][10][11] Interhospital transfers of patients with EGS conditions are common, occurring in up to 13% of all EGS encounters, but little is known about the transfer patterns within the network. 1,12 Acute care hospitals with EGS services are not evenly distributed with regard to population density or need for care and, at present, there are no standardized guidelines in place to direct patients to those hospitals. 1,11,12 This problem creates gaps in access to EGS care that have disproportionate consequences for underserved and rural communities, furthering a need for standardized interhospital transfer guidelines and quality measures. 6,7,12 Benefits of organized regionalization of care that includes standardized triage guidelines and verified specialty-specific centers of excellence have been found for trauma and complex oncologic surgeries. [13][14][15] To design successful EGS care delivery, a better understanding of current transfer patterns and associated hospital characteristics is needed to identify targets for quality improvement, including tools to aid transfer decision-making and the selection of destination hospitals. 12,16 One way to examine existing EGS transfer patterns and associated hospital factors is through network analysis. Network analysis has been used to model the spread of infection in hospital social networks and to examine interhospital transfers of patients with myocardial infarction or critical care needs. [17][18][19][20][21][22][23][24] In a network analysis framework, the extent to which patients are consistently transferred to high-resourced, high-volume, and high-performing hospitals would be both a measure of the network's current performance and a strategy for identifying inefficiencies to improve outcomes through regional care delivery. Therefore, we assessed interhospital transfer patterns for EGS in the US using a network analysis approach. We specifically examined the association between hospital characteristics (eg, size, resources, EGS volume, and outcomes) and transfer patterns within the network. We hypothesized that in a successful regionalized network, interhospital transfers of patients with EGS conditions would be directed to hospitals with high resources, volume, and performance.

Data Source
This study was approved by the University of Pittsburgh Institutional Review Board and deemed exempt from human subjects review under exemption 4 of US Code of Federal Regulations because the study used data that were publicly available or recorded in a manner in which participants could not be identified. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. 25 We performed a retrospective cohort study using all-payer claims data from the 2016 Healthcare Cost and Utilization Project (HCUP) state inpatient and emergency department (ED) databases in 8 states (Arkansas, Florida, Maryland, Massachusetts, Nebraska, New York, Vermont, and Wisconsin). These states were chosen because they provide HCUP with unique patient identifiers that enabled us to track patients across hospitals over time. All inpatient hospital stays with discharge dates between January 1 and December 31, 2016, including inpatient admissions from the ED, are captured in the state inpatient databases 26 and were defined as inpatient encounters. All ED visits between January 1 and December 31, 2016, that did not result in an admission at the same hospital are captured in the state ED databases 26 and were defined as ED-only encounters. Both databases include patient demographic characteristics, admission and discharge times, diagnostic and procedural codes from the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM), patient outcomes, and hospital identifiers. 27

Identification of EGS Episodes of Care
We defined EGS episodes of care as temporally adjacent ED-only encounters and inpatient encounters that occurred in different hospitals, in which the inpatient encounter included at least 1 diagnosis of an EGS condition as defined by the American Association for the Surgery of Trauma. 3 Based on published methods, patient encounters were considered temporally adjacent if the discharge date of the first hospital encounter was on the same day or the previous day of the admission date of the second hospital encounter. 17 We used general equivalence mappings from the Centers for Medicare & Medicaid Services to convert previously defined EGS diagnostic codes 3 from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) to the ICD-10-CM because the 2016 HCUP data included ICD-10-CM codes only. 27 We included EGS episodes for patients 18 years and older that involved interhospital transfers from the ED (ie, ED to inpatient transfers) or the inpatient setting (ie, inpatient to inpatient transfers).
The ED to inpatient transfers included ED to ED transfers that resulted in inpatient admission at the destination hospital. We identified operations and procedures based on ICD-9-CM procedural codes defined by the HCUP Surgery Flags software. 28 Because EGS care includes both nonoperative and operative management as well as care received at both pretransfer and posttransfer facilities, patients who were operatively and nonoperatively managed were both included.

Network Analysis of Hospital Centrality Measures
A network consists of nodes and directional links between nodes. 17,21,24,[29][30][31] In our networks, hospitals were considered nodes, and interhospital transfers were considered directional links between nodes, weighted by the number of transfers between the hospitals. 17,29-31 We constructed 5 separate networks: (1) all transfers, (2) ED to inpatient transfers only, (3) inpatient to inpatient transfers only, (4) all transfers involving at least 1 operation or procedure, and (5) all transfers not involving any operation or procedure. For each network, we defined 3 centrality variables for each hospital: in-degree centrality (the normalized number of links directed to a node), out-degree centrality (the normalized number of links directed from a node), and centrality ratio (in-degree centrality divided by out-degree centrality, in which higher values indicated more incoming transfers per outgoing transfer) 29-32 (Figure 1). Centrality measures were normalized by the maximum possible number of links based on the number of nodes in the network. In-degree and out-degree centrality measures ranged from 0 to 1 and were scaled by multiplying by 10 000. Centrality measures were calculated based on standard methods used for network analysis. 30-32 Based on standard definitions, in-degree centrality was calculated as the summation of links directed to a node in a given network, normalized by the maximum number of such links. Out-degree centrality was calculated as the summation of links directed away from a node in a given network, normalized by the maximum number of such links.
Centrality ratio was calculated as in-degree centrality divided by out-degree centrality.

Size and Resources
We obtained measures of hospital size and resources using the American Hospital Association Annual Survey, 33 and we obtained trauma center designations from the trauma center listing of the American College of Surgeons. 34 We identified the presence of trauma and/or surgical critical care fellowships using the fellowship listing from the Eastern Association for the Surgery of Trauma and

EGS Volume and Outcomes
We defined hospital EGS volume as the total number of inpatient EGS episodes that did not result in transfer within 1 calendar year. We defined hospital outcomes as risk-adjusted in-hospital mortality and failure to rescue for EGS episodes. 39,40 Failure to rescue has been recognized by the National Quality Forum as a surgical quality measure. 14 drainage) based on ICD-9-CM procedure codes defined by the HCUP Surgery Flags software. 28 We

Statistical Analysis
We reported patient demographic characteristics and descriptive statistics of the overall EGS interhospital transfer networks. Using 1-way analysis of variance and post hoc pairwise comparisons, we compared hospital centrality ratio across the other hospital characteristics. For the primary analysis, we performed multivariable regression modeling with clustering by state, and we conducted partial F tests to examine the association of centrality ratio with measures of hospital resources, volumes, and outcomes (reported as β coefficients with 95% CIs). The centrality ratio was logarithmically transformed before conducting the analysis of variance and multivariable regression analysis because of nonnormal distribution. We performed sensitivity analyses using the same multivariable regression model for the following subnetworks: (1) ED to inpatient transfers, (2) inpatient to inpatient transfers, (3) transfers involving at least 1 operation or procedure, and (4) transfers not involving any operation or procedure.  difference, 3.61 [95% CI, 3.11-4.10]; P < .001), which indicated that they received more incoming transfers per outgoing transfer (Figure 2 and Figure 3) (Figure 2 and Figure 3). In the post hoc pairwise comparison, hospitals in successively higher in-hospital mortality quartiles (ie, quartile 1    Table 2).

JAMA Network Open | Surgery
The sensitivity analyses yielded mostly similar results for the EGS interhospital transfer networks, regardless of transfer origin (ED vs inpatient) or the involvement of an operation or procedure. For all 4 subgroups (ED to inpatient, inpatient to inpatient, with operation or procedure, and without operation or procedure), a higher log-transformed centrality ratio was associated with greater resources, including more ICU beds and teaching hospital status (eTable 1 and eTable 2 in the Supplement). Of note, higher EGS volume was not associated with a higher log-transformed centrality ratio for ED to inpatient transfers, but the volume-centrality ratio association was maintained for inpatient to inpatient transfers (eTable 1 in the Supplement). In addition, higher EGS volume was not associated with a higher log-transformed centrality ratio for transfers involving operations or procedures, but the volume-centrality ratio association was maintained for transfers not involving an operation or procedure (eTable 2 in the Supplement). Nevertheless, for all subgroups, a higher log-transformed centrality ratio was again not associated with better outcomes (ie, lower failure to rescue and lower in-hospital mortality) (eTable 1 and eTable 2 in the Supplement).

Discussion
This cohort study was based on the concept that understanding existing interhospital transfer patterns is an important first step to guide targeted improvement strategies for the design of successful regionalized care and to improve outcomes for EGS. Yet the interhospital transfer network for EGS is not well understood, including the extent to which patients are transferred to hospitals with high resources, volume, and performance, as expected in a well-coordinated regionalized network. Applying a network analysis approach, we found that patients with EGS conditions were generally transferred to hospitals with high resources and volume, as expected. However, patients with EGS conditions were not consistently transferred to hospitals with better outcomes, as measured by lower risk-adjusted EGS in-hospital mortality and failure to rescue. These findings highlight the need for further research into the appropriate selection of transfer destinations as a strategy to improve EGS outcomes. 16 Network analysis has been used to examine overall interhospital transfers, ICU transfers, and transfers of patients with acute myocardial infarction, [17][18][19][20][21]31 but the approach in the examination of interhospital transfers for EGS is, to our knowledge, novel. In a regional critical care transfer network, higher centrality was associated with more hospital resources. 17 Our results revealed parallel associations between a higher centrality ratio and greater hospital resources and EGS volume in the EGS transfer network. However, hospitals that received more incoming transfers per outgoing transfer (ie, hospitals with a higher centrality ratio) were not always those with better risk-adjusted EGS outcomes.  Our measures of hospital outcomes might be subject to discharge bias, a phenomenon in which patients with complex medical conditions are transferred before death, which artificially lowers mortality rates. 56-58 However, most EGS episodes of care (93.6%) used to calculate our hospital-level risk-adjusted outcomes were episodes that did not involve interhospital transfer, making discharge bias less applicable. In addition, because each hospital may have different established interinstitutional relationships and collaboration with local emergency transport services, further region-specific research on transfer guidelines, and multidisciplinary coordination among clinicians, hospitals, and call centers is needed to tailor interventions to their respective health care networks.

Conclusions
Interhospital transfers in EGS were consistently directed to high-volume hospitals with more resources but not necessarily to hospitals with better risk-adjusted outcomes. Further research to develop guidelines and infrastructure to improve transfer decisions and the appropriate selection of transfer destinations on both the clinician and health care system levels is needed and could improve outcomes for patients with EGS conditions.