99 - Association of network structure with acute infant transport

Publication Number: 3095.99

Jochen Profit, Stanford University, Palo Alto, CA, United States; Daniel Helkey, Stanford University School of Medicine, Reno, NV, United States; Sarah Kunz, Boston Children's Hospital, Boston, MA, United States; Molly Passarella, Childrens Hospital of Philadelphia, Philadelphia, PA, United States; Michelle Dai, Harvard Medical School, Somerville, MA, United States; Ciaran Phibbs, Stanford University School of Medicine, Stanford, CA, United States; Scott A. Lorch, The Children's Hospital of Philadelphia, University of Pennsylvania, Philadelphia, PA, United States; Marinka Zitnik, Harvard Medical School, Boston, MA, United States; Jeannette A. Rogowski, The Pennsylvania State University, University Park, PA, United States

Background: Acute infant transport has been associated with adverse infant outcomes. The structure of regionalized perinatal networks varies widely between birth locations.

Objective: Test whether network features of shape and structure are independently and jointly associated with likelihood of acute infant transport.

Design/Methods: We identified very preterm ( < 32 weeks) infants using linked infant records from 2000 to 2019 for five states. We excluded non-acute transports and transports occurring outside state-lines.
We computed network features for each two-year period by identifying local networks centered on AAP Level 3+ hospitals in the top 90th percentile of annual acute transports and including all connected transfer hospitals. We calculated network centrality, density, and modularity for each network. Centrality is the tendency of networks to transport directly to the central hospital. Density measures the proportion of network hospitals connected, and modularity measures the tendency of hospitals with each local network to cluster together.
We used Poisson regression to quantify the effect of each risk-adjustor, with acutely transported infants as the outcome of interest and state included as a fixed-effect. Risk factors were grouped by network, hospital, infant, and maternal socio-economic. We computed relative risk ratios and 95% confidence intervals, adding variables by category into the regression sequentially starting with network factors (Figure 1). Coefficients for the fully-adjusted model were used to describe significant associations (Table 2).

Results: The study included 163,850 live born infants between five states (Table 1). Risk-factors of all categories were significantly associated with likelihood of acute transport. Hospital and infant level factors were most strongly associated with likelihood of transport.
In the joint regression, network factors remained significantly associated with likelihood of acute transport, with effect sizes from 3-12% per unit standardized increase in network metric (Table 2). Higher centrality was associated with increased likelihood of transport. Higher network density and modularity were both associated with a lower likelihood of acute transport. Likelihood of transfer and the effects of covariates on transfer likelihood vary significantly by state (Figure 1).

Conclusion(s): Attributes of neonatal regionalization as measured by network analysis are associated with the likelihood of acute transfer. Lower likelihood of transfer in higher density and modularity networks may indicate improved antenatal selection of birth location.

Table 1
Characteristics of the study population, liveborn infants and any associated acute neonatal transports for infants born in 5 states between 2000-2019.

Table 2
Relative risk ratios and significance of risk-adjustors in the joint regression. Comparison categories for categorical risk factors are 0-15 annual VPT births, birth hospital AAP level IIIa, no high school, gestational age 31 weeks, private health insurance, and maternal age 25-34, White mothers, and the state with the highest observed transfer rate.

Figure 1
Relative risk ratios for each state compared with the state with the highest observed transfer rate. Sequential risk-adjustment, going from the unadjusted model, model with only network factors, and model adjusting for all factors (network, hospital, infant, and maternal socio economic).