Daniel K. Sewell

Latent Space Models for Dynamic Networks

Dynamic networks are used in a variety of fields to represent the structure and evolution of the relationships between entities. We present a model which embeds longitudinal network data as trajectories in a latent Euclidean space. We propose Markov chain Monte Carlo (MCMC) algorithm to estimate the model parameters and latent positions of the actors in the network. The model yields meaningful visualization of dynamic networks, giving the researcher insight into the evolution and the structure, both local and global, of the network. The model handles directed or undirected edges, easily handles missing edges, and lends itself well to predicting future edges. Further, a novel approach is given to detect and visualize an attracting influence between actors using only the edge information. We use the case-control likelihood approximation to speed up the estimation algorithm, modifying it slightly to account for missing data. We apply the latent space model to data collected from a Dutch classroom, and a cosponsorship network collected on members of the U.S. House of Representatives, illustrating the usefulness of the model by making insights into the networks. Supplementary materials for this article are available online.

Latent space models for dynamic networks with weighted edges

Abstract Longitudinal binary relational data can be better understood by implementing a latent space model for dynamic networks. This approach can be broadly extended to many types of weighted edges by using a link function to model the mean of the dyads, or by employing a similar strategy via data augmentation. To demonstrate this, we propose models for count dyads and for non-negative real dyads, analyzing simulated data and also both mobile phone data and world export/import data. The model parameters and latent actors’ trajectories, estimated by Markov chain Monte Carlo algorithms, provide insight into the network dynamics.

Teamwork on the rocks: Rethinking interprofessional practice as networking

Optimising interprofessional practice has been identified as one of the key methods for improving health outcomes across the globe (Institute of Medicine, 2001). Predominantly, in the United States, better interprofessional practice has been conceptualised as creating high-functioning teams that communicate and collaborate efficiently and effectively to meet the Triple Aim of improved health outcomes, lower cost, and enhanced patient experience (Earnest & Brandt, 2014). Yet, despite nearly two decades of work in the arenas of interprofessional practice and education, progress towards defining impactful, reproducible interventions that improve teamwork and lead to better health outcomes has been slow (Institute of Medicine, 2015; Reeves et al., 2016; Zwarenstein, Goldman, & Reeves, 2009). Recently, we identified some reasons why optimising interprofessional practice might be so challenging. Using data from the electronic health record (EHR), we identified the healthcare professionals involved in the care of a hundred patients with colorectal cancer. Based on data of when and how healthcare professionals access, enter, and review information related to each patient from the date of diagnosis in our university’s cancer registry through 60 days after this date, we created networks of electronic collaboration among the healthcare professionals caring for each patient. The size and complexity of these networks provided some startling insights into the barriers to interprofessional practice which we briefly discuss in this editorial.

The Seasonal Variability of Surgical Site Infections in Knee and Hip Arthroplasty

BACKGROUND Surgical site infections (SSIs) after total knee (TKA) and total hip (THA) arthroplasty are devastating to patients and costly to healthcare systems. The purpose of this study is to investigate the seasonality of TKA and THA SSIs at a national level. METHODS All data were extracted from the National Readmission Database for 2013 and 2014. Patients were included if they had undergone TKA or THA. We modeled the odds of having a primary diagnosis of SSI as a function of discharge date by month, payer status, hospital size, and various patient co-morbidities. SSI status was defined as patients who were readmitted to the hospital with a primary diagnosis of SSI within 30 days of their arthroplasty procedure. RESULTS There were 760,283 procedures (TKA 424,104, THA 336,179) in our sample. Our models indicate that SSI risk was highest for patients discharged from their surgery in June and lowest for December discharges. For TKA, the odds of a 30-day readmission for SSI were 30.5% higher at the peak compared to the nadir time (95% confidence interval [CI] 20-42). For THA, the seasonal increase in SSI was 19% (95% CI 9-30). Compared to Medicare, patients with Medicaid as the primary payer had a 49% higher odds of 30-day SSI after TKA (95% CI 32-68). CONCLUSION SSIs following TKA and THA are seasonal peaking in summer months. Payer status was also a significant risk factor for SSIs. Future studies should investigate potential factors that could relate to the associations demonstrated in this study.

Latent space approaches to community detection in dynamic networks

Embedding dyadic data into a latent space has long been a popular approach to modeling networks of all kinds. While clustering has been done using this approach for static networks, this paper gives two methods of community detection within dynamic network data, building upon the distance and projection models previously proposed in the literature. Our proposed approaches capture the time-varying aspect of the data, can model directed or undirected edges, inherently incorporate transitivity and account for each actors individual propensity to form edges. We provide Bayesian estimation algorithms, and apply these methods to a ranked dynamic friendship network and world export/import data.

Warmer Weather as a Risk Factor for Cellulitis: A Population-based Investigation

Background The incidence of cellulitis is highly seasonal and this seasonality may be explained by changes in the weather, specifically, temperature. Methods Using data from the Nationwide Inpatient Sample (years 1998 to 2011), we identified the geographic location for 773719 admissions with the primary diagnosis (ICD-9-CM code) of cellulitis and abscess of finger and toe (681.XX) and other cellulitis and abscess (682.XX). Next, we used data from the National Climatic Data Center to estimate the monthly average temperature for each of these different locations. We modeled the odds of an admission having a primary diagnosis of cellulitis as a function of demographics, payer, location, patient severity, admission month, year, and the average temperature in the month of admission. Results We found that the odds of an admission with a primary diagnosis of cellulitis increase with higher temperatures in a dose-response fashion. For example, relative to a cold February with average temperatures under 40° F, an admission in a hot July with an average temperature exceeding 90°F has 66.63% higher odds of being diagnosed with cellulitis (95% confidence interval [CI]: [61.2, 72.3]). After controlling for temperature, the estimated amplitude of seasonality of cellulitis decreased by approximately 71%. Conclusion At a population level, admissions to the hospital for cellulitis risk are strongly associated with warmer weather.

Heterogeneous susceptibilities in social influence models

Network autocorrelation models are widely used to evaluate the impact of social influence on some variable of interest. This is a large class of models that parsimoniously accounts for how ones neighbors influence ones own behaviors or opinions by incorporating the network adjacency matrix into the joint distribution of the data. These models assume homogeneous susceptibility to social influence, however, which may be a strong assumption in many contexts. This paper proposes a hierarchical model that allows the influence parameter to be a function of individual attributes and/or of local network topological features. We derive an approximation of the posterior distribution in a general framework that is applicable to the Durbin, network effects, network disturbances, or network moving average autocorrelation models. The proposed approach can also be applied to investigating determinants of social influence in the context of egocentric network data. We apply our method to a data set collected via mobile phones in which we determine the effect of social influence on physical activity levels, as well as classroom data in which we investigate peer influence on student defiance. With this last data set, we also investigate the performance of the proposed egocentric network model.

Weather-Dependent Risk for Legionnaires' Disease, United States

Using the Nationwide Inpatient Sample and US weather data, we estimated the probability of community-acquired pneumonia (CAP) being diagnosed as Legionnaires’ disease (LD). LD risk increases when weather is warm and humid. With warm weather, we found a dose-response relationship between relative humidity and the odds for LD. When the mean temperature was 60°–80°F with high humidity (>80.0%), the odds for CAP being diagnosed with LD were 3.1 times higher than with lower levels of humidity (<50.0%). Thus, in some regions (e.g., the Southwest), LD is rarely the cause of hospitalizations. In other regions and seasons (e.g., the mid-Atlantic in summer), LD is much more common. Thus, suspicion for LD should increase when weather is warm and humid. However, when weather is cold, dry, or extremely hot, empirically treating all CAP patients for LD might contribute to excessive antimicrobial drug use at a population level.

The Landscape of Enteric Pathogen Exposure of Young Children in Public Domains of Low-Income, Urban Kenya: The Influence of Exposure Pathway and Spatial Range of Play on Multi-Pathogen Exposure Risks

Background Young children are infected by a diverse variety of enteric pathogens in low-income, high-burden countries. Little is known about which conditions pose the greatest risk for enteric pathogen exposure and infection. Young children frequently play in residential public areas around their household, including areas contaminated by human and animal feces, suggesting these exposures are particularly hazardous. Objectives The objective of this study was to examine how the dose of six types of common enteric pathogens, and the probability of exposure to one or multiple enteric pathogens for young children playing at public play areas in Kisumu, Kenya is influenced by the type and frequency of child play behaviors that result in ingestion of soil or surface water, as well as by spatial variability in the number of public areas children are exposed to in their neighborhood. Methods A Bayesian framework was employed to obtain the posterior distribution of pathogen doses for a certain number of contacts. First, a multivariate random effects tobit model was used to obtain the posterior distribution of pathogen concentrations, and their interdependencies, in soil and surface water, based upon empirical data of enteric pathogen contamination in three neighborhoods of Kisumu. Then, exposure doses were estimated using behavioral contact parameters from previous studies, and contrasted under different exposure conditions. Results Multi-pathogen exposure of children at public play areas was common. Pathogen doses and the probability of multi-pathogen ingestion increased with: higher frequency of environmental contact, especially for surface water; larger volume of soil or water ingested; and with play at multiple sites in the neighborhood versus single site play. Discussion Child contact with surface water and soil at public play areas in their neighborhood is an important cause of exposure to enteric pathogens in Kisumu, and behavioral, environmental, and spatial conditions are determinants of exposure.

Simultaneous and temporal autoregressive network models

While logistic regression models are easily accessible to researchers, when applied to network data there are unrealistic assumptions made about the dependence structure of the data. For temporal networks measured in discrete time, recent work has made good advances \citep{almquist2014logistic}, but there is still the assumption that the dyads are conditionally independent given the edge histories. This assumption can be quite strong and is sometimes difficult to justify. If time steps are rather large, one would typically expect not only the existence of temporal dependencies among the dyads across observed time points but also the existence of simultaneous dependencies affecting how the dyads of the network co-evolve. We propose a general observation driven model for dynamic networks which overcomes this problem by modeling both the mean and the covariance structures as functions of the edge histories using a flexible autoregressive approach. This approach can be shown to fit into a generalized linear mixed model framework. We propose a visualization method which provides evidence concerning the existence of simultaneous dependence. We describe a simulation study to determine the methods performance in the presence and absence of simultaneous dependence, and we analyze both a proximity network from conference attendees and a world trade network. We also use this last data set to illustrate how simultaneous dependencies become more prominent as the time intervals become coarser.