Network Methods in Biomedical Research

Methods from psychology and community ecology enhance network models of comorbidity

Networks are increasingly used to organize and analyze large numbers of comorbid relations between diseases. These epidemiological comorbidities are usually detected and quantified from healthcare data based on their co-occurrences in large populations. Most comorbidity networks are built up from comorbid pairs—that is, the statistical tests of comorbidity are bivariate and do not take other possibly antecedent or mediating diseases into account.

We compared pairwise correlation models to two multivariate alternatives adapted from psychology and community ecology: partial correlation models and joint incidence–distribution models. These comparisons revealed significant differences both in what comorbidites are statistically detectable and in how whole comorbidity networks are structured. The models have strengths and weaknesses, and we recommend that the choice of model be motivated by research questions.

Network Comorbidity Models

Four network comorbidity models based on disease co-occurrence detected in the National Ambulatory Medical Care Survey (2011). Solid (respectively, dashed) links indicate positive (negative) associations.

Network Comorbidity Models

Integral calculus reveals the spatial organization of glomerular capillaries

The blood supply is filtered of excess and waste in capillary tufts called glomeruli, the main components of kidney nephrons. Most kidney disease is caused by obstruction or destruction of glomerular vessels, so glomerular structure is important to renal pathology. Recent developments in serial-section scanning electron microscopy and virtual reality have reduced the cost of reconstructing glomerular networks, and we have used spatial graph models to gain insight into their spatial and topological structure.

For example, using circuit analysis, centrality analysis, and the fundamental theorem of calculus, we are able to trace an average blood flow through each glomerulus, forming a sort of “average” path from afferent (incoming) to efferent (outgoing) arteriole. These average paths exhibit a consistent cul-de-sac pattern, with the implication that arteriolar paths avoid direct routes between the arterioles near the vascular pole, consistent with the expectation that such “shortcuts” would allow blood to pass without filtration. They also suggest that, on average, glomeruli exhibit lateral symmetry with respect to the plane defined by the arterioles and the vascular and tubular poles. These and other properties will inform our understanding of how glomeruli function when only partially obstructed or destroyed.

Average Path

A spatial graph model of a mouse glomerulus (black) with its average flow (purple).

Trajectory

Research Support

Jason Cory Brunson received postdoctoral support from the UConn–NIDCR T90/R90 Research Training Program, National Institutes of Health, through grant number 5T90DE021989-07 (PI: Mina Mina); 7/1/2017-04/30/2020

Selected Publications

Jason Cory Brunson, Reinhard C Laubenbacher (2018) Applications of network analysis to routinely collected health care data: a systematic review. Journal of the American Medical Informatics Association 25(2): 210–221, doi: 10.1093/jamia/ocx052

Jason Cory Brunson, Thomas P Agresta, Reinhard C Laubenbacher (2020) Sensitivity of comorbidity network analysis. JAMIA Open 3(1): 94–103, doi: 10.1093/jamiaopen/ooz067

Mark Terasaki, Jason Cory Brunson, Justin Sardi (2020) Analysis of the three dimensional structure of the kidney glomerulus capillary network. Under review. bioRxiv doi: 10.1101/677864

Jason Cory Brunson, Morphological symmetry of murine glomeruli. In preparation.

Luis Sordo Vieira, Jason Cory Brunson, Energy values of minimal functional routes. In preparation.