Potential value of health information exchange for people with epilepsy: crossover patterns and missing clinical data

Abstract

Context: For people with epilepsy, the potential value of health information exchange (HIE) is unknown.

Methods: We reviewed two years of clinical encounters for 8055 people with epilepsy from seven Manhattan hospitals. We created network graphs illustrating crossover among these hospitals for multiple encounter types, and calculated a novel metric of care fragmentation: "encounters at risk for missing clinical data."

Results: Given two hospitals, a median of 109 [range 46 - 588] patients with epilepsy had visited both. Due to this crossover, recent, relevant clinical data may be missing at the time of care frequently (44.8% of ED encounters, 34.5% inpatient, 24.9% outpatient, and 23.2% radiology). Though a smaller percentage of outpatient encounters were at risk for missing data than ED encounters, the absolute number of outpatient encounters at risk was three times higher (14,579 vs. 5041).

Conclusion: People with epilepsy may benefit from HIE. Future HIE initiatives should prioritize outpatient access.

Figure 1.

Technique to create hospital networks. In this simplified example, imagine a network of four hospitals (A – D) visited by six patients. Panel A. List the patients and mark which hospitals each has visited. Panel B. Turn the list into a “patient-hospital” matrix, with four columns (one for each hospital) and enough rows for each patient. “1” indicates a patient visited that hospital; otherwise enter “0”. Panel C. Multiply the patient-hospital matrix by its transpose to create a hospital network matrix. The rows and columns each correspond to the list of hospitals. Each off-diagonal cell (black) represents the number of patients who visited both the row hospital and the column hospital. Panel D. These values can be tabulated to show the number of patients who visited pairs of hospitals. Panel E. Plot the values as a network graph. Each vertex represents a hospital, and the thickness of the connection between each pair scales with the number of people. In the language of graph theory, this is mathematically equivalent to transforming a bipartite graph (patients and hospitals) into a unipartite graph (hospitals).

Figure 2.

Hospital Network Graphs Illustrating Crossover Patterns for 8055 People with Epilepsy for Eight Encounter Types Over Two Years. Each circle represents one of seven Manhattan hospitals. The thickness of the line connecting two circles represents the number of people with epilepsy who were seen at both hospitals, for the given encounter type, over two years. The “All Visits” graph includes any clinical or radiology encounter. The other seven graphs only include encounters of the given type. The connectedness of each graph (i.e. the network density) represents the percent of all hospital pairs that shared patients. For each graph, the median and range of patients shared across all 21 hospital pairs are also presented.