Fig 1.
Overview of data and analysis.
(A) Preprocessing of data and overview of analysis methods. We prepared the primary data set for the US by combining five data sets of the USSID, which are released annually (2006~2010). Similarly, the National Inpatients Set of Korea (NISK) was utilized to integrate three sets of the released version from 2009 to 2011. NISK covers all of the outpatient records who have admission records in a year. (B-C) Overall modality of utilized USSID (B) and NISK (C).
Table 1.
Data statistics of State Inpatient Dataset of California, USA (USSID)* and National Inpatient Set of Korea (NISK)**.
Fig 2.
Distributions of diagnoses and associated fatal outcomes.
(A) Distribution of diagnoses in the USSID by age. The bottom plot presents the distribution of the diagnoses with fatal outcomes from the USSID. (B) Distribution of diagnoses in the NISK by age. The bottom plot presents the distribution of the diagnoses with fatal outcomes among the discharges from the NISK.
Fig 3.
Trajectory of disease diagnosis and death.
(A) Conceptual introductions to visualize timelines from disease (a circle node) to death (a square node). The linked edge (i.e. the line) between two nodes denotes significant transitions in patients within a year (FDR <0.1). (B) A simplified example of disease-to-death trajectory based on following patients from the start of the disease diagnosis.
Fig 4.
Trajectory of disease and death.
(A) The length distributions of 300 identified trajectories for 59,794 deaths in the USSID. (B) The length distributions of 405 identified trajectories in the NISK. The aligned histograms of diagnosis proportion presented by initial and final diagnoses with death outcomes. Each color denotes the type of disease, as determined by ICD codes.
Fig 5.
Decomposing diagnosis trajectory as a pairwise network.
The traced trajectories in the USSID and NISK were decoupled into a pairwise network consisting of nodes (disease and death) and edges (directed associations). (A-B) Overview of a network of disease diagnoses and associated death outcomes in the USSID (A) and NISK (B).
Fig 6.
Topological features of a pairwise network of diagnosis and death.
(A-B) Distributions of patient age by the type of edges including disease pairs and death-related edges in the USSID. (D-E) Distributions of patient age by the type of edges including disease pairs and death-related edges in the NISK. (C-F) Bubble plot of diagnoses nodes by the topological features. The x-axis indicates number of outlined edges, and y-axis presents degrees of in-linked edges.
Fig 7.
Risk factors of lethal diagnoses using the pairwise network in the USSID.
The decomposed network of diagnoses identified a prior diagnoses of the diseases. (A) Detected risk factors for the top-ranked diagnosis using the Case Fatality Ratio (CFR). (B) Detected risk factors for lethal diagnoses (i.e., top-ranked CFRs) in the elderly. (C) Detected risk factors for lethal diagnoses (i.e., top-ranked CFRs) in younger patients.
Fig 8.
Risk factors of lethal diagnoses using the pairwise network in the NISK.
(A) Detected risk factors for prioritized diagnosis using Case Fatality Ratio (CFR). (B) Detected risk factors for lethal diagnoses (i.e., strong CFRs) in the elderly. (C) Detected risk factors for lethal diagnoses (i.e., top ranked CFRs) in younger patients.