An analytical approach to characterize morbidity profile dissimilarity between distinct cohorts using electronic medical records

Abstract

We describe a two-stage analytical approach for characterizing morbidity profile dissimilarity among patient cohorts using electronic medical records. We capture morbidities using the International Statistical Classification of Diseases and Related Health Problems (ICD-9) codes. In the first stage of the approach separate logistic regression analyses for ICD-9 sections (e.g., "hypertensive disease" or "appendicitis") are conducted, and the odds ratios that describe adjusted differences in prevalence between two cohorts are displayed graphically. In the second stage, the results from ICD-9 section analyses are combined into a general morbidity dissimilarity index (MDI). For illustration, we examine nine cohorts of patients representing six phenotypes (or controls) derived from five institutions, each a participant in the electronic MEdical REcords and GEnomics (eMERGE) network. The phenotypes studied include type II diabetes and type II diabetes controls, peripheral arterial disease and peripheral arterial disease controls, normal cardiac conduction as measured by electrocardiography, and senile cataracts.

Figure 1

Example Morbidity Dissimilarity Indices (MDI) for four configurations. MDIs were drawn for $({\sigma }_1^2/{\sigma }_2^2,\rho )$ equal to (1, 0), (3,0), (1, 0.75), and (3, 0.75) in panels a), b), c), and d), respectively, where ${\sigma }_1^2=\mathit{\text{Var}}({\beta }_1)\text{ and }{\sigma }_2^2=\mathit{\text{Var}}({\beta }_2)$ and ρ is the correlation between β₁ and β₂. Different values of correlations (ρ) effectively alter the angle between the axes, as shown in panels c and d. The solid and dashed contours display the set of all (β₁, β₂), that yield MDI equal to 0.5 and 1.0, respectively.

Figure 2

Bland-Altman plots comparing unadjusted rates of ICD-9 categories, sections, and chapters for pairs of populations.

Figure 3

Adjusted odds ratios based comparing VU-T2D to VU-CON, NU-T2D to NU-CON, and MAY-PAD to MAY-CON. The symbol “X” denotes an extremely high odds ratio whose lower confidence limit exceeds 20.

Figure 4

Adjusted odds ratios based comparing GHC-CAT to MAR-CAT, NU-T2D to VU-T2D, and MAY-PAD to VU-QRS. The symbol “X” denotes an extremely high odds ratio whose lower confidence limit exceeds 20.