Fig. 1: Functional principal component analysis of COVID-19 patients’ C-reactive protein levels. | Nature Immunology

Fig. 1: Functional principal component analysis of COVID-19 patients’ C-reactive protein levels.

From: A patient-centric modeling framework captures recovery from SARS-CoV-2 infection

Fig. 1

a, The first two eigenfunctions represent the severity of inflammation and the recovery from inflammation, respectively, over the 7-week window of the FPCA, accounting for 78.5% and 20.8% of the variability, respectively. b, Scatterplot showing the FPCA scores, FPC1 and FPC2, corresponding to the first and second eigenfunctions, respectively; each point corresponds to one CovP. x axis indicates FPC1 ‘severity’ scores (the higher the more severe the inflammation); y axis indicates FPC2 ‘recovery’ scores (the higher the more pronounced the temporal resolution of inflammation). ‘Recovery’ groups 1, 2 and 3 were obtained by Gaussian mixture modeling; the opacity and diameter of the points are proportional to the estimated probability of assignment of each CovP to their recovery group. Log-transformed CRP trajectories for four CovPs with extreme severity or recovery scores (left and right). Gray bands, normal CRP levels corresponding to the IQR of HCs’ (log-transformed) CRP levels. Points correspond to the observed values; red, blue and gray curves, trajectories estimated using the FPCA framework; dashed curves delineate the 95% confidence bands (n = 113 CovPs). c, FPCA scores and groups by CovP clinical severity classes (B, screening symptomatic; C, hospital no oxygen required; D, hospital supplemental oxygen; E, hospital assisted ventilation). In the box plots, the center line indicates the median, box limits represent the upper and lower quartiles and whiskers indicate 1.5 times the IQR. One-versus-all two-sided t-tests: ****P < 0.0001, ***P < 0.001, **P < 0.01 and *P < 0.05; overall: analysis of variance (ANOVA) P < 0.0001, n = 113 CovPs. d, CRP trajectories conditional on the recovery groups 1–3 with 95% confidence bands, estimated with a longitudinal mixed model accounting for patients’ repeated measurements (likelihood ratio (LR) tests for the baseline group effect and group × time interaction effect, significance labels as in c). Points correspond to observed values. Group 1, inflammation absent or mild; group 2, early, resolving inflammation; group 3, persisting inflammation. e, Characterization of the recovery groups by age (one-versus-all two-sided t-tests, ANOVA P < 0.0001) and gender (Fisher exact test P = 0.0001). Box plots and significance labels as in c (n = 113 CovPs).

Back to article page