Phenotypic Classification of Multisystem Inflammatory Syndrome in Children Using Latent Class Analysis

Abstract

Importance: Multisystem inflammatory syndrome in children (MIS-C) is an uncommon but severe hyperinflammatory illness that occurs 2 to 6 weeks after SARS-CoV-2 infection. Presentation overlaps with other conditions, and risk factors for severity differ by patient. Characterizing patterns of MIS-C presentation can guide efforts to reduce misclassification, categorize phenotypes, and identify patients at risk for severe outcomes.

Objective: To characterize phenotypic clusters of MIS-C and identify clusters with increased clinical severity.

Design, setting, and participants: In this cohort study, MIS-C phenotypic clusters were inferred using latent class analysis applied to the largest cohort to date of cases from US national surveillance data from 55 US public health jurisdictions. Cases reported to the Centers for Disease Control and Prevention MIS-C national surveillance program as of April 4, 2023, with symptom onset on or before December 31, 2022, were retrospectively analyzed. Twenty-nine clinical signs and symptoms were selected for clustering after excluding variables with 20% or more missingness and 10% or less or 90% or more prevalence. A total of 389 cases missing 10 or more variables were excluded, and multiple imputation was conducted on the remaining cases.

Main outcomes and measures: Differences by cluster in prevalence of each clinical sign and symptom, percentage of patients admitted to the intensive care unit (ICU), length of hospital and ICU stay, mortality, and relative frequency over time.

Results: Among 8944 included MIS-C cases (median [IQR] patient age, 8.7 [5.0-12.5] years; 5407 [60.5%] male), latent class analysis identified 3 clusters characterized by (1) frequent respiratory findings primarily affecting older children (respiratory cluster; 713 cases [8.0%]; median [IQR] age, 12.7 [6.3-16.5] years), (2) frequent shock and/or cardiac complications (shock and cardiac cluster; 3359 cases [37.6%]; median [IQR] age, 10.8 [7.7-14.0] years), and (3) remaining cases (undifferentiated cluster; 4872 cases [54.5%]; median [IQR] age, 6.8 [3.6-10.3] years). The percentage of patients with MIS-C admitted to the ICU was highest for the shock and cardiac cluster (82.3% [2765/3359]) followed by the respiratory (49.5% [353/713]) and undifferentiated clusters (33.0% [1609/4872]). Among patients with data on length of stay available, 129 of 632 hospitalizations (20.4%) and 54 of 281 ICU stays (19.2%) in the respiratory cluster lasted 10 or more days compared with 708 of 3085 (22.9%) and 157 of 2052 (7.7%), respectively, in the shock and cardiac cluster and 293 of 4467 (6.6%) and 19 of 1220 (1.6%), respectively, in the undifferentiated cluster. The proportion of cases in both the respiratory cluster and the shock and cardiac cluster decreased after emergence of the Omicron variant in the US.

Conclusions and relevance: In this cohort study, MIS-C cases clustered into 3 subgroups with distinct clinical phenotypes, severity, and distribution over time. Use of clusters in future studies may support efforts to evaluate surveillance case definitions and identify groups at highest risk for severe outcomes.

Figure 1.. Prevalence of SARS-CoV-2 Test Results, Organ System Involvement, and Common Respiratory and Cardiac Presentations by Latent Class Analysis–Inferred Clusters

Prevalence was calculated based on nonmissing data except for cardiac dysfunction, for which prevalence was calculated based on all patients in a cluster, because missingness of echocardiographic results was likely not at random. Sample sizes in cluster labels indicate total numbers of patients within each cluster. Cardiac dysfunction was assessed by left or right ventricular dysfunction on echocardiogram, and shock was indicated in a clinical note or receipt of vasopressors. Cluster 1 indicates the respiratory cluster; cluster 2, shock and cardiac cluster; and cluster 3, undifferentiated cluster. ARDS indicates acute respiratory distress syndrome; BNP, B-type natriuretic peptide; NT-proBNP, N-terminal prohormone of brain natriuretic peptide; RT-PCR, reverse transciptase–polymerase chain reaction.

Figure 2.. Distributions of Levels of B-Type Natriuretic Peptide (BNP), Troponin, and C-Reactive Protein by Latent Class Analysis–Inferred Clusters

Distributions are scaled to have the same area. Horizontal lines indicate medians; edges of boxes, IQRs; and whiskers, upper and lower adjacent values. Sample sizes indicate the numbers of patients within each cluster with nonmissing data. Global differences across clusters were statistically significant (P < .001) as assessed using the Kruskal-Wallis test for continuous variables for all 3 clinical biomarkers. Cluster 1 indicates the respiratory cluster; cluster 2, shock and cardiac cluster; and cluster 3, undifferentiated cluster. To convert BNP to nanograms per liter, multiply by 1; to convert troponin to micrograms per liter, multiply by 1; to convert C-reactive protein to milligrams per liter, multiply by 10.

Figure 3.. Intensive Care Unit (ICU) Admission, Case Fatality, and Length of Stay in the ICU and Hospital by Latent Class Analysis–Inferred Cluster

Sample sizes indicate numbers of patients from each cluster included after removing patients with missing data. Tail probability is defined as the percentage of patients who were in the ICU (C) or hospitalized (D) beyond a certain length of time. Vertical orange line represents median length of stay (C and D).

Figure 4.. Monthly Percentages of Multisystem Inflammatory Syndrome in Children (MIS-C) Cases by Latent Class Analysis–Inferred Cluster and Number of Total MIS-C Cases, US, May 2020 to October 2022

Variant periods were defined using 50% or more of SARS-CoV-2 variant proportions from national genomic surveillance with an additional 2-week lag due to presumed minimal delay from SARS-CoV-2 infection to MIS-C onset: pre-Delta (July 9, 2021, and earlier), Delta (July 10 to December 31, 2021), and Omicron (January 1, 2022, and later).^, Months in which the total number of cases was less than 30 (April 2020 and earlier as well as November 2022 and later) were suppressed per National Center for Health Statistics standards.