Data mining approaches for detecting signals associated with post-acute pediatric sequelae of COVID-19

In a recent study published on medRxiv* preprint server, researchers described tree analysis statistics for the long-running pediatric coronavirus disease 2019 (COVID-19).

Background

The National Institutes of Health (NIH) launched the Researching COVID to Enhance Recovery (RECOVER) initiative in 2021 to use electronic health record (EHR) data to identify and classify patients with post-acute sequelae of COVID-19 (PASC), as described by the NIH as an inability to recover from SARS-CoV-2 infection or persistent symptomatology for more than 30 days.

According to the literature, PASC has been predicted in patients with COVID-19 and its origin, risk factors and outcomes have been described. So far, only a few studies have accurately described PASC in children.

About the study

In the current study, researchers aimed to uncover PASC signals using data mining instead of clinical experience.

Two comparisons dominated the analyzes presented. PASC cases were compared to patients infected and uninfected with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). The PASC evidence included a diagnosis code U09.9, an EHR Interface Terminology (IMO) term containing either the strings (“post” and “acute” and “covid”), or a diagnosis code B94.8.

Infected patients had a positive polymerase chain reaction (PCR), antigen, or serological test for SARS-COV-2. Serological tests for infection included immunoglobulin (Ig)-M, anti-nucleocapsid (N) IgG antibodies, anti-spike (S) or receptor binding domain (RBD) IgG antibodies, and undifferentiated antibodies IgA and IgG. Patients with COVID-19 in the hospital or emergency department (ED) were also classified as infected with SARS-CoV-2.

During the observation period of the study, testing for the virus was still routinely performed in healthcare facilities. Patients were considered uninfected with SARS-CoV-2 if (1) all diagnostic tests such as an antigen, PCR, and serology were negative during the study period, and (2) the patient had no diagnostic code indicating COVID-19, multisystem inflammatory syndrome in children (MIS-C) or PASC.

Cohort entry date for incident PASC infections was the same as the initial positive antigen or PCR test, four weeks before the initial positive serological test, or four weeks prior to initial PASC diagnosis in the absence of confirmatory testing.

Date of entry for non-PASC COVID-19 positive patients was based on first diagnosis or first encounter with COVID-19. For patients not infected with SARS-CoV-2, random negative tests were noted as cohort entry dates. At the start of the cohort, all the case and control patients were aged over 21 years. For each diagnostic code and each patient, the team constructed a binary indicator for the occurrence of an incident within 28 to 179 days of cohort admission.

The team used vocabulary from the 10th revision of the International Classification of Diseases (ICD-10) as input data. The hierarchy followed had seven levels of nodes corresponding to each level of the tree structure. The hierarchy was alternatively called a tree while the cluster containing an a node with its descendants was called the branch of a tree or a cut.

Results

A total of 13,750 patients were recruited for the three cohorts between March 1, 2020 and June 22, 2022, with 1,250 cases of PASC infection. Younger boys and girls were less likely to be part of the PASC cohort. Most cohorts entered in fall 2021.

Multiple statistical indications emerged when comparing people infected with PASC and COVID-19. At the highest level of the tree analysis, appreciable reductions were noted with ICD-10 codes for signs, symptoms, and clinical and laboratory findings that were not elsewhere classified, musculoskeletal diseases and connective tissue, nervous system disorders, respiratory disorders, mental illnesses. and behavioral disorders, nutritional, endocrine and metabolic disorders, diseases of the circulatory system, variables affecting health status and health services, subcutaneous and cutaneous diseases; and diseases of the digestive system.

Within the branch describing uncategorized signs and symptoms, the three main breaks corresponded to respiratory and circulatory symptoms, general symptoms, and cognitive, perceptual, emotional and behavioral symptoms.

Conclusion

The results of the study showed several disorders and bodily systems related to PASC. Because the study used data-driven methods, the team identified several new or underreported illnesses and symptoms.

The researchers believe that a more data-driven approach to knowledge discovery is needed due to the rapidly evolving nature of the pandemic and the lack of agreement on the precise symptoms that characterize PASC in children. This comprehensive analysis of diagnoses in a cohort of children with PASC adds much to the knowledge of the medical community about the complex symptoms of this disorder.

The study results may guide the design of future prospective studies to further explore the patterns found here, improve therapeutic practice, and focus research on the biochemical underpinnings of PASC.

*Important Notice

medRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be considered conclusive, guide clinical practice/health-related behaviors, or treated as established information.