A machine learning approach identifies 5-ASA and ulcerative colitis as being linked with higher COVID-19 mortality in patients with IBD

Abstract

Inflammatory bowel diseases (IBD), namely Crohn’s disease (CD) and ulcerative colitis (UC) are chronic inflammation within the gastrointestinal tract. IBD patient conditions and treatments, such as with immunosuppressants, may result in a higher risk of viral and bacterial infection and more severe outcomes of infections. The effect of the clinical and demographic factors on the prognosis of COVID-19 among IBD patients is still a significant area of investigation. The lack of available data on a large set of COVID-19 infected IBD patients has hindered progress. To circumvent this lack of large patient data, we present a random sampling approach to generate clinical COVID-19 outcomes (outpatient management, hospitalized and recovered, and hospitalized and deceased) on 20,000 IBD patients modeled on reported summary statistics obtained from the Surveillance Epidemiology of Coronavirus Under Research Exclusion (SECURE-IBD), an international database to monitor and report on outcomes of COVID-19 occurring in IBD patients. We apply machine learning approaches to perform a comprehensive analysis of the primary and secondary covariates to predict COVID-19 outcome in IBD patients. Our analysis reveals that age, medication usage and the number of comorbidities are the primary covariates, while IBD severity, smoking history, gender and IBD subtype (CD or UC) are key secondary features. In particular, elderly male patients with ulcerative colitis, several preexisting conditions, and who smoke comprise a highly vulnerable IBD population. Moreover, treatment with 5-ASAs (sulfasalazine/mesalamine) shows a high association with COVID-19/IBD mortality. Supervised machine learning that considers age, number of comorbidities and medication usage can predict COVID-19/IBD outcomes with approximately 70% accuracy. We explore the challenge of drawing demographic inferences from existing COVID-19/IBD data. Overall, there are fewer IBD case reports from US states with poor health ranking hindering these analyses. Generation of patient characteristics based on known summary statistics allows for increased power to detect IBD factors leading to variable COVID-19 outcomes. There is under-reporting of COVID-19 in IBD patients from US states with poor health ranking, underpinning the perils of using the repository to derive demographic information.

Recent advances in clinical practice: management of inflammatory bowel disease during the COVID-19 pandemic

“…Further studies have confirmed that mortality rate from COVID-19 was similar between patients with IBD and those without (RR 0.95, 95% CI 0.71 to 1.26), however, studies that stratified by disease subtype reported that mortality may be higher in patients with UC compared with Crohn’s disease (RR 1.94, 95% CI 1.22 to 3.10) 24 30. Multivariable analyses from SECURE-IBD suggest that this association is likely to be confounded by age, sex, smoking status and comorbidity 34 49 50…”

Risk Factors for COVID-19 in Inflammatory Bowel Disease: A National, ENEIDA-Based Case–Control Study (COVID-19-EII)

“…The authors stated that the number of reported cases in the earliest assessments was too low to fully evaluate the association of 5-ASA in COVID-19 evolution. However, further analysis using a machine learning approach showed that 5-ASA was highly associated with COVID-19/IBD mortality [25], but this finding was not confirmed by another similar predictive model [26]. Considering this information, researchers of the SECURE-IBD registry were very cautious and concluded that the association of 5-ASA and COVID-19 was likely due to reporting bias and/or reporting delays [18].…”

Update des Addendums zu den S3-Leitlinien Morbus Crohn und Colitis ulcerosa: Betreuung von Patienten mit chronisch-entzündlichen Darmerkrankungen in Bezug auf COVID-19 (Version 2.0)

“…Vereinzelte Studien ließen vermuten, dass die Mortalität bei Patienten mit Colitis ulcerosa höher ist als bei Patienten mit Morbus Crohn sei (RR 1,94, CI 1,22-3,1) [17,30]. Eine multivariate Analyse des SECURE-IBD-Registers legt jedoch nahe, dass die Assoziation vermutlich durch weitere Risikofaktoren bedingt ist [24,31,32].…”