By Vijay Kumar Malesu May 3 2024 Reviewed by Lily Ramsey, LLM

In a recent study published in The Lancet Digital Health, a group of researchers analyzed severe coronavirus disease 2019 (COVID-19) outcomes in patients with immune-mediated inflammatory diseases (IMIDs), focusing on the effects of medications, comorbidities, and vaccination status during different pandemic phases.

Study:  Machine learning to understand risks for severe COVID-19 outcomes: a retrospective cohort study of immune-mediated inflammatory diseases, immunomodulatory medications, and comorbidities in a large US health-care system . Image Credit: fizkes/ Background 

As of February 28, 2024, the global COVID-19 pandemic has resulted in over 7 million deaths, posing significant risks, particularly to those with IMIDs, such as rheumatoid arthritis, multiple sclerosis, and allergic asthma.

These conditions are marked by chronic inflammation and immune dysregulation, potentially exacerbating COVID-19 severity. Contributing factors include immune dysfunction, use of immunomodulatory medications (IMMs), and prevalent comorbidities like heart disease and diabetes.

Interestingly, allergic asthma may lessen severe COVID-19 risks, indicating complex interactions between IMIDs and the virus. Further research is needed to clarify these relationships to enhance patient-specific guidelines and care. About the study 

In the present study, clinical data were sourced from electronic health records (EHRs) of Providence St Joseph Health (PSJH), which operates 51 hospitals and 1,085 clinics across Alaska, California, Montana, Oregon, New Mexico, Texas, and Washington.

The study was divided into two periods: the pre-omicron phase (March 1, 2020- December 25, 2021) featuring the wild-type and early variants like alpha and delta, and the omicron-predominant phase (December 26, 2021-August 30, 2022).

Patients with IMIDs and controls without IMIDs were identified from medical records, ensuring data on diseases, medications, and comorbidities were recorded before their first COVID-19 infection. Related StoriesAging affects immune response and virus dynamics in COVID-19 patients, study findsStudy suggests lingering coronavirus in tissues may contribute to long COVID symptomsStudy reveals how SARS-CoV-2 hijacks lung cells to drive COVID-19 severity

The study, adhering to Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines and approved by an Institutional Review Board, tracked outcomes like hospitalization based on the date of a valid COVID-19 test.

Statistical analysis utilized machine learning models to parse the data, focusing on patient demographics, vaccination status, active comorbidities, IMID diagnoses, and IMM use. Variables were normalized, and missing data was handled through median imputation.

The study employed logistic regression (LR) for its interpretability and extreme gradient boosting (XGB) due to its capability to handle non-linear data efficiently.

Performance was evaluated on a held-out test set with the area under the receiver operating characteristic curve as a metric. Additionally, the study examined variable importance and interactions using various statistical techniques to ensure robustness and reliability in findings. Study results 

In the large-scale analysis of 2,167,656 patients tested for Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), 290,855 (13.4%) were confirmed to have COVID-19, including 15,397 (5.3%) patients with IMIDs and 275,458 (94.7%) without.

During the pre-omicron period, 110,217 (64.8%) individuals testing positive for COVID-19 were not fully vaccinated, a trend that decreased slightly in the omicron-predominant period with 64,864 (53.7%) unvaccinated.

Notably, in the omicron-predominant period, both the overall patient cohort and those testing positive had higher comorbidity rates and increased vaccination coverage compared to earlier in the pandemic.

In the pre-omicron period, 169,993 (11.2%) of 1,517,295 tested individuals were COVID-19 positive. Among them, 23,330 (13.7%) were hospitalized, 1,072 (0.6%) required mechanical ventilation, and 5,294 (3.1%) died.

IMID patients had higher rates of hospitalization (1,176 [14.6%] vs. 22,154 [13.7%]; p=0.024) and mortality (314 [3.9%] vs. 4,980 [3.1%]; p<0.0001) compared to controls. During the omicron-predominant period, the positive test rate increased to 18.6%, but hospitalizations (14,504 [12.0%]), mechanical ventilation (567 [0.5%]), and deaths (2,001 [1.7%]) declined.

IMID patients continued to show higher hospitalization (1,082 [14.8%] vs. 13,422 [11.8%]; p<0.0001) and mortality rates (187 [2.6%] vs. 1,814 [1.6%]; p<0.0001) than controls.

Age and certain comorbidities like atrial fibrillation, coronary artery disease, heart failure, chronic kidney disease, chronic obstructive pulmonary disease (COPD), chronic liver disease, and cancer consistently emerged as risk factors for severe COVID-19 outcomes across both time periods.

Conversely, vaccination and booster statuses were associated with significantly better outcomes. Interestingly, asthma and psoriasis were linked to reduced risks of severe consequences, highlighting the complexity of interactions between IMIDs and COVID-19.

Analysis via LR and XGB revealed insights into these associations. The XGB model, in particular, demonstrated superior performance in classifying health outcomes, with an area under the receiver operating characteristic curve ranging from 0.77 to 0.92, outperforming LR by approximately 7.5%.

Further detailed analysis confirmed the importance of variables such as opioid dependence and specific IMIDs like rheumatoid arthritis and multiple sclerosis in predicting severe COVID-19 outcomes. Journal reference:

Qi Wei, Philip J Mease, Michael Chiorean, et al. (2024) Machine learning to understand risks for severe COVID-19 outcomes: a retrospective cohort study of immune-mediated inflammatory diseases, immunomodulatory medications, and comorbidities in a large US health-care system, The Lancet Digital Health. doi: