Research Specialist Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital Chelmsford, United States
Background: A dynamic model to predict rapid disease progression in patients hospitalized with COVID-19 infection was recently published using data prior to November 2020. However, the performance of the model following the availability of vaccines and anti-viral treatments against COVID-19 is unclear.
Objectives: We aimed to validate and to update a dynamic prognostic model that incorporates changes in biomarkers to predict rapid clinical deterioration in patients hospitalized for COVID-19 following COVID-19 vaccine and treatment availability.
Methods: Using electronic health records from a large integrated care delivery network in Massachusetts, USA, from March 2020 to November 2021, we established a cohort of hospitalized patients aged ≥ 18 years with laboratory-confirmed COVID-19. We calculated C-statistic and Brier score for the published dynamic model using data after November 2020. To update the model to incorporate more recent data, we divided the entire dataset into development (70%) and validation (30%) cohorts. We used LASSO regression with 10-fold cross-validation to select predictors of rapid severity progression (defined as progression by two levels of a published severity scale in the next 24 hours) from a total of 115 predictors (including demographic features, pre-existing comorbidities and medications and time-varying vital and laboratory parameters). We calculated C-statistic and Brier score in development and validation datasets.
Results: Our study cohort consisted of 8,185 patients (5,730 in the development cohort [mean age: 62; 44% female] and 2,455 in the validation cohort [mean age: 62; 45% female]). The previously published model had a suboptimal performance using data after Nov 2020 (N=4,973, C-statistic=0.6. Brier score=1.07). We thus retrained our model using development set from Mar 2020 to Nov 2021. The updated model included 38 predictors with 18 as changing values in specific biomarkers. Patients hospitalized after Jun 1st, 2021 (when Covid-19 vaccines became widely available in Massachusetts) were on average younger, predominantly white, and had fewer comorbidities than those hospitalized before. The C-statistic was 0.79 in the development cohort and 0.76 in the validation cohort. Brier score was 0.016 and 0.017 in the development and validation cohorts respectively.
Conclusion: We developed and validated a new dynamic model predicting rapid severity progression among patients hospitalized with COVID-19. Characteristics of hospitalized patients and the predictors of the outcome have evolved substantially over time.
