COVID‐19 Patients Benefitting From Remdesivir for Improved Survival: A Neural Network‐Based Approach

ABSTRACT

Conflicting results from randomized trials regarding the efficacy of remdesivir for COVID‐19 have been reported. We aimed to develop a neural network (NN) to identify COVID‐19 patients who would derive the greatest survival benefit from remdesivir. This multicenter observational study included adults hospitalized for COVID‐19 between February 2020 and February 2021. A derivation cohort from Hospital Clínic (Barcelona) was used to create the NN, which was split into 928/1160 (80%) for training and 232/1160 (20%) for internal validation. The model used three normalized input variables: Ct values from rRT‐PCR, lymphocyte count at diagnosis, and the duration of symptoms before testing. Effectiveness was assessed in an external validation cohort of 898 patients from Hospital Mútua Terrassa and Hospital Universitari La Fe, Valencia. In the derivation cohort (median age 66 years; IQR 55–78), symptom duration, Ct values, and lymphocyte count showed considerable variation. Overall, 60‐day mortality was 165/1160 (14.2%). In the training set, 385/928 (41.5%) patients were identified as benefiting from remdesivir, characterized by lower Ct values, reduced lymphocyte counts, and shorter symptom duration. Mortality in this subgroup was 93/385 (24.2%): 6/385 (7.2%) in patients receiving remdesivir versus 87/385 (28.8%) in those who did not (p < 0.001). In the test set, 296/898 (33%) patients were identified as high‐benefit, with 60‐day mortality rates of 8/296 (11%) for those patients treated with remdesivir compared to 49/296 (22%) for those not treated (p < 0.04). In conclusion, we successfully developed and validated a NN capable of identifying patients with distinct clinical phenotypes who are at higher risk of mortality without remdesivir.