Differences of SARS-CoV-2 serological test performance between hospitalized and outpatient COVID-19 cases

•Deep learning technology applied to diagnosis of Acute Promyelocytic Leukemia (APL).•Convolutional neural networks (Mask R-CNN) may diagnose APL in bone marrow smear images with a high precision.•Data augmentation and pre-trained approach improves diagnostic accuracy. Serological severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) antibody assays differ in the target antigen specificity, e.g. of antibodies directed against the viral spike or the nucleocapsid protein, and in the spectrum of detected immunoglobulins. The aim of the study was to evaluate the performance of two different routinely used immunoassays in hospitalized and outpatient COVID-19 cases. The test characteristics of commercially available spike1 protein-based serological assays (Euroimmun, EI-assays), determining IgA or IgG and nucleocapsid-based assays (Virotech, VT-assays) determining IgA, IgM or IgG were compared in 139 controls and 116 hospitalized and outpatient COVID-19 cases. Hospitalized COVID-19 patients (n = 51; 115 samples) showed significantly higher concentrations of antibodies against SARS-CoV-2 and differed from outpatient cases (n = 65) by higher age, higher disease severity scores and earlier follow up blood sampling. Sensitivity of the two IgG assays was comparable in hospitalized patients tested ≥ 14 days (EI-assay: 88%, CI95% 67.6–99.9; VT-assay: 96%, CI95% 77.7–99.8). In outpatient COVID-19 cases sensitivity was significantly lower in the VT-assay (86.2%, CI95% 74.8–93.1) compared with the EI-assay (98.5%, CI95% 90.6–99.9). Assays for IgA and IgM demonstrated a lack of specificity or sensitivity. Our results indicate that SARS-CoV-2 serological assays may need to be optimized to produce reliable results in outpatient COVID-19 cases who are low or even asymptomatic. Assays for IgA and IgM have limited diagnostic performance and do not prove an additional value for population-based screening approaches.