Growing evidence suggests the neurobiological mechanism upholding post-COVID-19 depression mainly relates to immune response and subsequent unresolved low-grade inflammation. Herein we exploit a broad panel of cytokines serum levels measured in COVID-19 survivors at one- and three-month since infection to predict post-COVID-19 depression. 87 COVID survivors were screened for depressive symptomatology at one- and three-month after discharge through the Beck Depression Inventory (BDI-13) and the Zung Self-Rating Depression Scale (ZSDS) at San Raffaele Hospital. Blood samples were collected at both timepoints and analyzed through Luminex. We entered one-month 42 inflammatory compounds into two separate penalized logistic regression models to evaluate their reliability in identifying COVID-19 survivors suffering from clinical depression at the two timepoints, applied within a machine learning routine. Delta values of analytes lowering between timepoints were entered in a third model predicting presence long-term depression. 5000 bootstraps were computed to determine significance of predictors. The cross-sectional model reached a balance accuracy (BA) of 76 % and a sensitivity of 70 %. Post-COVID-19 depression was predicted by high levels of CCL17, CCL22. On the other hand, CXCL10, CCL2, CCL3, CCL8, CXCL5, CCL15, CCL23, CXCL13, and GM-CSF showed protective effects. The longitudinal model obtained good performance as well (BA = 74 % and sensitivity = 68 %), revealing CXCL16 and CCL25 as additional drivers of clinical depression. Moreover, dynamic changes of analytes over time accurately predicted long-term depression (BA = 76 % and sensitivity = 75 %). Our findings unveil a putative immune profile upholding post-COVID-19 depression, thus reinforcing the need to deepen molecular mechanisms to appropriately target depression.
