Development and validation of a prognostic model for assessing long COVID risk following Omicron wave-a large population-based cohort study

Abstract

Background: Long coronavirus disease (COVID) after COVID-19 infection is continuously threatening the health of people all over the world. Early prediction of the risk of Long COVID in hospitalized patients will help clinical management of COVID-19, but there is still no reliable and effective prediction model.

Methods: A total of 1905 hospitalized patients with COVID-19 infection were included in this study, and their Long COVID status was followed up 4-8 weeks after discharge. Univariable and multivariable logistic regression analysis were used to determine the risk factors for Long COVID. Patients were randomly divided into a training cohort (70%) and a validation cohort (30%), and factors for constructing the model were screened using Lasso regression in the training cohort. Visualize the Long COVID risk prediction model using nomogram. Evaluate the performance of the model in the training and validation cohort using the area under the curve (AUC), calibration curve, and decision curve analysis (DCA).

Results: A total of 657 patients (34.5%) reported that they had symptoms of long COVID. The most common symptoms were fatigue or muscle weakness (16.8%), followed by sleep difficulties (11.1%) and cough (9.5%). The risk prediction nomogram of age, diabetes, chronic kidney disease, vaccination status, procalcitonin, leukocytes, lymphocytes, interleukin-6 and D-dimer were included for early identification of high-risk patients with Long COVID. AUCs of the model in the training cohort and validation cohort are 0.762 and 0.713, respectively, demonstrating relatively high discrimination of the model. The calibration curve further substantiated the proximity of the nomogram's predicted outcomes to the ideal curve, the consistency between the predicted outcomes and the actual outcomes, and the potential benefits for all patients as indicated by DCA. This observation was further validated in the validation cohort.

Conclusions: We established a nomogram model to predict the long COVID risk of hospitalized patients with COVID-19, and proved its relatively good predictive performance. This model is helpful for the clinical management of long COVID.

Keywords: Long COVID; Nomogram; Omicron; Prediction.

Fig. 1

Flow chart showing the selection of study participants

Fig. 2

Feature variable selection using LASSO regression in the training cohort. (A) Tuning parameter selection cross-validation error curve. (B) Plot of the LASSO coefficient profiles

Fig. 3

Development and application of the model for predicting the risk of long COVID. (A) The final nomogram prediction model. (B-C) Receiving operating characteristic curves showing the performance of the prediction model in predicting long COVID in the (B) training cohort and (C) validation cohort. Abbreviations: CKD, chronic kidney disease; PCT, Procalcitonin; WBC, white blood cell; IL-6, interleukin-6

Fig. 4

(A-B) Calibration curves for testing the stability of prediction model in the (A) training cohort and (B) validation cohort. (C-D) Decision curve analysis of prediction model in the (C) training cohort and (D) validation cohort