Prediction of COVID-19 deterioration in high-risk patients at diagnosis: an early warning score for advanced COVID-19 developed by machine learning

Abstract

Purpose: While more advanced COVID-19 necessitates medical interventions and hospitalization, patients with mild COVID-19 do not require this. Identifying patients at risk of progressing to advanced COVID-19 might guide treatment decisions, particularly for better prioritizing patients in need for hospitalization.

Methods: We developed a machine learning-based predictor for deriving a clinical score identifying patients with asymptomatic/mild COVID-19 at risk of progressing to advanced COVID-19. Clinical data from SARS-CoV-2 positive patients from the multicenter Lean European Open Survey on SARS-CoV-2 Infected Patients (LEOSS) were used for discovery (2020-03-16 to 2020-07-14) and validation (data from 2020-07-15 to 2021-02-16).

Results: The LEOSS dataset contains 473 baseline patient parameters measured at the first patient contact. After training the predictor model on a training dataset comprising 1233 patients, 20 of the 473 parameters were selected for the predictor model. From the predictor model, we delineated a composite predictive score (SACOV-19, Score for the prediction of an Advanced stage of COVID-19) with eleven variables. In the validation cohort (n = 2264 patients), we observed good prediction performance with an area under the curve (AUC) of 0.73 ± 0.01. Besides temperature, age, body mass index and smoking habit, variables indicating pulmonary involvement (respiration rate, oxygen saturation, dyspnea), inflammation (CRP, LDH, lymphocyte counts), and acute kidney injury at diagnosis were identified. For better interpretability, the predictor was translated into a web interface.

Conclusion: We present a machine learning-based predictor model and a clinical score for identifying patients at risk of developing advanced COVID-19.

Keywords: Advanced stage; COVID-19; Complicated stage; LEOSS; Machine learning; Predictive model.

Fig. 1

Machine-learning scheme and results. A Schematic workflow illustrating the iterative reduction of variables. First, the best performing predictor was selected based on all baseline variables (“base predictor”). Next, variables were removed following an iterative optimization procedure leading to the slim predictor and the minimalistic predictor. B Ranking of the variables for the “slim predictor” by their scaled importance. Values in parentheses depict the relative importance. C Performance (area under the curve [AUC] and accuracy) of predictors during the iterative optimization in a top down procedure. From right to left, the procedure started with n = 61 variables removing one variable at a time (displayed on the x-axis). The performance declined considerably at n = 21 variables which led to the selection of the minimalistic predictor just before this decline. D Receiver operating characteristics (ROC) curve of the minimalistic predictor on the test and validation set

Fig. 2

Stability and performance of the minimalistic predictor. A Performance of the minimalistic predictor when leaving out one variable at a time (displayed on the x-axis). Leaving out a single variable from the identified binary variables did not markedly influence the performance. B Receiver operating characteristics (ROC) curve of the minimalistic predictor on the validation set considering only patients without any missing value (in the list of n = 20 variables for the minimalistic predictor). C ROC curve of SACOV-19 based on patients of the validation set without missing values (in the list of the 11 patient variables of SACOV-19)