Machine-learning-based analysis of liver injuries in COVID-19 patients

Background For several months, global numbers of SARS-CoV-2 infections are rising. The COVID-19 pandemic is a huge challenge for the healthcare system worldwide as well as for the society. Clinicians have to make fast but informed decision. We analyzed a large amount of collected data from the intensive care unit with a machine-learning-assisted algorithm to find essential clinical correlations between parameters and a severe COVID-19 disease.

Methods We used machine-learning algorithms – a gradient boosting model and a neural network autoencoder - to analyze 56 COVID-19 patients from the intensive care unit (ICU) of the University Hospital of Regensburg. The dataset included 136 measurements of high-frequency vital- and respiratory parameters, laboratory measurements and demographical information. Our focus was set on the liver values glutamic-pyruvate transaminase (GPT), Bilirubin, alkaline phosphatase (ALP) and international normalized ratio (INR). We compared these values of 56 COVID-19 patients with the same of 589 non-COVID-19 Patients hospitalized at the ICU at 2019. On this basis, we defined a severity score to classify the liver injury of SARS-CoV-2 infected patients.

The grading of the severity score is defined by the deviation of the measured values to its normal values:

• Severity grade 1: deviation by 1,5-2x of normal value of GPT, Bilirubin and ALP and an INR of 1,15- 1,35

• Severity grade 2: deviation by 2-3x of normal value and 0,5 step of INR

• Severity grade 3: deviation by 3-4x of normal value and 0,5 step of INR

• Severity grade 4: deviation of 4 or more of normal value and 0,5 step of INR

Results 56 SARS-CoV-2 infected patients from the intensive care station were analyzed. The whole patient cohort shows indications of a liver involvement. 8 % show severity grade 1, 19 % severity grade 2 and 10 % severity grade 3. Noteworthy is that 60 % of the COVID-19 patients got the highest severity grade of 4. Furthermore 38 % of the patients with severity grade 4 died.

Conclusions In our study, we show that there is a correlation of a COVID-19 disease and liver involvement. All of the analyzed SARS-CoV-2-infected patients developed massive liver injuries. 60 % of the patients developed the highest severity grade 4 and 38 % of them died. In conclusion, elevated liver values overlap with increased mortality. The challenging task for the future is to find the molecular mechanism of the connection between SARS-CoV-2 and the liver.

Publication History

Article published online:
04 January 2021