Risk stratification in hospitalized COVID-19 patients

To the Editor:

Ding et al. recently published a retrospective analysis of potential predictors for mortality in 2,073 Chinese patients hospitalized with COVID-19.¹ As their main findings, they reported that increased liver parameters as well as liver injury were predictive for 28-day mortality, and proposed a nomogram that estimates the mortality risk of hospitalized patients with COVID-19 disease. However, as the understanding of COVID-19 disease improves, it becomes evident that the disease may present differently in different regions. Liver injury, as well as other gastrointestinal symptoms were associated with worse outcome in some, but not all studies.[2], [3], [4], [5], [6], [7] Therefore, we set out to validate the main findings in an Austrian cohort of 405 hospitalized patients with COVID-19 disease.

We retrospectively collected demographic and laboratory data as well as in-hospital mortality from all patients with a positive SARS-Cov-2 PCR test hospitalized at either the University Hospital Graz or the State Hospital Graz II between February 28^th, 2020 and May 30^th, 2020. Due to availability, the nomogram parameters direct bilirubin and troponin I were substituted with total bilirubin and troponin T, respectively. Cox-Regression was used to estimate hazard ratios for 7-, 14-, 21- and 28-day in-hospital mortality; p values below 0.05 were considered significant. Monte-Carlo simulation were run to define cut-offs with the highest overall accuracy.

Firstly, we analysed whether elevated liver enzymes (aspartate aminotransferase [AST], alanine aminotransferase [ALT]) or liver injury are predictive of 28-day mortality, as proposed in the original article. AST levels above the upper limit of normal were found in 189 of 305 (61,4%) patients, elevated ALT levels in 85 of 310 (27.4%) patients; 21 of 243 (8.6%) patients fulfilled the criteria for liver injury. We found that neither elevated AST (hazard ratio [HR] 1.23; 95% CI 0.77–1.96; p = 0.4), elevated ALT (HR 0.61; 95% CI 0.35–1.07; p = 0.09) nor liver injury (HR 0.74; 95% CI 0.27–2.05; p = 0.6) could predict hospital mortality in COVID-19 patients in Graz, Austria. Similar, non-predictive results were obtained for 7-, 14-, or 21-day mortality. GI symptoms in general, as well as diarrhoea in particular, were not associated with 28-day mortality (HR 0.58; 95% CI 0.15–2.16; p = 0.4 and HR 0.42; 95% CI 0.04–4.00; p = 0.5, respectively).

Secondly, we tested whether the proposed nomogram, including age, severe pneumonia, lymphocyte count, platelet count, C-reactive protein, D-dimer, creatinine, troponin, AST and bilirubin, could predict 28-day mortality in the Austrian cohort. A complete data set could be obtained from 151 patients, 37 (24.5%) of whom died within 28 days of hospitalization. The score derived from the nomogram was significantly associated with 28-day in-hospital mortality (HR 1.009; 95% CI 1.005–1.013; p <0.001) and showed a high predictive accuracy (AUROC 0.80; 95% CI 0.71–0.88; p <0.001). Similar results were obtained for 7-, 14-, or 21-day mortality. However, not all factors of the nomogram could predict in-hospital mortality individually. While age points (i.e. age∗1.11-1.11), severe pneumonia, C-reactive protein over 10 mg/L, D-dimer over 1.3 mg/L (adjusted cut-off), creatinine over 84 μmol/L and troponin T over 15.6 ng/ml were predictive, lymphocyte count, platelet count, AST and bilirubin were not.

Although the proposed nomogram is performing well, it is rather complicated to calculate which could be a considerable disadvantage in well-frequented COVID-19 wards. We propose a simplified version based on the predictors that we could reliably reproduce in the Austrian cohort. The simplified score is composed of the sum of the following: age decade, 4 points for severe pneumonia, 8 points for C-reactive protein over 10 mg/L, 4 points for creatinine over 84 μmol/L. This simplified version can still predict 28-day mortality (HR 1.24; 95% CI 1.16–1.34; p <0.001) and shows comparable accuracy to the full version of the nomogram (AUC difference: -0.001; 95% CI -0.05 to 0.05; p = 0.9). A score of 22 or higher best identified patients at risk of 28-day in-hospital mortality (Fig. 1 ).

Fig. 1.

Prediction of in-hospital mortality.

(A) AUROC for the proposed nomogram and the simplified version. (B) Kaplan-Meier curves according to the simplified nomogram (p <0.0001, Log-rank test).

In conclusion, we validated and simplified the nomogram proposed by Ding et al. ¹ although liver injury itself was not predictive in our cohort. The divergence in liver function-related findings is not unexpected. Several meta-analyses in the past 6 months showed that a certain fraction of individual studies indicated no association between liver function abnormalities and outcome, although, overall, they were associated with disease severity and outcome.^{3 , 5 , 6} Our retrospective study design is not suited to address the source of this discrepancy. However, direct comparison of the 2 study populations showed that patients in our cohort were 74 (±14) years old and therefore 12 years older on average compared to Ding et al., which might explain the considerably higher mortality rate as well as the higher prevalence of liver function abnormalities (74% vs. 61%, respectively) in our cohort. Our cohort is limited by the smaller sample size compared to Ding et al. and the unavailability of direct bilirubin which might also contribute to the observed discrepancies. Despite these limitations we could validate the nomogram proposed by Ding et al. and call for validation of the full¹ as well as our simplified version of the nomogram in other cohorts across the globe to improve the risk stratification of COVID-19 patients.

Financial support

No financial support was given in order to complete the study or write the manuscript.

Authors’ contributions

TL, NF, HW collected data, AH, VS analysed the data and wrote the manuscript.

Conflict of interest

The authors declare no conflict of interest.

Please refer to the accompanying ICMJE disclosure forms for further details.

Supplementary data

The following is the supplementary data to this article: