A total of 432 patients with COVID-19 were included in this retrospective study. In the univariate analysis, we found that the high level of NLR, LDH, D-dimer, and CT score have significant correlation with the severity of COVID-19. While after adjusting other statistically significant indexes, the predictive value of NLR > 3.82, LDH > 246U/L were persisted. This indicated that when NLR exceeded the cutoff point, the risk of serious disease increased by 2.163 times. And the risk of LDH over optimum cutoff increased by 2.298 times. While the value of D-dimer > 0.83 µg/ml and CT score > 7 in predicting the severity of disease was weak and could not be recommended as independent predictors. In addition, the risk of severity was also closely related to fatigue, chest tightness, hypertension and CRP. Meanwhile, combining NLR > 3.82 and LDH > 246U/L can improve the sensitivity of disease risk prediction.
Immune dysfunction plays an important role in the severity of COVID-19 [12]. Recent studies have elucidated that neutropenia and lower lymphopenia could be found in the severe group of COVID-19[13, 14]. NLR took lymphocyte and neutrophil into account at the same time. Several studies have shown the predictive value of NLR in distinguishing COVID-19 patients with severe and critical types. In a study of the dynamic changes of lymphocyte subsets and cytokine profiles in patients with COVID-19, NLR can be used as a prognostic factor for early identification of severe cases[15]. A cohort of patients with COVID-19 also proved that, after adjustment of confounding factors, each unit increase in NLR, the risk of in-hospital mortality increases by 8%[16]. Another study conducted by Yang X et. al, [6] in 93 patients with COVID-19 demonstrated that NLR can be used as independent indicators for poor clinical outcome and the largest AUC for NLR were 0.841,with specificity (63.6%) and sensitivity (88%). However, limited by sample diversity, the outcome needs further evaluation. In the present study, the predictive value of NLR is consistent with abovementioned studies. Meanwhile, the sample size and diversity were enriched by collecting data from two clinical centers, which will strengthen the reliability of conclusions. The optimum cutoff for NLR was 3.82 and AUC was 0.716. And the sensitivity and specificity of NLR > 3.82 were 50.40% and 84.04%, respectively. Meanwhile, in multivariate logistic regression, NLR > 3.82 can be used as an independent predictor for disease risk (OR = 2.163;95%CI = 1.162–4.026; p = 0.015).
The elevation of LDH was one of the most common laboratory abnormalities in patients with COVID-19. Acute lung injury was highly related to LDH[17]. A systematic literature review and meta-analysis had shown that LDH > 245U/L can predict the progress of COVID-19[7]. In a study of the risk factors for death in cancer patients with COVID-19, the elevated LDH was closely related to the increase of mortality [18]. Furthermore, in another retrospective analysis of 120 patients with COVID-19, comparing with mild patients, the severe patients have higher LDH levels (mean 200.8 U/L for mild vs mean 342.8 U/L for severe)[19]. The predictive value of LDH is further confirmed by our study. Meanwhile, ROC analysis showed that the AUC for LDH was 0.74 and the optimum cutoff was 246 U/L. The sensitivity was 59.2% and the specificity was 79.15%. And the logistic regression indicated that the risk of serious disease increased by 2.298 times when LDH over optimum cutoff (OR = 2.298; 95%CI = 1.327–3.979; p = 0.003). In addition, the sensitivity of disease risk prediction can be improved by combining LDH > 246U/L with NLR > 3.82. (NLR > 3.82 [50.40%] vs. combined diagnosis model [72.80%]; p = 0.0007; LDH > 246 [59.2%] vs. combined diagnosis model [72.80%]; p < 0.0001). However, the specificity were decreased (NLR > 3.82 [84.04%] vs. combined diagnosis model [69.71%]; p = 0.0007; LDH > 246[79.15%] vs. combined diagnosis model [69.71%]; p < 0.0001).
Moreover, the sensitivity, specificity, and AUC for NLR and LDH are not relatively high enough. Due to the different admission time of patients with COVID-19 and the acute aggravation of some patients in a period of time after admission, the value of admission indicators may be underestimated. However, compared with other articles[6, 15, 20], the sample size and diversity of patients with COVID-19 increase the reliability of the results in this study. Meanwhile, more importantly, the optimum cutoff can indicate the risk of acute aggravation of patients with COVID-19 in the present study. Furthermore, it provides more evidence for the establishment of multi-parameter diagnosis model.
Coagulation disorders are more common in severe patients than in light patients [21, 22]. A study conducted by Zhang L et al.[23] had proved that the level of D-dimer ≥ 2.0 µg/mL (fourfold increase) could effectively predict the mortality of patients with COVID-19. While after balancing the confounding factors, the logistic regression showed that D-dimer > 0.83 µg/ml could not be used as an independent predictor of disease risk in this study (OR = 1.209; 95%CI = 0.626–2.334; p = 0.571). In a dynamic study of hematological parameters in patients with COVID-19, the D-dimer of the severe group was higher than the non-severe group on days 1, 7 and 14 (p < 0.05) [24]. This suggests that due to different admission times, the ability of D-dimer to predict disease risk may be weakened. In another response to the prognostic value of D-dimer in patients with COVID-19, the predictive value of D-dimer might be affected by other factors, such as hormonotherapy, antibiotic therapy et al. Due to the baseline level of D-dimer varies greatly in patients, the value of dynamic monitoring of D-dimer may be higher in patients with COVID-19[25]. Further researches are still required to evaluate the significance of D-dimer in evaluating the severity of COVID-19.
COVID-19 patients have lung involvement with imaging changes [10, 26]. In different stages of the disease, the CT manifestations are different, which are important to the diagnosis and staging of patients [27]. With the same semi-quantitative scoring system, a multi-center paired cohort study conducted by J. Liu et al. [28] showed that CT changes are obvious in acute exacerbation of COVID-19, accompanied by an increase of CT score. This indicated that elevated CT score may predict the poor outcome. Another retrospective single-center study indicated that the CT score has a high diagnostic value in patients with severe COVID-19. ROC analysis showed that AUC for CT score was 0.918. The optimum cutoff of CT score was 7.5. The sensitivity was 82.6% and the specificity was 100% [11]. However, the study only analyzes imaging, without combined with clinical data. And significant differences in the number of patients between severe-critical patients and non-severe groups also affect the accuracy of the results. While in present study, after combining the clinical data, the CT score can’t be used as an independent predictor of disease risk (OR = 1.519; 95%CI = 0.71–3.247; p = 0.281). A study by Zhang B et. al[29], demonstrated that the severity of lung abnormalities evaluated by CT score might be associated with laboratory parameters. Therefore, due to correlation between CT score and laboratory parameters, the ability to independently predict disease risk of CT score may be attenuated. Additional investigations are warranted to assess whether CT score can be an independent predictor of disease risk.
There are some limitations in this study. First, owing to different severity of patients and different medical resources, the time from onset to admission might not be representative, which might affect the level of four parameters on admission. Meanwhile, the representativeness of CT score and D-dimer may also be affected by different admission times. Second, other clinical data and test results are not included in the analysis, which may cause bias, weakening the reliability of the results. Third, to a certain degree, the CT score as a semi-quantitative evaluation method was subjective.