DISCUSSION

In this study, we analyzed significant clinical risk factors and developed a model based on predicting rehabilitation duration for the refined risk management of mild-moderate COVID-19. To our knowledge, this study is the first to predict the rehabilitation duration for mild-moderate COVID-19 based on large-scale datasets of clinical factors. The rehabilitation duration can be easily calculated using the risk model for 1 mild-moderate case of COVID-19 (F = 11.055, P < 0.001, and adjusted R² = 0.361). According to the risk stratification of the local epidemic, tailored clinical decisions-making can be conducted, and strategic risk management can be derived. The model incorporated 5 clinical indices of WBC, PaCO², K, TBIL, and AST, from blood gas analysis, blood cell analyses, hepatic function, and electrolytes, respectively. These indices reflect information on various aspects of mild-moderate COVID-19 during the recovery course. Therefore, these 5 predictors are independent factors that are related to the rehabilitation risk of mild-moderate COVID-19, and can be used as simple and effective prognostic evaluation indicators.

Among the 5 predictors incorporated into the model, WBC counts are often considered as a marker of systemic inflammation and immune system health. ^{Reference Hall, Stanaway and Carrell17} The acute variation of WBC counts is related to infections or other environmental exposures. ^{Reference Goswami and Ghosh18} A study found ^{Reference Deng, Liu and Liu19} that the WBC count increased in the group of COVID-19 patients who died during hospitalization; comorbid bacterial or fungal infection might have occurred in these patients, causing the WBC count to increase. Our study also validated that WBC count can be used as a predictor of rehabilitation duration. Interestingly, CD4+ and CD8+ T cells play protective roles in coronavirus infection ^{Reference Ling, Xu and Lin13,Reference Chen, Qi and Liu20,Reference Sun, Li and Lu21} but were not incorporated into our model. Previous studies found that, during the acute phase of SARS-CoV infection in humans, CD4+ and CD8+ T cell counts were decreased. ^{Reference Wong, Wu and To22-Reference Li, Qiu and Zhang24} SARS-CoV-2 also presented similar characteristics, because viral infection causes persistent consumption and/or insufficient regeneration of lymphocytes. ^{Reference Deng, Liu and Liu19} Therefore, CD4+ and CD8+ T cell counts vary in response to SARS-CoV-2 infection in the acute phase and may be far from satisfactory for protection during the entire recovery period.

Compared with PaO2/FiO2, which is widely accepted for quantifying oxygenation in the evaluation and monitoring of severe cases of COVID-19, ^{Reference Shen, Wang and Zhao25} studies regarding PaCO2 on COVID-19 are scarce. PaCO2 reflects the balance between the production and elimination of carbon dioxide (CO2) and is related to alveolar ventilation. Increased PaCO2 levels indicate the inefficiency of the lungs to eliminate CO2 and increase in pulmonary dead space. ^{Reference Nuckton, Alonso and Kallet26} Therefore, several scholars have proposed to use intermediate tidal volume instead of low tidal volume to correct the inadequacy of ventilation in COVID-19 cases. ^{Reference Liu, Liu and Xu27} On the other hand, a study from Xixi Hospital in Hangzhou reported ^{Reference Liu, Xu and Zhang28} that patients with COVID-19 had low K level, especially 1 case of extended hospitalization, but these levels increased persistently after treatment. In our study, PaCO2 and K were significantly related to the rehabilitation duration, and the PaCO2 level of all 90 cases and K levels in most cases were normal. Related research regarding 2 predictors and COVID-19 is rare. Our study suggests that the value of PaCO2 and K on COVID-19 should be further explored, as they may become important predictors of clinical outcome, even if at the normal level, and may be suitable predictors for all COVID-19 cases rather than only severe cases.

At least 7 relatively large-scale case studies have reported ^{Reference Huang, Wang and Li29-Reference Guan, Ni and Hu35} that 2-11% of COVID-19 cases had liver comorbidities, and 14-53% of cases have reported abnormal liver function during disease progression. Severe COVID-19 cases have higher rates of liver dysfunction. ^{Reference Zhang, Shi and Wang36} Liu et al. determined ^{Reference Liu, Jiang and Shao37} that cytokine storm syndrome was likely to emerge in severe cases and could result in abnormal serum transaminase and bilirubin levels. AST and TBIL were incorporated into our model, which may reflect the levels of cytokine release in mild-moderate COVID-19 cases. Moreover, another study ^{Reference Yao, Wang and Lian38} found that there was no correlation between increases in transaminase and TBIL (R = −0.006; P = 0.972). In our study, it is not difficult to understand that TBIL was positively correlated with rehabilitation duration; but of interest, all transaminases (including AST) were negatively correlated with rehabilitation duration (shown in Supplemental Table S1). Transaminases mainly exist in hepatocytes and are released when hepatocellular damage occurs, but the excessively low level of transaminases is not a favorable signal in disease progression, such as in liver failure. ^{Reference He, Xia and Yu39} Because transaminases are indispensable catalysts for the normal operation of the liver, the moderate transaminase levels in mild-moderate cases reflects the activity of hepatocytes (not limited to COVID-19).

In this study, with the help of statistical analysis and comprehensive analysis of 5 conventional laboratory indicators, a mathematical model was developed to effectively predict the rehabilitation duration of mild-moderate COVID-19 cases without complicated and expensive detection techniques. As these tests can be performed in grass-root medical hospitals, and the results can be derived quickly, the risk model has wide clinical applicability. If the local rehabilitation duration is 14 ± 5 d (means ± SD), when the predicted rehabilitation duration of 3 mild-moderate cases with COVID-19 is 7, 10, and 20 d, the 3 cases have low, intermediate, and high risks, respectively. Accordingly, the strategy of individualized treatments should focus on containing the further spread in low-risk cases, and surveilling the complete recovery process and preventing the disease fluctuation in intermediate risk cases, while it is necessary to analyze the reasons for each high-risk case and perform aggressive interventions. If the predictive rehabilitation duration of 1 mild-moderate COVID-19 case varies during the recovery process, tailored decisions-making can be adjusted dynamically (Figure 2). Based on this refined risk management system, frontline medical staff can increase their levels of expertise regarding mild-moderate COVID-19, which fits the current trend toward personalized medicine. More importantly, this may help to assess the cost of local epidemic control in the context of big data.

There are several limitations in this study. First, limited by the outbreak scale of our local epidemic, the sample size in this study was not large enough. Second, the R² and adjusted R² of the model were not significant, and we cannot avoid potential biases as a retrospective study. Third, because all cases in this study were from our local region, it is unclear whether the characteristics were consistent with those of other regions. Finally, as a retrospective study, we did not obtain the re-examined test of 5 risk predictors at the same timepoint to validate the model from other different timepoints to discharge. This is the first study of its kind, there is no similar model for reference, and the proposed model may be further optimized after incorporating more valuable variables and larger cases. Therefore, in the next step, we will strive to conduct a multi-center and multi-area research for enrolling a larger number of cases to validate and improve our model for the rehabilitation duration prediction from any timepoint to discharge and early warning if the disease varies.