Risk Stratification Scores for Hospitalization Duration and Disease Progression in Moderate and Severe Patients With COVID-19

Jiaqi Huang; Yu Xu; Bin Wang; Ying Xiang; Na Wu; Wenjing Zhang; Tingting Xia; Zhiquan Yuan; Chengying Li; Xiaoyue Jia; Yifan Shan; Menglei Chen; Qi Li; Li Bai; Yafei Li


BMC Pulm Med. 2021;21(120) 

In This Article


Risk Scoring System to Stratify the Moderate Patients With Different Length of Hospital Stay

The demographic and clinical characteristics of moderate patients were summarized in Table 1. The cut-off value of length of hospital stay was defined as 13 days. There were 789 long-stay (> 13 days) patients (50.1% males, median age 61 years) and 892 short-stay patients (49.4% males, median age 56 years). During the observation period, 2 patients with a short-stay and 4 patients with a long-stay died (p = 0.332). The main symptoms including fever, cough, fatigue, asthma or dyspnea, and myalgia were more common in long-stay patients than in short-stay (p < 0.001). Traditional Chinese medicine (91.2%) and oxygen therapy (60.8%) were widely used; in addition, long-stay patients tended to receive more therapy than short-stay patients did (p < 0.001). Compared with short-stay patients, long-stay patients were significantly older, more likely to have higher levels of platelet count, NLR, CRP, ALT, LDH and D-dimer, as well as lower levels of lymphocyte count, hemoglobin, albumin and creatine kinase (p < 0.001).

The variables with significant association assessed by the univariate logistic regression were shown in Additional file 1: Table S2. In the final multivariate logistic regression model, the number of clinical symptoms > 3 (Odds ratio [OR]: 1.86, 95% confidence interval [CI]: 1.47–2.36), maximum body temperature during hospitalization ≥ 37.3 °C (OR: 2.59, 95% CI: 1.99–3.36), abnormal platelet count (compared with below the lower limit of the normal, normal range, OR: 1.91, 95% CI: 1.03–3.55; increase, OR: 3.33, 95% CI: 1.58–7.02), increased levels of CRP (OR: 1.79, 95% CI: 1.37–2.34), increased LDH levels (OR: 1.62, 95% CI: 1.28–2.04) and decreased levels of albumin (OR: 2.01, 95% CI: 1.57–2.59) were independently associated with the length of hospital stay in moderate patients (Figure 2a). The Hosmer and Lemeshow test of the final model showed an effective goodness-of-fit (p = 0.997). After excluding dead patients, the results were not materially altered (Additional file 1: Table S3).

In order to facilitate clinical application, we further built a risk scoring system to stratify the moderate patients with different length of stay. The risk scoring system was designated as STPCAL score including six variables: number of clinical symptoms, temperature, platelet count, CRP, albumin and LDH. The range of STPCAL score were 0 to 9 points. According to the STPCAL score, patients were classified into one of three risk categories for a longer hospital duration: low (Score 0–1, median = 8 days, with less than 20.0% probabilities), intermediate (Score 2–6, median = 13 days, with 30.0–78.9% probabilities), high (Score 7–9, median = 19 days, with more than 86.5% probabilities) (Table 2). The bootstrapping AUC of the STPCAL score was 0.72 (95% CI: 0.69–0.75). The calibration plot demonstrated the adequate agreement between observed outcome events and predictions by our score with calibration-in-the-large of 0.001 and calibration slope of 0.998 (Figure 3a).

Figure 3.

Calibration plots for predicting the probability of outcomes in COVID-19 patients. a STPCAL score for predicting hospitalization duration in moderate COVID-19 patients. b TRPNCLP score for predicting disease progression in severely ill COVID-19 patients. X-axis is predicted probability by risk scores, and y-axis is the actual probability of outcome events in our population. Dashed line represents the performance of the ideal scores. Dotted line is the apparent accuracy of our risk scores without overfitting correction. Solid line is the bootstrap-correction performance of our risk scores, representing dispersion estimation of future precision

Risk Scoring System to Predict Disease Progression of Severely Ill Patients

Up to the end of the follow-up, there were still 17 severely ill patients who have not been discharged. According to the outcomes from the last follow-up, the 17 patients were classified in progression (n=2) and non-progression group (n=15). Baseline epidemiological and clinical characteristics of severely ill patients were shown in Table 3. The median age of patients in the non-progression group was 64 (56, 71.25) years, and 319 (51.3%) were male. The median age of patients was 68.5 (62, 75) years and 63.1% were male in the progression group. The patients were significantly older (p = 0.003), with more comorbidities such as history of other cardiovascular disease (p = 0.026), history of respiratory disease (p < 0.001), history of other disease (p = 0.005) in the progression group. Traditional Chinese medicine treatment (88.3%) was the most common, followed by oxygen therapy (83.2%) and antiviral therapy (58.7%). There were significantly higher levels of leucocyte count, NLR, CRP, BUN, creatinine, LDH, prothrombin time and D-dimer, but lower levels of lymphocyte count and albumin (p < 0.05) in patients with disease progression than those with non-progression.

The variables with significant association assessed by the univariate logistic regression were shown in Additional file 1: Table S4. In the final multivariate logistic regression model, we found that the maximum body temperature during hospitalization ≥ 37.3 °C (OR: 5.24, 95% CI: 2.70–10.16), history of respiratory diseases (OR: 2.43, 95% CI: 1.20–4.92), decreased platelet count (OR: 2.39, 95% CI: 1.04–5.47), NLR > 5.00 (OR: 3.31, 95% CI: 1.74–6.30), increased levels of creatinine (OR: 2.94, 95% CI: 1.44–6.02), LDH (OR: 1.59, 95% CI: 1.16–2.18) and prothrombin time (OR: 2.51, 95% CI: 1.19–5.27) were independently associated with disease progression in severely ill patients (Figure 2b). The goodness of fit of the final model was acceptable according to Hosmer and Lemeshow test (p = 0.898).

We used beta coefficients of the above significant factors to construct a relative weighted score system, named as TRPNCLP score (temperature, respiratory disease, platelet count, NLR, creatinine, LDH, and prothrombin time score). The AUC of TRPNCLP score by bootstrapping was 0.88 (95% CI: 0.85–0.91), which was higher than that of MuLBSTA score (0.76, 95% CI: 0.73–0.79, p < 0.001). Similar differences were observed by NRI and IDI, indicating that TRPNCLP score had a significantly better reclassification than MuLBSTA score (Table 4). Furthermore, the TRPNCLP score was well-calibrated with calibration-in-the-large and calibration slope equal to 0.004 and 1.002, respectively (Figure 3B). The range of TRPNCLP score were 0 to 16 points. We further classified the TRPNCLP score into 3 levels to stratify the risk of disease progression: low risk (Score 0–5, n = 23 or 5.2%, with less than 12.7% probabilities), intermediate risk (Score 6–11, n = 43 or 34.4%, with 18.6–69.1% probabilities), and high risk (Score 12–16, n = 22 or 88.0%, with more than 77.9% probabilities) (Table 2).