Abstract
Objective
To identify predictive factors of unfavourable outcome among patients hospitalized for COVID-19.
Methods
We conducted a monocentric retrospective cohort study of COVID-19 patients hospitalized in Paris area. An unfavourable outcome was defined as the need for artificial ventilation and/or death. Characteristics at admission were analysed to identify factors predictive of unfavourable outcome using multivariable Cox proportional hazard models. Based on the results, a nomogram to predict 14-day probability of poor outcome was proposed.
Results
Between March 15th and April 14th, 2020, 279 COVID-19 patients were hospitalized after a median of 7 days after the first symptoms. Among them, 88 (31.5%) patients had an unfavourable outcome: 48 were admitted to the ICU for artificial ventilation, and 40 patients died without being admitted to ICU. Multivariable analyses retained age, overweight, polypnoea, fever, high C-reactive protein, elevated us troponin-I, and lymphopenia as risk factors of an unfavourable outcome. A nomogram was established with sufficient discriminatory power (C-index 0.75), and proper consistence between the prediction and the observation.
Conclusion
We identified seven easily available prognostic factors and proposed a simple nomogram for early detection of patients at risk of aggravation, in order to optimize clinical care and initiate specific therapies.
KEY MESSAGES
Since novel coronavirus disease 2019 pandemic, a minority of patients develops severe respiratory distress syndrome, leading to death despite intensive care. Tools to identify patients at risk in European populations are lacking.
In our series, age, respiratory rate, overweight, temperature, C-reactive protein, troponin and lymphocyte counts were risk factors of an unfavourable outcome in hospitalized adult patients.
We propose an easy-to-use nomogram to predict unfavourable outcome for hospitalized adult patients to optimize clinical care and initiate specific therapies.
Keywords: COVID-19, risk factors, nomogram
Introduction
Since December 2019, in Wuhan, China has experienced an outbreak of coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) [1,2]. This outbreak was declared a pandemic by World Health Organization [3] and has now spread to other regions of the world, including European countries as Italy, Spain, France, and the United Kingdom and the United States of America. For unknown reasons, a minority of patients develops severe respiratory distress syndrome after a week of mild disease, leading to death despite intensive care in a high proportion of patients. Several retrospective series from China investigated risk factors associated with unfavourable outcome [4–8], but the extrapolation of these results to the occidental population is questionable. Besides, because there is a narrow gate between the initial respiratory symptoms (typically cough and fever followed by dyspnoea) leading to a visit to the emergency room and the onset of respiratory distress syndrome requiring admission in the intensive care unit (ICU), rapidly identifying patients at high risk of an unfavourable outcome is needed, to optimise clinical care during this critical period.
Therefore, this study aimed to identify, in a retrospective cohort of adult patients with COVID-19 admitted to the hospital, simple, quantitative, and easily available predictive factors of unfavourable outcome through multivariate Cox regression analysis and a nomogram model.
Patients and methods
Study population and organization of care
This is a retrospective cohort study of patients aged 18 or over with COVID-19 requiring hospitalization at Beaujon hospital between March 15th and April 14th, 2020. Beaujon Hospital is a 450 beds University hospital located in the northern suburbs of Paris, an area of low socio-economic level. Patients presenting to the Emergency Department, with suspected or confirmed COVID-19 requiring oxygen therapy or who could not be followed in outpatient’s clinic were hospitalized in the COVID-19 unit.
Rapidly after the onset of COVID-19 epidemics, organization and functioning of the hospital were modified to increase the pool of dedicated beds, up to a maximum of 98 beds in addition to an extension of ICU to a maximum of 44 dedicated beds during the study period. Local guidelines for the management of COVID-19 patients were edited by infectious diseases and intensive care specialists, according to the current state of knowledge. This organization allowed a homogenization of the management of the patients.
For the present study, we included all hospitalized patients aged 18 or over, for whom a diagnosis of COVID-19 was made, based on positive SARS-CoV-2 real-time reverse transcriptase-polymerase chain reaction (RT-PCR) on nasal swabs and/or typical abnormalities on chest computed tomography (CT). Patients whose diagnosis was not confirmed, as patients who were directly admitted to ICU without being hospitalized in our COVID-19 unit, were excluded from the analyses. Patients were treated according to the local recommendations, and some patients were included in randomized controlled trials, evaluating tocilizumab (NCT04331808) or anakinra (NCT04341584) versus standard of care. The study was approved by the local institutional review board (IRB 00006477).
Data collection
Demographic, clinical, and laboratory data at admission, treatments, and outcome were extracted from electronic medical records using a standardized anonymized data collection form. The most intense level of oxygen support within the first 24 h of admission was recorded. Overweight was defined by a body mass index (BMI) ≥25 kg/m2. Patients were classified as follows: mild (requiring <3 L/min to maintain oxygen saturation [SpO2] ≥ 97%), moderate (requiring between 3 L/min and 5 L/min of oxygen to maintain SpO2 ≥ 97%) and severe (respiratory rate ≥30 breaths/minute and/or requiring >5 L/min of oxygen). Pulmonary extension of the lesions on chest CT fell into five standardized categories: minimal (<10%), moderate (10–25%), extended (25–50%), severe (50–75%) and critical (>75%). Patients who were discharged within the 14 days following admission were contacted by phone by the investigators to update their clinical status.
Outcomes
An unfavourable outcome was defined by the need for artificial ventilation or death within the 14 days following admission, whichever occurred first. Artificial ventilation included non-invasive mechanical ventilation (NIV) (continuous positive airway pressure and/or high flow nasal cannula [HFNC]) and/or invasive mechanical ventilation.
Statistical analysis
Descriptive analyses were conducted using mean (SD) or median (IQR) for quantitative variables and count (percent) for qualitative variables. Survival until unfavourable outcome was estimated using the Kaplan–Meier method, from the time between admission and the use of NIV, HFNC, and/or invasive mechanical ventilation, death, or day 14, whichever occurred first. Missing values were handled using multivariate imputations by chained equations (“mice” library in R), based on a fully conditional specification, where each incomplete variable is imputed by a separate model [9]. We made the assumption that the missing data were missing at random. Variables with the highest rate of missing data were ferritin (71.3%), us troponin I (43.6%), BMI (30.8%), and liver enzymes (12.5%).
To identify risk factors associated with unfavourable outcomes, variables including baseline characteristics, laboratory findings, and imaging data were analysed by univariable Cox proportional hazard regression analyses. Continuous predictors (i.e. respiratory rate, biological features) were categorized according to clinically relevant values. Hazard ratios (HRs) were presented with 95% confidence intervals (95%CI). Variables associated with p < .10 in univariate analyses were candidates for the multivariable Cox proportional hazard regression model. Because we focussed on easily available clinical and biological factors, imaging data and treatments were not included in the multivariable model.
The selection of the variables included in the final model was based on 1000 bootstrapped samples from the original sample with replacement. On each sample, a backward stepwise selection was used, and the final model was defined as the most frequently selected model among the 1000 bootstrapped samples [10].
Selected variables were then incorporated in a nomogram to predict the probability of 14-day survival without unfavourable outcome using statistical software (“rms” libraryin R). The nomogram function draws a partial nomogram for obtaining predictions from the fitted model manually: point scores are obtained for each predictor and the user must add the point scores manually to read predicted values on the final axis of the nomogram [11]. The maximum score of each variable was set as 100. The model performance was evaluated by the predictive accuracy for individual outcomes (discriminating ability) and by the accuracy of point estimates of the survival function (calibration). We used the Harrell’s concordance index (C-index) to evaluate the discriminative ability of the nomogram, using 100 bootstrap resampling [11]. The C-index estimates the probability of concordance in rank order between the predicted and observed outcomes. A C-index of 0.5 indicates the absence of discrimination, whereas a C statistic of 1.0 indicates perfect separation of patients with different outcomes. We developed a calibration plot to visualize the agreement between the predicted and observed 14-day survival without unfavourable outcome to assess the predicted accuracy of the nomogram. To take into account overfitting, we plotted an optimism-corrected calibration curve using 100 bootstrap resampling. In a well-calibrated model, the predictions should fall on a 45-degree diagonal line. In addition, patients in the validation set were assigned into different risk groups according to tertiles of prognostic scores. Survival curves for different risk groups were generated by the Kaplan–Meier analysis and compared using the log-rank test in order to investigate the discriminative ability of the nomogram.
All tests were two-sided, and a P-value <0.05 was considered statistically significant. All analyses were performed with R version 3.6.1. (R Foundation for Statistical Computing, Vienna, Austria)
Results
Study population
Between March 15th and April 14th, 2020, 334 patients were hospitalized for suspicion of COVID-19. We excluded 55 (16.5%) patients whose diagnosis was ruled out after a negative RT-PCR test and non-evocative chest CT. Finally, 279 patients met the inclusion criteria and were included in the analyses (Supplementary Figure 1).
The socio-demographic, clinical, biological features, treatments, and outcomes of the study population are described in Table 1. Median (IQR) age was 66 (53.6–76.4) years, and 183 (65.6%) patients were men. Most frequent comorbidities were hypertension for 131 (47%) patients and diabetes for 77 (27.6%). Notably, there were only 13 (4.7%) patients who declared being current smokers. Median (IQR) time between the first signs and admission was 7 (4–9) days. Most frequent clinical symptoms at admission were dyspnoea, cough, and fever, and 40 (14.3%) patients reported anosmia and/or gustatory dysfunction. Diarrhoea was reported in almost 20% of the cases. More than half of the patients were classified as mild forms, whereas 51 (18.3%) and 83 (29.7%) had moderate and severe forms, respectively. Almost all patients received antibiotics (94.6%) and pre-emptive or curative anticoagulation. Forty-four (15.8%) patients received intravenous glucocorticoids. Thirty patients were included in a randomised controlled trial: 4 (1.4%) were treated with anakinra, and 11 (3.9%) with tocilizumab.
Table 1.
Main characteristics of the study population (N = 279).
| Overall (N = 279) | No unfavourable outcome at D14 (N = 191) | Unfavourable outcome (N = 88) | |
|---|---|---|---|
| Age, years | 64.8 (16.1) | 62.9 (16.3) | 69.1 (15.0) |
| Gender | |||
| Female | 96 (34.4) | 72 (37.7) | 24 (27.3) |
| Male | 183 (65.6) | 119 (62.3) | 64 (72.7) |
| Smoking | |||
| Active smoker | 13 (4.7) | 7 (3.7) | 6 (6.8) |
| Non-smoker or former smoker | 186 (66.7) | 128 (67.0) | 58 (65.9) |
| Unknown | 80 (28.7) | 56 (29.3) | 24 (27.3) |
| Patient history | |||
| Diabetes | 77 (27.6) | 54 (28.3) | 23 (26.1) |
| Hypertension | 131 (47.0) | 85 (44.5) | 46 (52.3) |
| COPD | 16 (5.7) | 10 (5.2) | 6 (6.8) |
| Immunodeficiencya | 18 (6.5) | 14 (7.3) | 4 (4.5) |
| Coronary heart disease | 30 (10.8) | 19 (9.9) | 11 (12.5) |
| Clinical features | |||
| Time since first symptoms (days) | 6.8 (4.8) | 7.2 (5.2) | 5.8 (3.6) |
| Weight (kg) | 78.0 (17.4) | 77.5 (16.4) | 78.7 (19.0) |
| Body mass index (kg/m²) | |||
| <25 | 71 (25.4) | 54 (28.3) | 17 (19.3) |
| 25–30 | 65 (23.3) | 35 (18.3) | 30 (34.1) |
| ≥30 | 57 (20.4) | 38 (19.9) | 19 (21.6) |
| Missing | 86 (30.8) | 64 (33.5) | 22 (25.0) |
| Respiratory rate, per minute | 26.2 (6.7) | 25.2 (6.1) | 28.2 (7.5) |
| <30 | 220 (78.9) | 164 (85.9) | 56 (63.6) |
| ≥30 | 59 (21.1) | 27 (14.1) | 32 (36.4) |
| Body temperature ≥ 38 °C | 110 (39.4) | 65 (34.0) | 45 (51.1) |
| Cough | 190 (68.1) | 135 (70.7) | 55 (62.5) |
| Dyspnoea | 198 (71.0) | 124 (64.9) | 74 (84.1) |
| Myalgia | 58 (20.8) | 41 (21.5) | 17 (19.3) |
| Diarrhoea | 55 (19.7) | 40 (20.9) | 15 (17.0) |
| Confusion | 23 (8.2) | 12 (6.3) | 11 (12.5) |
| Anosmia or gustatory dysfunction | 40 (14.3) | 32 (16.8) | 8 (9.1) |
| Classification | |||
| Mild | 145 (52.0) | 122 (63.9) | 23 (26.1) |
| Moderate | 51 (18.3) | 38 (19.9) | 13 (14.8) |
| Severe | 83 (29.7) | 31 (16.2) | 52 (59.1) |
| Laboratory findingsb | |||
| Lymphocyte count (×109/L) | 1.2 (1.0) | 1.3 (1.2) | 0.9 (0.6) |
| C-reactive protein (mg/L) | 126.3 (91.1) | 108.8 (82.3) | 163.8 (98.2) |
| Urea (mmol/L) | 8.3 (8.6) | 7.1 (7.4) | 11.1 (10.2) |
| Creatinine (µmol/L) | 108.2 (75.7) | 100.2 (72.8) | 125.7 (79.3) |
| AST (U/L)b | 71.2 (101.9) | 59.7 (45.6) | 95.1 (164.6) |
| ALT (U/L)b | 45.8 (59.1) | 42.4 (42.5) | 52.7 (83.6) |
| D-dimer (µg/mL)b | 3.4 (7.3) | 2.0 (3.9) | 5.2 (10.0) |
| Us Troponin I (ng/L)b | 72.7 (421.9) | 28.2 (49.7) | 153.0 (700.1) |
| Serum ferritin (µg/L)b | 1465.3 (1584.2) | 1234.7 (1786.1) | 1835.2 (1113.8) |
| Computed tomography extension | |||
| Minimal | 27 (9.7) | 23 (12.0) | 4 (4.5) |
| Moderate | 91 (32.6) | 81 (42.4) | 10 (11.4) |
| Extended | 75 (26.9) | 47 (24.6) | 28 (31.8) |
| Severe | 27 (9.7) | 13 (6.8) | 14 (15.9) |
| Critical | 8 (2.9) | 2 (1.0) | 6 (6.8) |
| Not evaluated | 51 (18.3) | 25 (13.1) | 26 (29.5) |
| Treatment | |||
| Oxygen need (L/min) | 3.06 (3.33) | 2.13 (2.14) | 5.07 (4.40) |
| Lopinavir/ritonavir | 49 (17.6) | 26 (13.6) | 23 (26.1) |
| Hydroxychloroquine | 13 (4.7) | 11 (5.8) | 2 (2.3) |
| Antibiotics | 264 (94.6) | 182 (95.3) | 82 (93.2) |
| Glucocorticoids | 44 (15.8) | 19 (9.9) | 25 (28.4) |
| Anakinra | 4 (1.4) | 1 (0.5) | 3 (3.4) |
| Tocilizumab | 11 (3.9) | 4 (2.1) | 7 (8.0) |
| Anticoagulation | |||
| None | 18 (6.5) | 13 (6.8) | 5 (5.7) |
| Curative | 36 (12.9) | 14 (7.3) | 22 (25.0) |
| Pre-emptive | 225 (80.6) | 164 (85.9) | 61 (69.3) |
| Unfavourable outcome | |||
| Deceased without artificial ventilation | 40 (14.3) | NA | 40 (45.4) |
| Artificial ventilation | 48 (17.2) | NA | 48 (54.5) |
| HFNC or NIV | 28 (10.0) | NA | 28 (31.8) |
| Invasive mechanical ventilation | 27 (9.7) | NA | 27 (30.7) |
| Deceased after artificial ventilation | 11 (3.9) | NA | 11 (12.5) |
Results are expressed as number (%) for categorical variables and as mean (standard deviation) for quantitative variables. AST: aspartate aminotransferase; ALT: alanine aminotransferase; D14: day 14; HFNC: high flow nasal cannula; NA, non available; NIV: non-invasive mechanical ventilation.
aImmunodeficiency included solid cancers, haematological malignancies, transplantations and patients treated with immunosuppressant drugs.
bSGOT and SPOT were available for 244 (87.5%) patients, us troponin I levels for 157 (56.3%), and ferritin for 112 (29.6%) patients.
Primary endpoint
During the study period, 88 (31.5%) patients had an unfavourable outcome: 48 (17.2%) had artificial ventilation and 40 (14.3%) died without being admitted to ICU for artificial ventilation, following “do not resuscitate” order.
Among the 48 patients who had artificial ventilation, 28/48 (58.3%) were treated with NIV and/or HFNC, 27/48 (56.3%) required invasive mechanical ventilation (following NIV and/or HFNC for 7), and 11/48 (22.0%) patients died within the 14 days following hospital admission.
Median (IQR) time between admission in the COVID unit and unfavourable outcome was 3 (2–6) days (Supplementary Figure 1).
Factors associated with unfavourable outcome
Results of the univariable and multivariable models are shown in Table 2. The multivariable Cox regression model retained age (HR per decade 1.23; 95%CI 1.06–1.43), an elevated BMI ≥25 kg/m2 (HR 2.14; 95%CI 1.32–3.47), a high respiratory rate ≥30/minute (HR 1.91; 95%CI 1.21–3.03), body temperature ≥38 °C (HR 1.95; 95%CI 1.26–3.02), an elevated C-reactive protein ≥100 mg/L (HR 2.34; 95%CI 1.40–3.90), an elevated us troponin I (HR 1.62; 95%CI 1.03–2.56), and a lymphocyte count <1 G/L (HR 1.66; 95%CI 1.03–2.67) as factors associated with an unfavourable outcome (Table 2, Figures 1 and 2). Kaplan–Meier curves for these prognostic factors are shown in Supplementary Figure 2.
Table 2.
Hazard ratio (HR) and 95% confidence interval (CI) for the risk of unfavourable outcome (N = 279).
| Univariable analysis |
Multivariable analysis |
||
|---|---|---|---|
| HR (95%CI); p Value | HR (95%CI); p Value | ||
| Age, per decade | 1.18 (1.04–1.35); p = .012 | 1.23 (1.06–1.43); p = .007 | |
| Gender | |||
| Female | Ref | Ref | |
| Male | 1.57 (0.98–2.50); p = .061 | 1.53 (0.95–2.4); p = .080 | |
| Smoking | |||
| Active smoker | Ref | – | |
| Non-smoker or former smoker | 1.02 (0.64–1.63); p = .933 | – | |
| Patient history | |||
| Diabetes | 0.91 (0.57–1.46); p = .696 | – | |
| Hypertension | 1.27 (0.83–1.93); p = .267 | – | |
| COPD | 1.26 (0.55–2.88); p = .587 | – | |
| Immunosuppression | 0.64 (0.23–1.74); p = .382 | – | |
| Myocardial ischaemia | 1.16 (0.62–2.18); p = .646 | – | |
| Clinical features | |||
| Body mass index (kg/m²) | |||
| <25 | Ref | Ref | |
| ≥25 | 1.57 (1.00–2.48); p = .050 | 2.14 (1.32–3.47); p = .002 | |
| Respiratory rate (per minute) | |||
| <30 | Ref | Ref | |
| ≥30 | 2.75 (1.78–4.24); P < .001 | 1.91 (1.21–3.03); p = .006 | |
| Body temperature ≥ 38 °C | 1.81 (1.19–2.75); p = .005 | 1.95 (1.26–3.02); p = .003 | |
| Cough | 0.76 (0.49–1.17); p = .217 | – | |
| Anosmia—gustatory dysfunction | 0.57 (0.28–1.18); p = .130 | – | |
| Laboratory findings* | |||
| Lymphocyte count ≥1 × 109/L | 2.08 (1.31–3.31); p = .002 | 1.66 (1.03–2.67); p = .038 | |
| C-reactive protein ≥100 mg/L | 2.80 (1.73–4.53); P < .001 | 2.34 (1.40–3.90); p = .001 | |
| Urea ≥ 7 mmol/L | 2.32 (1.53–3.54); P < .001 | – | |
| AST ≥ 3xULN | 1.72 (1.13–2.62); p = .011 | – | |
| Us Troponin I ≥ 16 ng/L | 1.65 (1.08–2.51); p = .020 | 1.62 (1.03–2.56); p = .037 | |
Univariable analyses were computed with Cox regression model. Survival was calculated from the time of admission to transfer in intensive care unit for non-invasive mechanical ventilation including continuous positive airway pressure, high flow nasal cannula and/or invasive mechanical ventilation, or death, whichever occurred first. Variables included in the final model were those associated with the unfavourable outcome in univariable analyses at p < .10 and which were the most frequently selected in a backward-stepwise selection algorithm based on 1000 bootstrapped samples with replacement. Imaging data and treatment were not included in the multivariable models. AST: aspartate aminotransferase; HR: hazard ratios; 95%CI: 95% confidence interval; Ref: Reference; ULN: upper limit normal value.
Figure 1.
Description of the outcome according to age categories, respiratory rate, C-reactive protein, and lymphocyte count at hospital admission (N = 279).
Figure 2.
Forrest plot showing hazard ratios (95% confidence interval) for the risk of unfavourable outcome by multivariate analysis (N = 279). BMI: body mass index; 95% CI: 95% confidence interval.
Nomogram
A nomogram to predict the 14-day survival without unfavourable outcome probability is shown in Figure 3. The nomogram was created based on the seven prognostic factors determined in the multivariable analysis. Higher total points based on the sum of the assigned number of points for each factor in the nomogram were associated with a worse prognosis. The distribution of the nomogram prediction score is shown in Supplementary Figure 3.
Figure 3.
Nomogram predicting the probability of 14-day survival without unfavourable outcome in hospitalized adult patients for COVID-19. The nomogram was based on 7 prognostic factors (see the Model Specifications and Predictors in the Methods section). Total points based on the sum of the assigned number of points for each factor in the nomogram are used to predict 14–day survival without unfavourable outcome probability. As an example, a 60 year-old man (50 points), with a body mass index at 28 kg/m2 (49.5 points), with a respiratory rate at 22 breaths/min (0 point), a body temperature at 39 °C (40 points), lymphopenia (33 points), an elevated C-reactive protein (55 points) and a low us-troponin I level (0 point) will have a total of 227.5 points, and thus a 14-day probability of survival without unfavourable outcome of 50%. BMI: body mass index; CRP: C-reactive protein; troponin: us troponin I; prob: probability.
To further assess the discriminative ability of the model, the predicted probability of survival without unfavourable outcome was then plotted as Kaplan–Meier curves stratified by three groups (<150; 150–200; >200) of the predicted probability calculated from the nomograms, based on the tertiles (Figure 4). Patients with the lowest predicted survival without unfavourable outcome (score >200) had a substantially worse outcome (39.1% 14-day survival without unfavourable outcome) compared with patients with a score <150 or between 150–200 (85.6% and 78.0% survival without unfavourable outcome, respectively) (p < .001).
Figure 4.
Kaplan–Meier curves demonstrating survival without unfavourable patients in hospitalized adult patients according to the nomogram-prediction score.
The discriminative ability of the final model was sufficient, with a C-statistics of 0.75. The calibration plot for the prediction of 14-day survival without unfavourable outcome demonstrated a good consistence between the prediction and the observation (Supplementary Figure 4).
Discussion
In this French monocentric retrospective study based on 279 hospitalized patients, we identified simple and rapidly available risk factors that predict the likelihood of an unfavourable outcome, defined by the need for artificial ventilation and/or death.
Among the study population, disease manifestations were generally similar to those described in large case series from Chinese hospitals [1,12–14]. Most of our patients were men with traditional cardiovascular risk factors, including hypertension, diabetes mellitus, overweight, and/or obesity, which confirms observations suggesting that underlying cardiovascular disease is associated with an increased risk of severe forms of COVID-19 [15]. In contrast, there were very few patients with known immunosuppression such as neoplasia or solid organ transplant, although many of those patients are usually followed in our hospital. The low prevalence of active smokers was also noted, which agrees with some recent data suggesting a potential protective role of nicotine against COVID-19 [16]. Gastrointestinal symptoms appeared to be more common than in Chinese series, but similar to the prevalence found in a large American series [17]. Similarly, as described in some sizeable European series [18], higher rates of anosmia and/or gustatory dysfunction were observed compared to the Chinese series. Theses geographic differences could reflect different epidemiological or environmental factors, different genetic backgrounds and/or differential reporting.
In our series of hospitalized patients, almost one third required artificial ventilation or died within the few days following admission. This percentage was more than five times higher that in some Chinese series [13]. Potential explanations include higher disease severity in our cohort since testing and hospitalizations in France were limited mainly to patients with severe disease requiring oxygen therapy or to patients with comorbidities. The low socio-economic level of our population with potential difficulties in care access leading to diagnosis delays might also contribute to this high rate.
For hospitalized patients, we identified seven prognostic factors: age, BMI, respiratory rate, temperature, lymphocyte count, C-reactive protein, and troponin levels. Those factors are easily and rapidly available, within a few minutes for clinical characteristics, and within a few hours for laboratory findings. Thus, identification of patients at high risk can be easily made in the emergency room. Those predictive factors reflect several aspects of COVID-19 pathophysiology, including the viral response phase (lymphopenia), the host inflammatory response phase (fever, elevated inflammatory markers) and the cardiac tropism (troponin) on patients at risk (elevated age and overweight) [19]. These results also suggest that COVID-19 unfavourable outcome might be due to virus-activated “cytokine storm syndrome”, exacerbated inflammatory responses, or myocardial involvement [20–22].
To our knowledge, only a few studies have proposed a score and/or a nomogram to consider multiple predictive factors. In a series of 208 patients admitted to Fuyan hospital, Ji et al. suggested a nomogram and a score to predict disease progression (severe disease or requirement of mechanical ventilation, or worsening of lung CT findings), which included age, comorbidities, lymphocyte count and lactate dehydrogenase levels [7]. In another Chinese series of 1590 hospitalized patients, Chen et al. developed a nomogram predicting mortality with age, cardiovascular disease, dyspnoea and biological findings such as procalcitonin or liver enzymes [6]. While those findings present several similarities with our study, they were assessed in Chinese patients with different outcomes, and thus might not be extrapolated to European populations, where indications for testing and hospitalizations might be different.
We acknowledge some limitations to our study. First, this is a monocentric retrospective study, and some data might be missing, especially some baseline biologic data, when local guidelines did not include systematic laboratory testing recommendations. However, we tried to minimize the subsequent bias by performing multiple imputations. Moreover, as a retrospective study, we could not set up a validation cohort to assess the predictive accuracy of our nomogram. However, we used a bootstrap resampling cohort, and found a sufficient C-index, and a good consistence between the prediction and the observation in the calibration curves.
To conclude, we identified several prognostic factors in patients hospitalized for COVID-19 and propose an easy-to-use nomogram to predict unfavourable outcome in a European cohort. If validated, our findings could be used to early detect patients for whom optimization of clinical care and initiation of specific therapies should be considered during this critical period.
Supplementary Material
Acknowledgements
The authors are indebted to all persons (physicians, surgeons, radiologists, biologists, medical students, and paramedical staff) who were involved in the Beaujon COVID-19 Unit. The authors would like to thank Louise Arnaud, Victoria Bruce, Michel-Gabriel Cazenave, Lou Chantriaux, Marianne Fontaine, and Emma Solignac for their help on data collection.
Author contributions
All authors have made substantial contributions to this work and have approved the final version of the manuscript. Concept and design: YN, FC, VH, BF, AG. Acquisition of data: FC, AG. Statistical analysis: YN. Interpretation of data: YN, FC, VH, BF, AG. Writing original draft: YN, BF, AG. Writing review and editing: all authors.
Disclosure statement
None of the authors declared any competing interest in link with this study.
References
- 1.Huang C, Wang Y, Li X, et al. Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. The Lancet. 2020;395(10223):497–506. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Lu R, Zhao X, Li J, et al. Genomic characterisation and epidemiology of 2019 novel coronavirus: implications for virus origins and receptor binding. The Lancet. 2020;395(10224):565–574. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.WHO announces COVID-19 outbreak a pandemic 2020. http://www.euro.who.int/en/health-topics/health-emergencies/coronavirus-covid-19/news/news/2020/3/who-announces-covid-19-outbreak-a-pandemic. [accessed May 2, 2020].
- 4.Wu C, Chen X, Cai Y, et al. Risk factors associated with acute respiratory distress syndrome and death in patients with coronavirus disease 2019 pneumonia in Wuhan, China. JAMA Intern Med. 2020;180(7):934. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Ruan Q, Yang K, Wang W, et al. Clinical predictors of mortality due to COVID-19 based on an analysis of data of 150 patients from Wuhan, China. Intensive Care Med. 2020;46(5):846–848. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Chen R, Liang W, Jiang M, et al. Risk factors of fatal outcome in hospitalized subjects with coronavirus disease 2019 from a nationwide analysis in China. Chest. 2020;158(1):97–105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Ji D, Zhang D, Xu J, et al. Prediction for progression risk in patients with COVID-19 pneumonia: the CALL Score. Clin Infect Dis. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Zhang J, Wang X, Jia X, et al. Risk factors for disease severity, unimprovement, and mortality of COVID-19 patients in Wuhan, China. Clin Microbiol Infect. 2020. DOI: 10.1093/cid/ciaa414/5818317 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Buuren S. v. Groothuis-Oudshoorn K. Mice: multivariate imputation by chained equations in R. J Stat Softw. 2011;45:1–67. [Google Scholar]
- 10.Austin PC, Tu JV.. Bootstrap methods for developing predictive models. Am Stat. 2004;58(2):131–137. [Google Scholar]
- 11.Harrell FE. Regression modeling strategies: with applications to linear models, logistic and ordinal regression, and survival analysis. Cham: Springer International Publishing; 2015. [Google Scholar]
- 12.Zhou F, Yu T, Du R, et al. Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study. The Lancet. 2020;395(10229):1054–1062. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Guan W, NiZ, Hu Y, et al. Clinical characteristics of coronavirus disease 2019 in China. N Engl J Med. 2020;382(18):1708–1720. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Wang D, Hu B, Hu C, et al. Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus–infected pneumonia in Wuhan, China. JAMA. 2020;323(11):1061–1069. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Mai F, Del Pinto R, Ferri C. COVID-19 and cardiovascular diseases. J Cardiol 2020. DOI: 10.1016/j.jjcc.2020.07.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Changeux J-P, Amoura Z, Rey F, et al. A nicotinic hypothesis for Covid-19 with preventive and therapeutic implications. C R Biol. 2020;343:33–39. [DOI] [PubMed] [Google Scholar]
- 17.Goyal P, Choi JJ, Pinheiro LC, et al. Clinical characteristics of covid-19 in New York City. N Engl J Med. 2020;382(24):2372–2374. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Lechien JR, Chiesa-Estomba CM, Place S, et al. Clinical and epidemiological characteristics of 1,420 European patients with mild-to-moderate coronavirus disease 2019. J Intern Med 2020. DOI: 10.1111/joim.13089 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Siddiqi HK, Mehra MR. COVID-19 illness in native and immunosuppressed states: a clinical-therapeutic staging proposal. J Heart Lung Transplant. 2020;39(5):405–407. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Mehta P, McAuley DF, Brown M, et al. COVID-19: consider cytokine storm syndromes and immunosuppression. The Lancet. 2020;395(10229):1033–1034. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Hadjadj J, Yatim N, Barnabei L, et al. Impaired type I interferon activity and exacerbated inflammatory responses in severe Covid-19 patients. Science. 2020;369(6504):718-724. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Qin C, Zhou L, Hu Z, et al. Dysregulation of immune response in patients with coronavirus 2019 (COVID-19) in Wuhan, China. Clin Infect Dis. 2020;71:762–768. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.