Characteristics of patients with COVID-19 who have deteriorating chest X-ray findings within 48 h: a retrospective cohort study

The severity of chest X-ray (CXR) findings is a prognostic factor in patients with coronavirus disease 2019 (COVID-19). We investigated the clinical and genetic characteristics and prognosis of patients with worsening CXR findings during early hospitalization. We retrospectively included 1656 consecutive Japanese patients with COVID-19 recruited through the Japan COVID-19 Task Force. Rapid deterioration of CXR findings was defined as increased pulmonary infiltrates in ≥ 50% of the lung fields within 48 h of admission. Rapid deterioration of CXR findings was an independent risk factor for death, most severe illness, tracheal intubation, and intensive care unit admission. The presence of consolidation on CXR, comorbid cardiovascular and chronic obstructive pulmonary diseases, high body temperature, and increased serum aspartate aminotransferase, potassium, and C-reactive protein levels were independent risk factors for rapid deterioration of CXR findings. Risk variant at the ABO locus (rs529565-C) was associated with rapid deterioration of CXR findings in all patients. This study revealed the clinical features, genetic features, and risk factors associated with rapid deterioration of CXR findings, a poor prognostic factor in patients with COVID-19.

Chest imaging facilitates the diagnosis and management of patients 19 .In clinical settings, CXR can be easily performed using mobile CXR units in a dedicated, isolated room to reduce the transmission risk 19 .Although CXR is considered less sensitive for detecting pulmonary involvement in early-stage disease 20 , it is useful for monitoring the progression of lung abnormalities in COVID-19, especially in critically ill ICU patients 21 .The requirement for ventilatory support is associated with worsening findings early after admission 22 ; moreover, mortality can be predicted based on CXR findings before and after ICU admission 23 .However, the impact of deteriorating CXR findings on outcomes other than ventilatory support needs and the clinical and genetic characteristics of patients with deteriorating CXR findings remain unclear.
A nationwide multi-center consortium was established to address the COVID-19 pandemic in Japan 24,25 .Since the pandemic's onset, the network has been collecting DNA, RNA, and plasma samples, as well as detailed clinical information, from patients with COVID-19 throughout Japan on a long-term basis.The first Japanese large-scale genome-wide association study (GWAS) on COVID-19 reported that genetic variants, including DOCK2, had a population-specific association with oxygenation requirements by patients with COVID-19 24 .
It is important for clinicians to be able to predict the severity of a patient's illness, especially during a pandemic, in order to use medical resources appropriately.We hypothesised that rapid deterioration in CXR findings after hospitalisation is associated with subsequent worsening of the disease.We aimed to investigate the prognosis, clinical characteristics, and genetic characteristics of patients with worsening CXR findings during early hospitalization.

Study design and settings
This retrospective cohort study recruited hospitalized COVID-19 cases through the Japan COVID-19 Task Force 24 .From February 2020 to May 2021, data obtained from consecutive patients aged ≥ 18 years, who were diagnosed with COVID-19 based on polymerase chain reaction tests and agreed to participate in the study, were entered into electronic case record forms by attending physicians at the affiliated research institutions.The exclusion criteria were as follows: (1) patients from other countries (2) patients with incomplete medical records, e.g., insufficient data regarding the presence/absence of chest radiographic deterioration within 48 h of admission or critical outcomes (Fig. 1).All patients provided written or oral informed consent.The study design was approved by the ethics committees of Keio University School of Medicine (20,200,061) and related research institutions.All methods were performed in accordance with the relevant guidelines and regulations.

Statistical analysis
Regarding baseline characteristics, categorical variables were presented as frequencies and proportions, while continuous variables were presented as means and standard deviations.We compared data according to the presence/absence of lung infiltrates in > 50% of the fields within 48 h of admission using the t-test and Chi-square test, as appropriate.The median hospitalization duration was estimated using the Kaplan-Meier method and compared using the log-rank test.
To investigate the association between rapid deterioration of imaging and radiographic findings, CXR findings (unilateral/bilateral ground-glass opacity [GGO]/consolidation) were adjusted, followed by multivariate logistic regression analysis.Additionally, a Cochran-Armitage trend test was performed to determine the rapid deterioration of CXR findings as well as the tendency of radiographic findings to exhibit no, unilateral, or bilateral shadows.
To assess the association between the rapid deterioration of CXR findings and clinical outcomes (death, most severe disease, IMV use, and ICU treatment), we performed multivariate logistic regression analyses with adjustment for baseline CXR findings, number of days from symptom onset to hospitalisation, and characteristics known as predictors of COVID-19 severity (age, body mass index [BMI], hypertension, diabetes mellitus, cardiovascular disease, malignancy, chronic obstructive pulmonary disease [COPD], asthma, chronic liver disease, and chronic kidney disease) [2][3][4][5] .
To identify the clinical characteristics of patients with rapid deterioration of CXR findings, we adopted a holdout method, where data from two-thirds of the cases were used as training data, while the remaining data were used as test data to validate the performance of the prediction model for patients with COVID-19 who have rapid deterioration of CXR findings.Receiver operating characteristic (ROC) curve analysis was performed to determine appropriate cut-off values of continuous variables for rapid deterioration of CXR findings using the Youden index.We performed multivariable analysis using a logistic regression model with a backward selection procedure to select the combinations of variables.The variables were selected based on a threshold p-value of 0.05.The Bayesian information criterion was also applied to select the optimal model among the existing models.We performed ROC curve analysis for the model, with the performance of the prediction models being assessed using the area under the curve (AUC).
The dosage effects of the variants on rapid deterioration of CXR findings were evaluated using logistic regression models, with age (included only in all age analyses) and sex as covariates.To evaluate the association between the ABO blood groups and rapid deterioration of CXR findings, we performed a multivariate logistic regression analysis of the A/B/AB/O blood groups and other blood groups, with adjustment for age and sex.
To ensure the reliability of the data, two clinicians from different affiliations, M.W. (National Hospital Organization Tokyo Medical Center) and T.K. (Keio University School of Medicine), independently reviewed the initial and follow-up CXRs and assessed for rapid deterioration of CXR findings (n = 45).Agreement between the two clinicians' readings was analysed using Cohen's kappa coefficient (κ), with κ values graded as slight (0.00-0.20), fair (0.21-0.40), moderate (0.41-0.60), substantial (0.61-0.80), and almost perfect (0.81-1.00) according to Landis and Koch criteria 36 .
Data were presented as adjusted odds ratios (aORs) with 95% confidence intervals (CIs).Statistical significance was set at p < 0.05.All statistical analyses were performed using the JMP 16 program (SAS Institute) and SAS software (version 9.4; SAS Institute).

Statement of Ethics
All patients involved in this study provided written or oral informed consent, and the study design was approved by the ethics committees of Keio University School of Medicine (20,200,061) and the affiliated research institutions.
Compared with patients with two CXRs within 48 h, those without were generally younger (P < 0.001) and comprised more women (P = 0.001), and patients with less severity (P = 0.003) (Supplementary Table 3).

Association between baseline CXR findings and rapid deterioration
Figure 2 shows the comparison of the baseline CXR findings between patients with and without rapid deterioration of CXR findings.For patients with GGO and consolidation on both modalities, the frequency of rapid deterioration of CXR findings increased in the following order: no shadow, unilateral shadow, and bilateral shadow; CXR GGO (P < 0.001), CXR consolidation (P < 0.001).Multivariate logistic regression analysis revealed that unilateral shadow, bilateral shadow, GGO, and consolidation were associated with an increased likelihood of rapid deterioration of CXR findings (Table Supplementary Table 5).

Association between rapid deterioration of CXR findings and outcomes
Table 2 shows the results of multiple logistic regression analysis using parameters that included rapid deterioration of CXR findings, GGO, consolidation on CXR, and number of days from symptom onset to hospitalisation, in addition to the previously reported prognostic factors [2][3][4][5] .Rapid deterioration of CXR findings was an independent risk factor for death (aOR

Association of the ABO blood groups with rapid deterioration of CXR findings
Table 5 shows the association between rapid deterioration of CXR findings and the ABO blood groups.Blood group AB was associated with an increased risk of rapid deterioration of CXR findings compared with the non-AB blood types (aOR [95% CI] = 1.84 [1.15-2.95]),after adjustment for age and sex.

Discussion
Our study provided three novel findings with clinical relevance.First, patients with COVID-19 with rapid CXR deterioration had poorer clinical outcomes than those without; accordingly, and they may require more aggressive treatment.Second, we identified predictors of CXR deterioration upon admission.Therefore, clinicians should pay more attention to CXR deterioration after hospitalization of patients with these risk factors.Third, we identified the genetic risk factors for CXR deterioration in Japanese patients with COVID-19.The CXR severity score on admission is a risk factor for death, severe disease, IMV use, and ICU treatment in patients with COVID-19 9,11,12,14,15 .Deterioration of CXR findings after hospitalization influences the requirement for ventilatory support after admission 22 .Moreover, studies have investigated mortality prediction using CXR findings before and after ICU admission 23 and the association of the worst CXR scores during hospitalization with discharge and death 37 .However, we found that rapid deterioration of CXR findings was an independent risk factor for death, most severe disease, ICU admission, and tracheal intubation.Furthermore, multivariate analysis using baseline CXR findings revealed that these relationships were robust.Compared with baseline CXR findings, rapid deterioration of CXR findings had a higher aOR of predicting worse outcomes.Our findings suggest clinicians should be aware of CXR deterioration, especially within 48 h of admission, regardless of the baseline CXR findings.
We identified the risk factors for rapid CXR deterioration.Several reported factors contribute to severe COVID-19 development [2][3][4][5][6][7][8] .In our study, patients with rapid CXR deterioration showed many of these risk factors for death and severe disease.CXR consolidation; concomitant cardiovascular disease and COPD; and elevated body temperature, AST, serum potassium, and CRP levels, were independent risk factors for rapid deterioration of CXR findings.A prediction model for CXR deterioration using these risk factors showed high Table 2. Predictors of death, most severe disease, IMV use, and ICU treatment.The adjusted odds ratio was estimated by logistic regression.For continuous variables, the unit odds ratio for each 1-unit change is shown.95% CI, 95% confidence interval; aOR, adjusted odds ratio; BMI, body mass index; COPD, chronic obstructive pulmonary disease; CXR, chest X-ray; GGO, ground-glass opacity; ICU, intensive care unit; IMV, invasive mechanical ventilation.Previous GWAS reports have suggested an association between the genetic characteristics of patients with COVID-19 and the severity of COVID-19 24,38 .We performed analyses using risk variants and ABO blood groups to characterize the genetic characteristics related to rapid deterioration of CXR findings.We identified wholepopulation and population-specific risk variants at the ABO locus (rs529565-C) among the 15 genes extracted from previous reports 24,[27][28][29][30][31][32][33][34][35] .The DOCK2 locus has been associated with severe disease in patients with COVID-19 aged < 65 years 24 .Therefore, we performed an analysis of the association between DOCK2 and acute worsening of imaging in patients with COVID-19 aged < 65 years, but found no significant differences.Regarding the risk variant of the ABO locus, we observed a risk of rapid deterioration of CXR findings in patients with COVID-19 in the AB blood group.In patients with COVID-19, an association between the AB blood group and most severe disease has been reported 39 .The AB blood group is reported to be more susceptible to a variety of infections 40 and at higher risk for thrombosis 41 , which may contribute to the same mechanisms as most severe disease and rapid deterioration of CXR findings.However, further studies are needed to explore this point.
Our study had three main limitations.First, we could not determine the mechanisms underlying rapid deterioration of CXR findings from a virological perspective.Therefore, our findings may not reflect the current Table 3. Predictors of rapid deterioration in imaging findings.The adjusted odds ratio was estimated by logistic regression.95% CI, 95% confidence interval; aOR, adjusted odds ratio; AST, aspartate aminotransferase; COPD, chronic obstructive pulmonary disease; CRP, C-reactive protein; CXR, chest X-ray; K, potassium.

Figure 1 .
Figure 1.Participant selection process Overall, 1976 patients with COVID-19 were hospitalized during the study period.We excluded 51 non-Japanese patients and 269 patients with incomplete medical records.COVID-19, coronavirus disease.

Figure 2 .
Figure 2. Radiographic findings in patients with rapid deterioration of CXR findings.Proportion of cases showing rapid deterioration of CXR findings according to the distribution of GGO/consolidation on CXR.CXR, chest X-ray; GGO, ground-glass opacity.

Table 1 .
Baseline characteristics of the patients.Data are shown as mean ± standard deviation or percentage values.Data were analyzed using the χ 2 test, t-test, or log-rank test, as appropriate.BMI, body mass index; CXR, chest X-ray; COPD, chronic obstructive pulmonary disease.

Table 4 .
Association of COVID-19 risk variants with rapid deterioration in CXR findings.The adjusted odds ratio was estimated by logistic regression, with adjustment for age and sex.The odds ratio represents the incremental odds for each unit increase in allele dosage.The significance threshold based on Bonferroni's correction was set at p < 0.0033.95% CI, 95% confidence interval; aOR, adjusted odds ratio; EA, effect allele; RA, reference allele; COVID-19, coronavirus disease; CXR, chest X-ray.

Table 5 .
The association of the ABO blood group with rapid deterioration of CXR findings.The adjusted odds ratio was estimated by logistic regression with adjustments for age and sex.95% CI, 95% confidence interval; aOR, adjusted odds ratio; CXR, chest X-ray.