Validation of a specialized evaluation system for COVID-19 in Japan: A retrospective, multicenter cohort study

Abstract

Introduction

Evaluation of a severity grade (SG) is important to classify patients for efficient use of limited medical resources. This study validates two existing evaluation systems for the prevention of the coronavirus disease 2019 (COVID-19) in Japan: a criterion of SG and a list of 14 specialized underlying diseases (SUDs).

Methods

A retrospective cohort was created using electronic medical records from 18 research institutes. The cohort includes 6,050 COVID-19 patients with two types of diagnosis information as follows: SG at hospitalization among mild, moderate I, moderate II, and severe and aggravation after hospitalization.

Results

A crude mortality rate and an aggravation rate increased by the worsening of SG in the COVID-19 cohort. The transition of the aggravation rate was notable for COVID-19 patients with SUD. A conditional probability of the mortality given the aggravation in the COVID-19 cohort was 87.4% compared to mild or moderate patients (approximately 21%–45%) who have the possibility of the aggravation. An odds ratio of the mortality and aggravation information about the SUD list was higher than other variables.

Conclusions

We demonstrated the possibility of improving the criteria of SG by including the SUD list for more effective operation of the criteria of SG. Furthermore, we demonstrated the importance of the prevention of the aggravation based on the conditional probability, and the possibility of predicting the aggravation using the risk factors.

1. Introduction

1.1. Background

Evaluation of a severity grade (SG) is important to classify patients with an infectious disease not limited to the coronavirus disease 2019 (COVID-19) for isolating them. In fact, one study summarized various factors to evaluate the severity of COVID-19 in early stage of its pandemic [1]. In addition, various studies have analyzed details of these factors such as cytokines [2,3], hypertension [4], coexistence of cancer [5], and lymphopenia [6]. Moreover, studies have demonstrated that a typical negative lifestyle habit could constitute the risk factor such as obesity [[7], [8], [9]] and smoking [10]. Considering that one study discussed a prediction model as a helpful tool for evaluating the severity and prognosis of a patient [11], these factors could become explorative variables of this model. However, applying this model to an actual diagnosis by each new inpatient is difficult because these variables are always not recorded in electronic medical records (EMRs) for the execution of this model. Similarly, some studies have applied the existing risk scoring system of pneumonia on COVID-19 patients, such as the A-DROP score [[12], [13], [14]], regarding various meta-analysis studies that have evaluated the negative effects of typical risk factors on COVID-19 patients with pneumonia [[15], [16], [17], [18], [19]].

On the other hand, various guidelines for COVID-19 have been published for situations such as an overview [20,21], a treatment plan by each severity [22,23] and that of an individual situation (e.g., perinatal medicine [24], inpatients without an intensive care [25], and drug recommendation [26]). Similarly, the Ministry of Health, Labour and Welfare provides the guideline of COVID-19 within a criterion of SG among mild, moderate I, moderate II, and severe [27]. In particular, we believe that the criterion of severe patients is important for the efficient operation of limited medical resources because most severe patients have to wear a mechanical ventilator (MV) or be admitted to an intensive care unit (ICU). Therefore, it is necessary to validate the criterion of SG to be able to classify patients irrespective of their nationality.

In addition to the criterion of SG, the Ministry of Health, Labor and Welfare lists 14 specialized underlying diseases (SUDs) [28] to classify those who should prioritized for vaccination. We believe that SUD is useful to classify severe patients more efficiently because previous studies have demonstrated that certain diseases could become the risk factor in COVID-19 patients [[4], [5], [6]]. However, the criterion of SG and the SUD list are now independently operated in Japan. Therefore, we expect to improve the criterion of SG by including the list.

1.2. Study objective

We aim to validate two existing evaluation systems—the criteria of SG and the SUD list—that were designed in Japan toward an efficient and appropriate provision of medical treatment under conditions of limited medical resources.

2. Patients and methods

2.1. Study design and participants

A retrospective cohort of patients treated from April 1, 2019, to September 30, 2021, was created using EMRs collected from 26 cooperative research institutes. These records contain the Diagnostic Procedure Combination (DPC) data submitted by acute hospitals for the social insurance system in Japan. The DPC data included fundamental patient information, such as sex, age, and disease, and a record of hospital charges by treatment. Although 27 institutes originally participated in our study, one institute was excluded due to the lack of the DPC data. Since we aim to evaluate the situation of each institute, this study does not use a sample size estimation that a typical clinical trial is required to include.

Fig. 1 depicts the analysis dataset creation process for our data analysis using the cohort from the 26 institutes. There are four exclusion criteria: (1) any research institute completely lacks monthly data of hospital charges; (2) any individual record not showing COVID-19; (3) any missing value in a body mass index (BMI) that the SUD list requires to record; or (4) any past hospitalization in patients whose hospitalization time is multiple. The 10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10) is used to extract patients with COVID-19. Those who recorded U07.1 or U07.2 in the main disease, diseases in which the most medical resources were used, or diseases that were a direct trigger of hospitalization are considered as COVID-19 patients. The analysis dataset was created following this exclusion and was divided following four SGs evaluated at hospitalization (see Table 1 ).

Fig. 1.

Data processing flowchart.

Table 1.

Definition of severity criteria.

Severity	Symptom	Participants (n = 6,050)
		Number	%
Mild	Nothing in below	4,662	77.1
Moderate I	All	346	5.7
	With dyspnea	4	0.1
	With pneumonia	343	5.7
Moderate II	Necessity of oxygen inhalation	788	13.0
Severe	All	254	4.2
	Admission to ICU	0	0.0
	Necessity of MV	254	4.2

Abbreviations: ICU, Intensive care unit; MV, Mechanical ventilator.

2.2. Definition of the criterion of SG and the SUD list using data items

Table 1 highlights a definition of SG of COVID-19 at hospitalization using existing data items. The description “all” in the table refers to those who fall under one or more detail items (e.g., with dyspnea and with pneumonia in moderate I). Initially, patients who wore MV on the first day of hospitalization or who were admitted to the ICU, retrospectively, were considered severe patients. Next, those who were administered with oxygen on the first day of hospitalization were considered as moderate II patients. Finally, patients whose comorbidity is dyspnea (ICD-10 is R06.0) or pneumonia (ICD-10 is J10.0 or J11.0, or the first three headings of ICD-10 is J12 to J18) were considered as moderate I patients, and the remaining patients were considered as mild patients. Although saturation of percutaneous oxygen (SpO₂) is used to classify mild or moderate patients, this study did not use it because most patients did not record its value upon hospitalization in our existing database within the DPC data or otherwise, such as through a laboratory inspection record.

Table 2 highlights a definition of the 14 SUDs based on the code of ICD-10 of the comorbidity at hospitalization except SUD8, SUD9, SUL10, SUD12, and SUD14 due to the lack of corresponding codes for these four diseases. Furthermore, we classified patients into four subgroups using their age and BMI (I: age <65 & BMI <30; II: age <65 & BMI $\ge$ 30; III: age $\ge$ 65 & BMI <30; and IV: age $\ge$ 65 & BMI $\ge$ 30).

Table 2.

Definition of fourteen SUDs.

SUD		Corresponding code (ICD-10)	Participants (n = 6,050)
			Number	%
SUD1	Chronic respiratory disease	J31.x, J32.x, J35.x, J37.x, J40.x-J47.x, J68.4, J70.1, J70.3, J95.3, J96.1	282	4.7
SUD2	Chronic heart disease within high blood pressure	I10.x-I15.x, I25.x, I31.x, I48.2	826	13.7
SUD3	Chronic nephric disease	N18.x	214	3.5
SUD4	Chronic liver disease	K71.3, K71.4, K71.5, K72.1, K73.x	6	0.1
SUD5	Diabetes	E10.x-E14.x	1,030	17.0
SUD6	Blood disease except iron deficiency anemia	D51.x-D77.x	125	2.1
SUD7	Immunodeficiency within cancer	Cxx.x, D80.x-D89.x	328	5.4
SUD8	Treatment with something to make immune function be worse	Unavailable
SUD9	Neural or neuromuscular disease caused by immunodeficiency	Unavailable
SUD10	Physical decline caused by neural or neuromuscular disease	Unavailable
SUD11	Chromosomal aberration	Q90.x-Q99.x	2	0.0
SUD12	Patients with both physical and intellectual severe disability	Unavailable
SUD13	Sleep apnea syndrome	G47.3	20	0.3
SUD14	Severe mental disorder	Unavailable

Abbreviations: ICD-10, The 10th revision of the International Statistical Classification of Diseases and Related Health Problems; SUD, Specialized underlying disease.

Notes: The digit “x” in the corresponding code such as “J31.x” and “Cxx.x” means that the numeric code in ICD-10 is not used to decide whether patients contract SUD or not.

2.3. Statistical methods

Our primary analysis compared a crude mortality rate (CMR) and an aggravation rate (AGR) by the cohort to evaluate whether the criteria of SG can classify patients. Here, the aggravation is defined as those who were not severe at hospitalization but characterized as being severe patients in Table 1 after hospitalization. Therefore, AGR is defined for only mild or moderate patients at hospitalization as a ratio of aggravated patients and all patients.

First, we compared CMR by SG at hospitalization. We expected this comparison to demonstrate the increase of CMR by the worsening of SG. Second, we compared AGR and CMR in aggravated patients. We expected this comparison to demonstrate the influence of the aggravation on the increase of CMR. In similar, we dealt with the comparison by whether patients contract any SUD or not. Here, a P value is calculated using Fisher's exact test for the comparison above. Finally, we calculated a conditional probability (CP) of the mortality given SG at hospitalization or the aggravation using the Bayes' theorem as follows:

$$P\left(S_i|M\right)=\frac{P\left(M\right)P\left(M|S_i\right)}{P\left(M\right)P\left(M|S_1\right)+P\left(M\right)P\left(M|S_2\right)+P\left(M\right)P\left(M|S_3\right)}$$

$$P\left(A|M\right)=\frac{P\left(M\right)P\left(M|A\right)}{P\left(M\right)P\left(M|A\right)+P\left(M\right)P\left(M|\neg A\right)}$$

$$P\left(M|X\right)=\frac{P\left(M\cap X\right)}{P\left(X\right)}=\left\{\frac{N\left(M\cap X\right)}{N\left(T\right)}\right\}/\left\{\frac{N\left(X\right)}{N\left(T\right)}\right\}$$

whereas $S_i$ means three SGs (1: mild, 2: moderate I, and 3: moderate II), $M$ means the mortality, $A$ means the aggravation, $X$ mean $S_i$ or $A$, and $N(\bullet )$ means the number of those who fall under the value in brackets, where especially $N\left(T\right)$ means the number of all patients. Each $N(\bullet )$ can be seen in Table 4 . We expected this calculation to demonstrate which is more important in mild or moderate patients, the mortality by SG at hospitalization or the aggravation following hospitalization. Therefore, CP was calculated without severe patients.

Table 4.

CP of the mortality given SG or the aggravation.

Given variable	Category	Participants (n, %)						CP (%)
		Survival		Mortality		Total
SG	Mild	4,428	76.4	234	4.0	4,662	80.4	21.3
	Moderate I	319	5.5	27	0.5	346	6.0	33.1
	Moderate II	703	12.1	85	1.5	788	13.6	45.7
Aggravation	No	5,207	89.8	239	4.1	5,446	94.0	12.6
	Yes	243	4.2	107	1.8	350	6.0	87.4
Total		5,450	94.0	346	6.0	5,796	100.0

Abbreviations: CP, Conditional probability; SG, Severity grade.

Our secondary analysis estimated an odds ratio (OR) of the mortality and the aggravation to explore a risk factor using a logistic regression in those who were not severe at hospitalization. Since this analysis would explore a potential risk of the mortality or the aggravation, severe patients were excluded for this analysis. This analysis includes both univariate and multivariate analyses. If all categories record OR with statistical significance (i.e., P value is less than 0.05), the corresponding explorative variable is included in the multiple analysis. However, sex, age, and BMI are excluded despite being statistically significant in the univariate analysis if the explorative variables classified in the SUD or A-DROP score created from these three variables (e.g., subgroup in SUD vs age and BMI) also show the statistical significance, retrospectively. The reason is that there is a high possibility of missed OR caused by a strong correlation between these three variables and variables in SUD or A-DROP score.

In the secondary analysis, there are three classifications of these data items. The first is a data item about SUD: (1) the four subgroups defined by age and BMI; (2) number of contracting SUDs. The second is two items about the A-DROP score that can be created \using data items in our database as follows: (1) “A: Age” extracts male patients whose age is $\ge$ 70 or female patients whose age is $\ge$ 75; (2) “O: Orientation” extracts patients who record any value in the Japan Coma Scale at hospitalization. The third is a data item about patient basic characteristics: (1) sex; (2) age; (3) BMI; (4) a smoking index; (5) an activities daily living (ADL) at hospitalization.

All statistical analyzes were performed using the R programming version 4.1.2, with the P value of <0.05 indicating statistical significance.

3. Results

This study included 6,050 participants in the data analysis. Table 3 shows CMR and AGR by each SG. The number of aggravated patients is an internal number of all patients. Table 3 indicates that CMR and AGR increased by the worsening of SG despite the consideration of SUD. On the other hand, there was no difference of CMR in patients with aggravation by each SG. Furthermore, AGR in patients with SUD was higher than in those without SUD.

Table 3.

CMR and AGR by SG and SUD.

SUD	SG	All patients					Aggravated patients
		Participants (n, %)		Death (n)	CMR (%)	P value	Participants (n, %)		AGR (%)	Death (n)	CMR (%)	P value
												AGR	CMR
All	All	6,050	100.0	416	6.9		350	100.0	5.8	107	30.6
	Mild	4,662	77.1	234	5.0	<0.001	239	68.3	5.1	69	28.9	<0.001	0.122
	Moderate I	346	5.7	27	7.8		29	8.3	8.4	6	20.7
	Moderate II	788	13.0	85	10.8		82	23.4	10.4	32	39.0
	Severe	254	4.2	70	27.6
No	All	3,942	100.0	220	5.6	<0.001	181	100.0	4.6	55	30.4	<0.001	0.733
	Mild	3,125	79.3	113	3.6	<0.001	126	69.6	4.0	31	24.6	<0.001	0.011
	Moderate I	186	4.7	14	7.5		12	6.6	6.5	3	25.0
	Moderate II	491	12.5	54	11.0		43	23.8	8.8	21	48.8
	Severe	140	3.6	39	27.9
Yes	All	2,108	100.0	196	9.3		169	100.0	8.0	52	30.8
	Mild	1,537	72.9	121	7.9	<0.001	113	66.9	7.4	38	33.6	0.003	0.416
	Moderate I	160	7.6	13	8.1		17	10.1	10.6	3	17.6
	Moderate II	297	14.1	31	10.4		39	23.1	13.1	11	28.2
	Severe	114	5.4	31	27.2

Abbreviations: AGR, Aggravation rate; CMR, Crude mortality rate; SG, Severity grade; SUD, Specialized underlying disease.

Notes: The P value in the rows that the column SG shows “Mild” means the results of comparison between the corresponding rate by SG at hospitalization. The word “All” means a comparison between the corresponding rates by whether patients contract any SUD.

Table 4 is a cross-tabulation to calculate CP of the mortality given SG or the aggravation in those who are not severe at hospitalization. CP is known as an inverse probability, one of the most popular frameworks of the Bayes’ theorem. We calculated CP, considering a probability of the cause (SG or the aggravation) given the effect (the mortality). Table 4 indicates that CP of the mortality given the aggravation is remarkably higher than that of mild or moderate patients.

Table 5 shows OR of the mortality and the aggravation by each risk factor in those who were not severe at hospitalization. If OR was >2 with the statistical significance, the corresponding cell was enhanced. Supplementally, Appendix A shows CMR and AGR by each risk factor in Table 5. The appendix B shows summary statistics of these risk factors in a numeric value. Table 5 indicates the risk factors of both the mortality and aggravation using OR when the reference category that we assumed was the lowest risk category. In particular, subgroups (II and IV) and ADL (abnormal) would be important to detect higher risk patients because their OR was >2 in both objective (mortality and aggravation) and regression styles (univariate and multiple).

Table 5.

OR estimation by each risk factor.

Risk factor			Univariate		Multiple
Classification	Variable	Category	Estimation [95% CI]	P Value	Estimation [95% CI]	P Value
(a) Mortality
SUD	Subgroup	I	Ref.		Ref.
		II	2.24 [1.35, 3.71]	0.001	2.08 [1.25, 3.46]	0.004
		III	6.10 [4.52, 8.22]	<0.001	3.30 [2.21, 4.93]	<0.001
		IV	8.56 [4.48, 16.33]	<0.001	4.68 [2.36, 9.27]	<0.001
	Number of	0	Ref.		Ref.
	SUDs	1	1.67 [1.31, 2.14]	<0.001	1.05 [0.81, 1.36]	0.704
		>1	2.14 [1.57, 2.93]	<0.001	1.21 [0.87, 1.69]	0.245
A-DROP Score	A: Age	No	Ref.		Ref.
		Yes	3.88 [3.10, 4.86]	<0.001	1.20 [0.87, 1.65]	0.274
	O: Orientation	No	Ref.		Ref.
		Yes	2.75 [2.05, 3.70]	<0.001	1.28 [0.94, 1.75]	0.111
Basic characteristic	Sex	Male	Ref.
		Female	0.85 [0.68, 1.07]	0.164
	Age	<20	0.03 [0.02, 0.03]	<0.001
	(years)	20 to 64	Ref.
		65 to 74	3.55 [2.58, 4.88]	<0.001
		>74	5.71 [4.34, 7.52]	<0.001
	BMI	<18.5	0.06 [0.05, 0.07]	<0.001
		18.5 to <25	Ref.
		25 to <30	1.11 [0.86, 1.45]	0.419
		30 to <35	0.86 [0.54, 1.35]	0.502
		35 to <40	1.00 [0.43, 2.32]	0.996
		>=40	2.27 [1.01, 5.07]	0.045
	Smoking index	0	Ref.		Ref.
		>0	1.43 [1.15, 1.78]	0.001	1.42 [1.13, 1.78]	0.002
	ADL	Normal	Ref.		Ref.
		Abnormal	5.57 [4.23, 7.33]	<0.001	3.51 [2.62, 4.70]	<0.001
(b) Aggravation
SUD	Subgroup	I	Ref.		Ref.
		II	2.49 [1.74, 3.56]	<0.001	2.22 [1.54, 3.20]	<0.001
		III	2.04 [1.60, 2.61]	<0.001	1.17 [0.90, 1.53]	0.241
		IV	4.62 [2.53, 8.44]	<0.001	2.67 [1.42, 4.99]	0.002
	Number of	0	Ref.		Ref.
	SUDs	1	1.74 [1.36, 2.21]	<0.001	1.31 [1.02, 1.69]	0.036
		>1	2.14 [1.57, 2.93]	<0.001	1.52 [1.10, 2.12]	0.012
A-DROP Score	A: Age	No	Ref.
		Yes	1.19 [0.95, 1.50]	0.128
	O: Orientation	No	Ref.		Ref.
		Yes	1.73 [1.24, 2.41]	0.001	0.98 [0.69, 1.38]	0.898
Basic characteristic	Sex	Male	Ref.		Ref.
		Female	0.53 [0.42, 0.68]	<0.001	0.62 [0.47, 0.80]	<0.001
	Age	<20	0.05 [0.04, 0.06]	<0.001
	(years)	20 to 64	Ref.
		65 to 74	2.46 [1.91, 3.17]	<0.001
		>74	1.05 [0.79, 1.39]	0.746
	BMI	<18.5	0.05 [0.04, 0.06]	<0.001
		18.5 to <25	Ref.
		25 to <30	2.04 [1.59, 2.63]	<0.001
		30 to <35	2.18 [1.51, 3.13]	<0.001
		35 to <40	3.05 [1.66, 5.58]	<0.001
		>=40	2.96 [1.32, 6.64]	0.008
	Smoking index	0	Ref.		Ref.
		>0	1.72 [1.38, 2.15]	<0.001	1.42 [1.11, 1.80]	0.004
	ADL	Normal	Ref.		Ref.
		Abnormal	5.01 [3.84, 6.53]	<0.001	4.74 [3.58, 6.27]	<0.001

Abbreviations: ADL, Activities daily living; BMI, Body mass index; CI, Confidence interval; OR, Odds ratio; SUD, Specialized underlying disease.

4. Discussion

4.1. Study contribution

Initially, there was the effective operation of the criterion of SG because CMR worsened SG upon hospitalization. In particular, CMR in severe patients with COVID-19 was remarkably higher (27.6%) than among all patients (6.9%). This was appropriate because this value was identical to those who were admitted to the ICU (25.7%) [29]. Furthermore, CMR in COVID-19 patients (6.9%) was remarkably lower than that of the avian influenza virus infection (approximately 60%) [30], the Middle East respiratory syndrome (upper 20%, retrospectively) [31], and severe acute respiratory syndrome (approximately 11%, retrospectively) [32].

In addition, we demonstrated that the aggravation after hospitalization was more important than the severe diagnosis at hospitalization with regarding Table 4. The aggravation and the severe diagnosis may be a same condition as wearing MV or admitting to ICU. However, its meaning is totally different because this condition cannot be expected for the aggravation despite being expectable for severe diagnosis at hospitalization. In other words, the appropriate time for the MV or ICU is different for aggravated patients (after hospitalization) and patients with the severe diagnosis (upon hospitalization). This difference indicates inefficient use of limited medical resources within MV and ICU, because the use of MV or ICU for aggravated patients is an unexpected situation in comparison to the initial treatment plan at hospitalization that neither MV nor ICU would be implemented on these patients. During the COVID-19 pandemic, their efficient use is a critical issue to improve the quality of treatment due to a huge demand of MV with inequity problem of its distribution [33]. Moreover, various diseases discussed a new type of MV with both a lower cost of introduction and user friendliness [[34], [35], [36], [37], [38]]. Therefore, we believe that the prevention of the aggravation is required to solve this issue because the prevention can decrease unexpected use of MV and ICU. For example, the inclusion of the SUD list in existing SG criteria (cf. Table 3) could detect those who have a higher possibility of the aggravation at hospitalization because we demonstrated that both CMR and AGR were higher in patients with SUD compared to those without SUD (see Table 2). Moreover, Category IV in the four subgroups in the SUD list showed a higher OR of aggravation with statistical significance (see Table 5). Since this category means patients whose age is $\ge$ 65 and BMI is $\ge$ 30 form the highest risk group, this result would be clinically appropriate and contribute to improving the existing evaluation criteria of SG.

Finally, we discuss a risk prediction model of the aggravation in addition to mortality as a solution of the above issue using various risk factors in Table 5. Although there are some cases of the model for the mortality by COVID-19 using methods of machine learning [[39], [40], [41]], few studies have developed that of the aggravation. In fact, one study recently explored a possibility of a biomarker of the aggravation as a candidate for an explorative variable in the model [42]. However, we demonstrated that certain risk factors could be available to predict the aggravation as same as the mortality such as the four groups in SUD and ADL. Since these factors are recorded though creating the DPC database as a routine process of claiming medical fees, we believe that these factors could be easily implemented on the prediction model compared to the previous studies. Besides the four groups in SUD and ADL, we also explored the variables that could efficiently detect higher-risk patients using the data items in our database because of opaque risk factors in COVID-19 as an emerging infectious disease. For instance, we combined some risk factors into one explorative variable such, as A: age in the A-DROP score classification, using sex and age. Although explorative variables other than the two variables below did not show a remarkable value (OR was >2) in the mortality and the aggravation, they showed the value in only one of the mortality and the aggravation. Therefore, we believe that these other variables would have a possibility of improving the prediction model.

4.2. Limitations

A major limitation was not being able to use a value from a laboratory inspection despite including SpO₂ in the criterion of SG. In comparison to data items in the DPC database, this value was not recorded every day. This occurred as a consequence of the insufficient introduction of EMRs in Japan. In fact, studies have discussed that small- or moderate-scale hospitals cannot introduce EMRs like large scale hospitals [43,44]. However, this limitation did not have a critical negative influence on our data analyses because SpO₂ was not necessary to extract severe patients as shown in Table 1.

On the contrary, there were some minor limitations in this study. The first was that our secondary analysis did not consider the adjustment of confounding factors. However, this could be solved in future work because there are various data analyzing methods for the prediction model within this adjustment by a rapid progress of an information technology. The second minor limitation was that the influence of future variants of COVID-19 that was detected after was not discussed because our database only recorded information until September 30, 2021, and there were no data items that could identify an individual strain such as the Alpha strain. The third was that it was not possible to observe the influence of vaccination history because the database did not include it, though this history was strongly expected to decrease mortality, especially among the elderly.

4.3. Conclusion

In conclusion, we demonstrated that the criteria of SG could effectively identify patients with a higher risk of mortality as those who were severe at hospitalization. Also, we defined patients with aggravation as those who were not severe at hospitalization but wore MV or were admitted to ICU after hospitalization. Moreover, we found that the occurrence of aggravation after hospitalization was a serious risk factor of mortality rather than a severe diagnosis at hospitalization. Finally, we explored the risk factors of mortality and aggravation. This exploration indicated the possibility of our future work building a high-performance prediction model.

Funding

This study was supported by a scholarship donation from the Kansai Economic Federation to the Kyoto University. Kyoto University provided a part of this donation with the Japan Medical Network Association as a research cooperation institution.

Author statements

HF developed the study design, collected data, tackled the data analysis, and wrote the draft of the manuscript. KA was the chief investigator and supervised overall process of the study. All authors contributed to the writing of the final manuscript.

Ethics approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of the Kyoto University Graduate School and Faculty of Medicine, Ethics Committee (approval number, R2963) and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Informed consent

Informed consent was obtained by an opt-out method. The research executive office noted details of this study on their website. The office can consider that the informed consent is obtained unless participants request the office not to use their information.

Declaration of competing interest

HF has no competing interests. KA received a research fund as the director of the Japan Medical Network Association from the Kyoto University.

Appendices.

Table A.

CMR and AGR by each risk factor

Risk factor			All patients (n = 6,050)				Without severe patients (n = 5,796)
Classification	Variable	Category	Participants (n, %)		Mortality (n)	CMR (%)	Participants (n, %)		Aggravation (n)	AGR (%)
SUD	Subgroup	I	2,878	47.6	70	2.4	2,792	48.2	108	3.9
		II	547	9.0	32	5.9	520	9.0	47	9.0
		III	2,526	41.8	299	11.8	2,394	41.3	181	7.6
		IV	99	1.6	15	15.2	90	1.6	14	15.6
	Number of	0	3,942	65.2	220	5.6	3,802	65.6	181	4.8
	SUDs	1	1,492	24.7	131	8.8	1,415	24.4	113	8.0
		>1	616	10.2	65	10.6	579	10.0	56	9.7
A-DROP	A: Age	No	4,123	68.1	166	4.0	3,961	68.3	227	5.7
Score		Yes	1,927	31.9	250	13.0	1,835	31.7	123	6.7
	O: Orientation	No	5,488	90.7	314	5.7	5,342	92.2	307	5.7
		Yes	562	9.3	102	18.1	454	7.8	43	9.5
Basic	Sex	Male	3,624	59.9	270	7.5	3,445	59.4	254	7.4
characteristic		Female	2,426	40.1	146	6.0	2,351	40.6	96	4.1
	Age	<20	245	4.0	0	0.0	245	4.2	0	0.0
	(years)	20 to 64	3,180	52.6	102	3.2	3,067	52.9	155	5.1
		65 to 74	1,091	18.0	100	9.2	1,017	17.5	118	11.6
		>74	1,534	25.4	214	14.0	1,467	25.3	77	5.2
	BMI	<18.5	733	12.1	52	7.1	707	12.2	29	4.1
		18.5 to <25	3,185	52.6	203	6.4	3,069	53.0	137	4.5
		25 to <30	1,486	24.6	114	7.7	1,410	24.3	123	8.7
		30 to <35	473	7.8	31	6.6	447	7.7	41	9.2
		35 to <40	112	1.9	8	7.1	105	1.8	13	12.4
		≥40	61	1.0	8	13.1	58	1.0	7	12.1
	Smoking index	0	3,083	51.0	175	5.7	2,980	51.4	137	4.6
		>0	2,967	49.0	241	8.1	2,816	48.6	213	7.6
	ADL	Normal	3,149	52.0	68	2.2	3,133	54.1	72	2.3
		Abnormal	2,901	48.0	348	12.0	2,663	45.9	278	10.4

Abbreviations: ADL, Activities daily living; AGR, Aggravation rate; BMI, Body mass index; CMR, Crude mortality rate; SUD, Specialized underlying disease.

Table B.

Summary statistics of each risk factor in a numerical value

Variable	Mean	SD	Median	Min	Max
Number of SUDs	0.5	0.7	0	0	5
Age (years)	58.8	20.8	60	0	104
BMI	23.73	5.44	23.4	0.0	60.5
Smoking index	1,982.4	3,750.8	0	0	9999

Abbreviations: BMI, Body mass index; SD, Standard deviation; SUD, Specialized underlying disease.