Epidemiological Characteristics and Factors Associated with Critical Time Intervals of COVID-19 in Eighteen Provinces, China: A Retrospective Study.

BACKGROUND: As COVID-19 ravages continuously around the world, more information on the epidemiological characteristics and factors associated with time interval between critical events is needed to contain the pandemic and to assess the effectiveness of interventions.
METHODS: Individual information on confirmed cases from January 21 to March 2 was collected from provincial or municipal health commissions. We identified the difference between imported and local cases in the epidemiological characteristics. Two models were established to estimate the factors associated with time interval from symptom onset to hospitalization (TOH) and length of hospital stay (LOS) respectively.
RESULTS: Among 7,042 cases, 3392 (48.17%) were local cases and 3304 (46.92%) were imported cases. Since the first intervention was adopted in Hubei on January 23, the daily reported imported cases reached a peak on January 28 and gradually decreased since then. Imported cases were on average younger (41 vs. 48), and had more male (58.66% vs. 47.53%) compared to local cases. Furthermore, imported cases had more contacts with other confirmed cases (2.80 ± 2.33 vs. 2.17 ± 2.10), which were mainly within family members (2.26 ± 2.18 vs. 1.57 ± 2.06). The TOH and LOS were 2.67 ± 3.69 and 18.96 ± 7.63 days respectively, and a longer TOH was observed in elderly living in the provincial capital cities that were higher migration intensity with Hubei.
CONCLUSIONS: Measures to restrict traffic can effectively reduce imported spread. However, household transmission is still not controlled, particularly for the infection of imported cases to elderly women. It is still essential to surveil and educate patients about the early admission or isolation.

Introduction

As of September 20, 2020, a total of more than 30 million confirmed cases of coronavirus disease 2019 (COVID-19), as well as more than 900,000 deaths had been reported by World Health Organization (WHO) in the worldwide (World Health Organization, 2020a, World Health Organization, 2020b). At the same time, China had reported 85,291 lab confirmed cases with 4,634 deaths (China National Health Commission of the People’s Republic of China, 2020a). Despite the WHO and international community declared and took many efforts to control this pandemic in time, our knowledge about the COVID-19 is still very limited, and the number of daily reported cases is still increasing sharply worldwide (Organization, 2020b).

In the context of the rapid spread of COVID-19, a full understanding of the epidemiological characteristics of this infectious disease is crucial in epidemic control and public policy practices. Several studies conducted in China, Italy and the United States have reported some epidemiological characteristics of COVID-19 in the initial phase (Grasselli et al., 2020, Liang et al., 2020, Price-Haywood et al., 2020, Richardson et al., 2020, Wu and McGoogan, 2020), However, there is still a lack of research on the space-time characteristics in the populations of imported and local cases respectively which is of great significance. Imported cases play a very important role in the disease spreading, especially it is an indicator for predicting new clusters of infections. Understanding its epidemiological characteristics would help us to assess the possible effect of non-pharmaceutical interventions (NPIs), such as travel restrictions (Desjardins et al., 2020, Gilbert et al., 2020). Furthermore, considering the changes in susceptible populations, exposure opportunity and intervention of disease over epidemic progresses and locations, the epidemiological characteristics of disease should hence be estimated spatiotemporally in order to better describe the epidemic (Zhang et al., 2020a). For example, the space-time characteristics of COVID-19 revealed by previous studies can prioritize locations and the best time for different NPIs (Desjardins et al., 2020, Lai et al., 2020b, Masrur et al., 2020). Therefore, exploring the epidemiological characteristics of imported cases from a space-time perspective is critical and provides guidance for countries on interventions taken at different periods and regions, specifically in resource-scarce countries and regions.

As a highly contagious disease, early detection, isolation, hospitalization and diagnosis of COVID-19 are also important for control and they can effectively reduce the risk of disease transmission (Bi et al., 2020, Rong et al., 2020, Thompson, 2020). Delay in hospitalization or isolation may lead to prolonged periods of infectiousness, and increase the difficulty and burden of infectious disease control. Previous studies have described some characteristics of patients with COVID-19 including the time interval between key events (Liang et al., 2020, Tian et al., 2020, Yu et al., 2020). In addition, existing literature also brought to light the reduction in the time interval from symptom onset to hospitalization/isolation after various interventions (Li et al., 2020, Zhang et al., 2020a). However, little is known about individual-level influence factors associated with delaying hospital admission and length of hospital stay. Identifying these factors would not only help us predict the medical burden and reasonably allocate medical resources, but also would inform response efforts across the world.

In this study, we described the spatiotemporal distribution of the COVID-19 in eighteen provinces of China (outside Hubei province) and investigated the epidemiological characteristics in the population of imported cases and local cases, from the beginning of this epidemic until it was under good control. We further assessed the critical influence factors associated with time interval from symptom onset to hospitalization (TOH) and length of hospital stay (LOS), including demographic and temporal and spatial characteristics.

Materials and Methods

Data Source and Study Population

We constructed a retrospective cohort study for COVID-19 confirmed cases, based on the detailed information published by the provincial or municipal health commissions in eighteen provinces of China (outside Hubei province) from January 21 to March 2. The details of sampling and data collection are shown in Fig. 1 . Data collectors were trained and divided into five groups of two according to provinces to collect timely epidemiological data of confirmed cases. LinkMed EDC were used for data entry, the two collectors in each group entered the same data, and we conducted data verification and consistency test in real-time.

Fig. 1

Flowchart of population selection.

Definition of Available Variables

Specifically, demographic characteristics, epidemiological history and date of critical event were extracted from the official report of the confirmed case details. (1) Demographic information including age, gender, residence at the time of diagnosis and type of symptoms were included in our analysis. (2) Epidemiological history includes history of travel or residence in other regions and contact history of confirmed cases. According to whether the patient had a travel or residence history in other regions within 14 days before diagnosis and likely exposure to pathogens in that regions, the patient was divided into imported and local cases. Similarly, we can identify whether patients had contacted with confirmed cases of family and non-family members. (3) The dates of events include the date of symptoms onset, hospitalization/isolation, CDC diagnosis and recovery/death. Hospitalization/isolation is defined as a patient receiving regular hospital treatment (not includes small medical institutions such as clinics and community health service centers), or a mandatory isolation measure implemented by the community. In this study, we used the time interval between two events to analyze this data, including time interval from symptom onset to hospitalization (TOH) and length of hospital stay (LOS).

Additionally, we also collected information on the intensity of migration from Hubei to these 18 provinces in the week before January 23, which was obtained from the Baidu migration map (Baidu Map, 2020). Migration intensity between provinces and Hubei was categorized into four levels: strong connection (≥0.15%), medium connection [0.05%-0.15%), weak connection [0.03-0.05%) and very weak connection (<0.03). Finally, according to the daily trend of new cases and date of intervention, we divided the entire epidemic into five periods from the beginning of the epidemic (Jan 21) to Mar 2. The first period is before January 23, when Wuhan took measures of traffic restrictions and lockdown, since then every week works as one period, until the last period is a recession of this epidemic after February 14.

Statistical Analyses

We described the epidemic scale in 18 provinces and the proportion of imported cases spatiotemporally. Meantime, the demographic characteristics of imported and local cases were reported. In addition, two models were established to identify and quantify the relevant sociodemographic factors to TOH and LOS respectively. In the first model, we estimated the factors associated with TOH using a generalized linear model with a Poisson distribution and a log link. Besides, the odds ratio (OR) and their 95% confidence intervals (CI) were calculated after incorporating multiple variables (Coxe et al., 2009, SAS Publishing, 2016). In the second model, an accelerated failure time (AFT) model was used to handle the survival data with both left and right censored (Kalbfleisch, 2002, Paul, 2010). In our study of analyzing factors associated with LOS, left censoring would occur if we know that a patient recovered before Marth 2, but the exact time cannot be obtained. Similarly, right censoring would occur for patients who are confirmed in the later phase of the epidemic. Moreover, we included the TOH in the model and used the hazard ratio (HR) and their 95% CIs to identify the difference in LOS among recovered patients with different characteristics. Based on the distribution of LOS which is denoted by T, we established the Weibull model, written as,where is a random disturbance term, and ,…, , and σ are parameters to be estimated. Then we applied a likelihood function with censored to estimate the parameter values.Here, is the time of the event or the time of censoring, is an indicator variable with a value of 1 if is uncensored or 0 if censored. The term ) denotes the individual's probability density function, is a survivor function representing the probability of outcome event time in the interval (a, b). For left censoring data,  = 1; while for right censoring data,  = 0. All statistical analysis was conducted using SAS 9.4 (SAS Institute Inc., North Carolina, USA). P < 0.05 was considered statistically significant.

Results

Spatiotemporal characteristics of the epidemic

Among 7,042 cases, 3392 (48.17%) of patients were local cases and 3304 (46.92%) of patients were imported cases, and less than 5% (346) of other patients were unable to confirm their travel history within 14 days before diagnosis. The temporal and spatial distribution of imported and local cases is shown in Fig. 2 . From panel A, we can see that the greater the intensity of migration with Hubei, the more cases in the province. For provinces with migration intensity greater than 0.03%, the proportion of imported cases to total cases was about 50%. However, for provinces including Tianjin, Ningxia and Hebei with very weak connection (<0.03%) with Hubei, they had more local cases than imported cases. Since the first intervention was adopted in Hubei on January 23, the daily reported imported cases reached the highest on January 28, and the proportion of imported cases to the total cases gradually decreased over time, reaching 50% on February 2 (Fig. 2B).

Fig. 2

Distribution of the number of COVID-19 cases by region and calendar date during January 21 to March 2 in 18 provinces in China.

Panel A. The map shows the influence of all imported cases in 18 provinces as of March 2, and these provinces have different migration identity from Hubei.

Panel B. The daily number of new cases in 18 provinces divided into imported case and local case.

Demographic characteristics and the time interval of key events

Table 1 presents the characteristics of analytic confirmed patients. Among all patients included in the analysis, the mean age was 45 years, and 53.13% were male. Imported cases were on average younger (41 vs. 48), had more male (58.66% vs. 47.53%), had more provincial capital residents (18.87% vs. 16.89%) and lower proportion of clearly confirmed case contact history (28.13% vs. 56.48%), whether family member confirmed case (23.47% vs. 39.04%) or non-family (7.13% vs. 23.15%), compared to local cases. After adjustment for measured confounders by logistic regression, similar results were still found. The most common symptoms included fever (79.99%) and cough (56.46%). These patterns are similar in both imported cases and local cases. Compared to imported cases from other provinces (outside Hubei), cases imported from Hubei had fewer provincial capital residents (17.73% vs. 23.79%), and lower proportion of clearly contacted with a confirmed case (24.41% vs. 44.13%). For time interval, the frequency and best-fitting probability density function for TOH and LOS are present in Fig. 3 respectively. As shown in Table 2 bottom, the mean TOH was 2.67 ± 3.69 days, and imported cases had shorter TOH than local cases (2.48 ± 3.55 vs. 2.89 ± 3.87). In addition, the mean LOS was 18.96 ± 7.63 days, and imported cases had similar LOS to local cases (19.22 ± 7.84 vs. 18.65 ± 7.32).

Table 1

Demographic characteristics in imported and local patients.

Characteristics	All Patients(n = 7042)		Import Patients(n = 3392)				Local Patients(n = 3304)		Unknown Patients(n = 346)	Odds Ratio (95% CI)*
			From HubeiProvince (n = 2753)	From other provinces (n = 639)	All(n = 3392)
no./no.of patients with data (%)
Male	3593/6763(53.13)		1566/2674(58.56)	374/633(59.08)	1940/3307(58.66)		1550/3261(47.53)a		103/195(52.82)	1.55(1.38-1.74)
Age group
≤19	363/6684(5.43)		146/2633(5.55)	21/626(3.35)	167/3259(5.12)		187/3235(5.78)a		9/190(4.74)	Reference
20-29	832/6684(12.45)		425/2633(16.14)	88/626(14.06)	513/3259(15.74)		298/3235(9.21)		21/190(11.05)	1.59(1.21-2.08)
30-39	1421/6684(21.26)		694/2633(26.36)	167/626(26.68)	861/3259(26.42)		519/3235(16.04)		41/190(21.58)	1.51(1.18-1.93)
40-49	1442/6684(21.57)		592/2633(22.48)	147/626(23.48)	739/3259(22.68)		656/3235(20.28)		47/190(24.74)	1.04(0.81-1.33)
50-59	1399/6684(20.93)		471/2633(17.89)	127/626(20.29)	598/3259(18.35)		756/3235(23.37)		45/190(23.68)	0.76(0.59-0.97)
≥60	1227/6684(18.36)		305/2633(11.58)	76/626(12.14)	381/3259(11.69)		819/3235(25.32)		27/190(14.21)	0.48(0.37-0.62)
Means ± SD	44.67 ± 16.57		41.12 ± 14.87	42.61 ± 13.83b	41.41 ± 14.68		48.00 ± 17.76b		43.72 ± 14.89
Median(P25-P50)	45(33-56)		40(31-52)	42(33-52)c	41(31-52)		49(35-60)c		44(34-55)
City of resident at the time of diagnosis
Other city	5810/7062(82.27)		2265/2753(82.27)	487/639(76.21)	2752/3392(81.13)		2746/3304(83.11)		312/346(85.25)	Reference
Provincial capital	1252/7062(17.73)		488/2753(17.73)	152/639(23.79)a	640/3392(18.87)		558/3304(16.89)a		54/346(14.75)	1.19(1.03-1.39)
Symptoms
Fever#	2938/3673(79.99)		1168/1435(81.39)	294/365(80.55)	1462/1800(81.22)		1409/1796(78.45)		67/77(87.01)	0.92(0.78-1.08)
Cough†	1293/2290(56.46)		505/877(57.58)	140/211(63.35)	645/1098(58.74)		629/1153(54.55)		19/39(48.72)	0.96(0.83-1.11)
Fatigue†	353/2290(15.41)		144/877(16.42)	35/211(15.84)	179/1098(16.30)		172/1153(14.92)		2/39(5.13)	0.82(0.63-1.07)
Cold†	320/2290(13.97)		120/877(13.68)	26/211(11.76)	146/1098(13.30)		170/1153(14.74)		4/39(10.26)	0.73(0.56-0.96)
Sore throat†	277/2290(12.10)		123/877(14.03)	29/211(13.12)	152/1098(13.84)		123/1153(10.67)a		2/39(5.13)	1.11(0.84-1.48)
Headache†	238/2290(10.39)		106/877(12.09)	21/211(9.50)	127/1098(11.57)		109/1153(9.45)		2/39(5.13)	0.94(0.69-1.29)
Muscle or joint pain†	161/2290(7.03)		54/877(6.16)	21/211(9.50)	75/1098(6.83)		84/1153(7.29)		2/39(5.13)	0.77(0.53-1.12)
Digestive symptoms†	156/2290(6.81)		50/877(5.70)	17/211(8.06)	67/1098(6.10)		85/1153(7.37)		4/39(10.26)	1.11(0.76-1.63)
Dyspnea†	67/2290(2.93)		19/877(2.17)	10/211(4.52)	29/1098(2.64)		37/1153(3.21)		1/39(2.56)	0.84(0.47-1.49)
Epidemiological contact history‡
unclear	4242/7062(60.07)		2081/2753(75.59)	357/639(55.87)a	2438/3392(71.88)		1438/3304(43.52)a		346/346(100)	Reference
clear	2820/7062(39.93)		672/2753(24.41)	282/639(44.13)	954/3392(28.13)		1866/3304(56.48)		0/346(0)	0.38(0.34-0.43)
Family-confirmed patient contact history‡
unclear	4976/7062(70.46)		2174/2753(78.97)	422/639(66.04)a	2596/3392(76.53)		2014/3304(60.96)a		346/346(100)	Reference
clear	2086/7062(29.54)		579/2753(21.03)	217/639(33.96)	796/3392(23.47)		1290/3304(39.04)		0/346(0)	0.56(0.49-0.63)
Non-family-confirmed patient contact history‡
unclear	6055/7062(85.74)		2602/2753(94.52)	548/639(85.76)a	3150/3392(92.87)		2539/3304(76.85)a		346/346(100)	Reference
clear	1007/7062(14.26)		151/2753(5.48)	91/639(14.24)	242/3392(7.13)		765/3304(23.15)		0/346(0)	0.26(0.22-0.31)

#no./no.of patient with symptoms and reported.

†no./no.of patient with symptoms and reported (Except for patients with only fever symptoms).

‡We can trace the history of contact with confirmed patient.

aP < 0.05, Chi-square test; bP < 0.05, Student t-test; cP < 0.05, Wilcoxon rank sum test. These three methods were used to compare the difference between imported and local case and the difference between imported from Hubei province and from other province outside Hubei.

*Odds Ratio represents the comparison between imported cases and local cases after adjusted for region, diagnosis date and other confounders in this table. CI denotes confidence interval.

Fig. 3

The time interval from symptom onset to hospitalization and length of hospital stay.

Panel A shows the frequency (blue histograms) and best-fitting probability density function (Poisson, red curves) for time interval from symptom onset to hospitalization(≥0). Panel B shows the frequency (blue histograms) and best-fitting probability density function (Weibull, red curves) for length of hospital stay.

Table 2

Description of confirmed case contact history of patients and several time intervals.

Characteristics	All Patients (n = 7042)		Import Patients(n = 3392)				Local Patients (n = 3304)		Unknown Patients (n = 346)
			From HubeiProvince (n = 2753)	From otherprovinces (n = 639)	All(n = 3392)
Number of confirmed cases the patient who has contacted#
1	1275/2820(45.21)		202/672(30.06)	106/282(37.59)a	308/954(32.29)		967/1866(51.82)a		42/105(40.00)
2	759/2820(26.91)		223/672(33.18)	79/282(28.01)	302/954(31.66)		457/1866(24.49)		29/105(27.62)
≥3	786/2820(27.87)		247/672(36.76)	97/282(34.40)	344/954(36.06)		442/1866(23.69)		34/105(32.38)
Means ± SD	2.38 ± 2.20		2.86 ± 2.26	2.66 ± 2.51 b	2.80 ± 2.33		2.17 ± 2.10b		0.49 ± 1.26
Median(P25-P50)	2(1-3)		2(1-4)	2(1-3) c	2(1-4)		1(1-2)c		0(0-0)
Number of confirmed cases of family members#
0	734/2820(26.03)		93/672(13.84)	65/282(23.05)a	158/954(16.56)		576/1866(30.87)a		21/105(20.00)
1	831/2820(29.47)		171/672(25.45)	68/282(24.11)	239/954(25.05)		592/1866(31.73)		36/105(34.29)
2	657/2820(23.30)		199/672(29.61)	81/282(28.72)	280/954(29.35)		377/1866(20.20)		21/105(20.00)
≥3	598/2820(21.21)		209/672(31.1)	68/282(24.11)	277/954(29.04)		321/1866(17.20)		27/105(25.71)
Means ± SD	1.80 ± 2.13		2.38 ± 2.14	1.98 ± 2.25 b	2.26 ± 2.18		1.57 ± 2.06b		0.57 ± 0.98
Median(P25-P50)	1(0-2)		2(1-3)	2(1-2) c	2(1-3)		1(0-2)c		0(0-1)
Number of confirmed cases of non-family members#
0	1813/2820(64.29)		521/672(77.53)	191/282(67.73)a	712/954(74.63)		1101/1866(59.00)a		71/105(67.62)
1	704/2820(24.96)		78/672(11.61)	62/282(21.99)	140/954(14.68)		564/1866(30.23)		13/105(12.38)
2	180/2820(6.38)		37/672(5.51)	8/282(2.84)	45/954(4.72)		135/1866(7.23)		16/105(15.24)
≥3	123/2820(4.36)		36/672(5.36)	21/282(7.45)	57/954(5.97)		66/1866(3.54)		5/105(4.76)
Means ± SD	0.58 ± 1.12		0.47 ± 1.20	0.68 ± 1.56 b	0.53 ± 1.32		0.60 ± 1.01		0.66 ± 1.25
Median(P25-P50)	0(0-1)		0(0-0)	0(0-1) c	0(0-1)		0(0-1)c		0(0-1)
Interval from symptom onset(days)
To hospitalization, n	3343		1337	370	1707		1546		90
Means ± SD	2.67 ± 3.69		2.25 ± 3.50	3.33 ± 3.58b	2.48 ± 3.55		2.89 ± 3.87b		2.57 ± 3.12
Median(P25-P50)	1(0-4)		1(0-4)	3(0-6)c	1(0-4)		2(0-5)c		2(0-4)
To CDC diagnosis, n	4697		1860	439	2299		2277		121
Means ± SD	5.92 ± 4.16		5.39 ± 3.89	6.95 ± 4.31b	5.69 ± 4.02		6.15 ± 4.28b		5.90 ± 4.08
Median(P25-P50)	5(3-8)		4(3-7)	7(4-9)c	5(3-8)		5(3-9)c		5(3-8)
To recovery, n	1066		529	114	643		390		33
Means ± SD	21.74 ± 8.13		21.43 ± 8.21	23.73 ± 8.95b	21.84 ± 8.38		21.62 ± 7.80		21.33 ± 7.01
Median(P25-P50)	20(16-26)		20(16-25)	23(17-29)c	20(16-26)		20(16-26)		21(18-24)
Interval from hospitalization(days)
To CDC diagnosis, n	4667		1910	486	2396		2157		114
Means ± SD	3.35 ± 2.98		3.24 ± 2.85	3.41 ± 3.09	3.27 ± 2.90		3.44 ± 3.06		3.32 ± 2.82
Median(P25-P50)	2(1-4)		2(1-4)	2(1-4)	2(1-4)		2(1-5)		2(2-4)
To recovery, n	1229		607	127	734		463		32
Means ± SD	18.96 ± 7.63		19.14 ± 7.77	19.61 ± 8.16	19.22 ± 7.84		18.65 ± 7.32		17.59 ± 7.32
Median(P25-P50)	17(14-22)		18(14-23)	19(14-23)	18(14-23)		17(14-22)		17(13-21)
Interval from CDC diagnosis(days)
To recovery, n	1432		700	139	839		545		48
Means ± SD	15.70 ± 7.30		15.82 ± 7.53	16.53 ± 7.76	15.94 ± 7.57		15.39 ± 6.82		15.10 ± 7.70
Median(P25-P50)	14(11-19)		14(11-19)	15(11-20)	14(11-19)		14(11-18)		13(10-19)

#no./no.of patients with clear contact history (%).

aP < 0.05, Chi-square test; bP < 0.05, Student t-test; cP < 0.05, Wilcoxon rank sum test. These three methods were used to compare the difference between imported and local case and the difference between imported from Hubei province and from other provinces outside Hubei.

Contact history of confirmed cases

The top half of Table 2 shows the contact history of confirmed cases, dividing into contact with confirmed cases of family members and non-family members. The overall median number of confirmed cases the patients who have been contacted were 2 (interquartile range, 1 to 3) in this epidemic. Among them, the patient had contact with 1.80 confirmed cases of family members and 0.58 confirmed cases non-family members on average. Imported cases had contacted more confirmed cases than local cases (2.80 ± 2.33 vs. 2.17 ± 2.10). Furthermore, imported cases had contacted with confirmed cases of family members more than local cases (2.26 ± 2.18 vs. 1.57 ± 2.06), while those had contacted with confirmed cases of non-family equal to local cases (0.53 ± 1.32 vs. 0.60 ± 1.00). In addition, imported cases from Hubei had contacted more confirmed cases than imported cases from other provinces (2.86 ± 2.26 vs. 2.66 ± 2.51), especially in contracted with confirmed cases of family members.

Factors associated with time interval from symptom onset to hospitalization

The left panel of Table 3 shows the results of the first model for the influence factors of TOH. A longer TOH was observed in older and provincial capital cases. The older the case is, the longer the TOH. As compared with the cases younger than 20, especially for cases older than 60 years old, the TOH increased by approximately twice times (OR = 1.87; 95% CI: 1.63, 2.13). Patients in the provincial capital had a 1.08-fold longer TOH than patients in other cities. Moreover, significant risk factors for longer TOH also were identified in the middle and later periods of the epidemic (OR = 1.40; 95% CI: 1.21,1.61, and OR = 1.88; 95% CI: 1.50,2.34, respectively). While, patients with fever had a shorter TOH than those without fever (OR = 0.84; 95% CI: 0.79,0.89). Besides, patients clearly contacted with family-confirmed case shortened this time significantly (OR = 0.89; 95% CI: 0.85,0.93). Furthermore, patients who lived in regions with lower migration intensity with Hubei province had shorter TOH. Particularly, as for patients living in regions where had the migration intensity more than 0.15%, migration intensity (1) between 0.05% and 0.15%, had down to 0.87 times decreased risk of longer time, (2) between 0.03% and 0.05%, had down to 0.74 times, (3) less than 0.03%, had down to 0.69 times. In addition, there is no significant differences in TOH between imported and local cases.

Table 3

Factors associated with time interval from symptom onset to hospitalization and length of hospitalization.

Characteristics	Interval from symptom onset to hospitalization								Length of hospital stay
	Description				Poisson				Description				Weibull
	n	Mean (SD)	Median (P25-P50)		OR*	95% CI*	P		n	Mean (SD)	Median (P25-P50)		HR*	95% CI*	P
Intercept					2.50	(2.04,3.07)	<.0001						19.13	(16.08,22.77)	<.0001
Gender
Male	2430	2.84(3.53)	2(0-4)		Reference				654	19.33(7.88)	18(14-23)		Reference
Female	2134	2.95(3.70)	2(0-5)		0.99	(0.95,1.04)	0.7555		542	18.62(7.33)	17(14-22)		0.96	(0.93,1.00)	0.0432
Age group
≤19	180	1.59(2.77)	0(0-2)		Reference				64	17.72(7.45)	16(12-21.5)		Reference
20-29	562	2.33(3.17)	1(0-4)		1.29	(1.12,1.49)	0.0004		194	17.27(6.5)	16(13-20)		0.97	(0.88,1.07)	0.5505
30-39	972	2.62(3.43)	1(0-4)		1.51	(1.32,1.72)	<.0001		260	18.37(7.62)	17(13-21)		1.00	(0.92,1.10)	0.9209
40-49	1027	3.04(3.56)	2(0-5)		1.65	(1.44,1.88)	<.0001		252	20.33(7.96)	19(14-24.5)		1.06	(0.97,1.16)	0.2340
50-59	972	3.39(3.88)	2(0-5)		1.76	(1.54,2.01)	<.0001		234	19.06(7.29)	18(14-23)		1.04	(0.95,1.13)	0.4516
≥60	825	3.14(3.89)	2(0-5)		1.87	(1.63,2.13)	<.0001		185	20.31(8.4)	18(14-24)		1.08	(0.99,1.18)	0.0995
City at the time of diagnosis
Other city	3745	2.89(3.63)	2(0-5)		Reference				1033	19.01(7.44)	17(14-22)		Reference
Provincial capital	831	2.93(3.55)	2(0-5)		1.08	(1.02,1.13)	0.0081		164	18.91(8.82)	17(13-24)		0.98	(0.93,1.03)	0.4337
Fever
No	982	3.12(3.88)	2(0-5)		Reference				221	19.79(7.86)	18(14-25)		Reference
Yes	3588	2.83(3.53)	2(0-5)		0.84	(0.79,0.89)	<.0001		976	18.82(7.59)	17(14-22)		1.00	(0.96,1.05)	0.9438
Family-confirmed patient contact history#
unclear	3312	2.95(3.60)	2(0-5)		Reference				798	18.93(7.65)	17(14-23)		Reference
clear	1264	2.74(3.64)	1(0-4)		0.89	(0.85,0.93)	<.0001		399	19.14(7.63)	18(14-22)		1.05	(1.01,1.09)	0.0128
Non-family-confirmed patient contact history#
unclear	3962	2.87(3.55)	2(0-5)		Reference				1057	19.04(7.62)	17(14-22)		Reference
clear	614	3.03(4.02)	1.5(0-5)		1.08	(1.02,1.14)	0.0119		140	18.67(7.82)	17(14-23)		0.98	(0.94,1.04)	0.5539
Case source type
Import Patients	2299	2.71(3.45)	1(0-4)		Reference				734	19.22(7.84)	18(14-23)		Reference
Local Patients	2277	3.08(3.77)	2(0-5)		0.96	(0.92,1.01)	0.1129		463	18.65(7.32)	17(14-22)		0.95	(0.91,0.99)	0.0134
Date of diagnosis
21-Jan to 23-Jan	100	2.82(3.38)	2(0-5)		Reference				33	17.09(7.91)	16(12-18)		Reference
24-Jan to 30-Jan	1371	2.51(3.12)	1(0-4)		0.90	(0.77,1.04)	0.1505		384	19.55(7.95)	18(14-23)		1.07	(0.93,1.23)	0.3498
31-Jan to 6-Feb	1873	2.78(3.31)	2(0-5)		1.05	(0.90,1.21)	0.5508		478	19.39(7.86)	18(14-23)		1.00	(0.87,1.14)	0.9493
7-Feb to 13-Feb	985	3.37(4.21)	2(0-5)		1.26	(1.08,1.46)	0.0029		245	18.45(6.65)	17(14-22)		0.87	(0.76,1.00)	0.0550
14-Feb to 02-Mar	447	3.27(4.16)	2(0-5)		1.50	(1.27,1.76)	<.0001		57	15.46(6.38)	14(11-17)		0.66	(0.57,0.77)	<.0001
Migration identity with Hubei provinces (%)
≥0.15	869	3.58(3.86)	3(0-6)		Reference				449	17.55(5.78)	17(14-20)		Reference
0.05-0.15	775	3.10(3.15)	2(1-5)		0.88	(0.83,0.93)	<.0001		23	30.00(10.18)	32(25-35)		1.00	(0.95,1.07)	0.8995
0.03-0.05	1890	2.77(3.64)	1(0-5)		0.58	(0.54,0.62)	<.0001		383	20.92(9.03)	19(15-26)		1.17	(1.11,1.22)	<.0001
<0.03	1042	2.40(3.59)	1(0-4)		0.66	(0.62,0.70)	<.0001		342	18.01(6.85)	17(13-22)		0.99	(0.94,1.04)	0.6426
Interval from symptom onset to hospitalization(days)
≤1	—	—	—		—	—	—		470	20.39(8.45)	19(15-24)		Reference
2-3	—	—	—		—	—	—		173	18.11(6.71)	17(14-21)		0.94	(0.89,0.99)	0.0138
≥4	—	—	—		—	—	—		310	18.09(7.36)	16(13-22)		0.95	(0.92,0.99)	0.0225

#We can trace the history of contact with a confirmed patient.

*OR denotes Odds Ratio, HR denotes Hazard Ratio, CI denotes confidence interval.

Factors associated with length of hospital stay

The right panel of Table 3 gives the HR estimates of related factors associated with LOS. There were no significant differences in LOS among different gender or age groups. It also showed that differences in LOS relative to city type and fever symptoms were not statistically significant. While, patients clearly contacted with family-confirmed case had a longer LOS (HR = 1.05; 95% CI: 1.01,1.09) than patients who did not clearly contact. Moreover, we found that local patients had a shorter hospital stay than imported cases (HR = 0.95; 95% CI: 0.91,0.99). Furthermore, patients reported in the later periods of this epidemic had a shorter hospital stay than patients in the initial epidemic (HR = 0.66; 95% CI: 0.57,0.77). Compared with patients whose TOH was less than or equal to one day, LOS of patients whose TOH was more than 4 days was reduced by 0.05 percentage. And the similar result appeared in patients whose TOH was 2-3 days (HR = 0.94; 95% CI: 0.89,0.99).

Discussion

Comprehensive epidemiological characteristics of the COVID-19 covering the entire periods of epidemic and summaries of the experience from China are useful in public health control. In this study, we described the epidemiological characteristics of imported and local cases, including temporal and spatial characteristics. Indeed, regions with greater migration intensity with Hubei had more imported cases. After the lockdown measures taken by cities in Hubei since January 23 towards the interruption of sustained COVID-19 transmission outside Hubei Province (Nie et al., 2020). We found the daily reported imported cases reached a peak on January 28 and gradually decreased since then. These suggest that traffic restrictions or lockdown in the epicenter can effectively reduce the export of cases (Islam et al., 2020, Zhang et al., 2020a). Moreover, outside of the epicenter, it is also obvious that timely restriction and quarantine of suspicious imported individuals with a travel history of epicenter can effectively reduce the transmission by imported cases in local (Cui et al., 2020, Kwok et al., 2020, Lai et al., 2020a). Even in the provinces that were not in close contact with Hubei, the surveillance of imported cases could not still be overlooked. Taking Tianjin, Ningxia and Hebei province as examples, local cases were twice as large as imported cases, which was related to the several local gathering events of imported cases (Chen et al., 2020, Dong et al., 2020, Zhang et al., 2020b).

This study confirms previously described characteristics (Liang et al., 2020, Wu and McGoogan, 2020), but also highlights the difference between imported and local cases. Throughout this epidemic, imported patients focused on younger, had a higher proportion of male and had more provincial capital residents compared to local cases. This may match the situation that labor exports are mainly the young and middle-aged male in China. This result also insinuates older women living in non-provincial capital cities were at greater risk of exposure when the epidemic spreads to the local. A study on household transmission also founded similar results (Xu et al., 2020). Moreover, the proportion of clearly confirmed case contact history in local cases was higher than that in imported cases. This may be due to the complicated epidemic chain in Hubei Province in the initial phase of the epidemic, which made it difficult to track the contact history of imported cases. Nonetheless, approximately 40% of local cases may be attributed to the household transmission. Among the patients who were clearly exposed to confirmed cases, imported cases had more contacts with other confirmed cases than local cases on average, and contacts were mainly family members. Although we are unable to determine the infectious relationship between them, it might partly explain household transmission caused by imported cases was more prominent. This suggests that after NPIs such as restricting population movement were taken. More effective interventions were still needed to be taken to control household transmission simultaneously, especially for the infection of imported cases to elderly woman in non-provincial capital cities. Indeed, the Chinese government encouraged people to stay at home as much as possible (Lai et al., 2020b). While, the cases that have migrated out from Hubei before January 23 still have the risk of household transmission in local. Therefore, emergency measures were taken by local governments across China to strengthen the tracking and isolation of recent travelers from Hubei (China National Health Commission of the People’s Republic of China, 2020b, China The State Council of the People’s Republic of China, 2020), which reduced this risk to a certain extent. Moreover, our study showed that the daily local cases reached a peak on the 14th day (February 6) after the lockdown, and then gradually declined. This also illustrates the early response of the government is very important for containing the local spread of imported cases.

Our findings show that there was a lag of 2.67 days from symptom onset to hospital admission, and the average length of hospital stay was about 19 days, which were similar to previous studies conducted in China (Khalili et al., 2020, Liang et al., 2020, Linton et al., 2020). Surprisingly, we found that the older the patients are, the longer the hospitalization delays. Considering the situation that medical resources outside Hubei Province had not reached saturation, this might be related to the hospital admission pattern of viral respiratory diseases or the lack of recognition of the disease in elderly patients (Petrilli et al., 2020). Besides, the TOH at the later phase of the epidemic showed a rebound trend. Cases reported in the later phase of the epidemic had a slack attitude in seeking medical resources and the decline in control efforts were possible reasons. However, research in China (outside Hubei province) during January 21 to February 17 demonstrated a shorter hospital admission delay from January 28 to February 17 (4.4 vs. 2.6 days) (Zhang et al., 2020a). Before adjusting for other factors, our research also showed a slightly shorter hospital admission delays in the week after January 23. Except for the different study population and period, we consider this result may be affected by the confounder. Our research included the later phase of the epidemic and adjusted other demographic factors.

This study also confirms that patients living in provincial capital that closely connected to the epicenter had a longer TOH. This provides new demands on the epidemic prevention and control, that is, in provincial capital cities close to the epicenter, case tracking, surveillance and education of immediate admission/isolation should be emphasized. A mathematical model study showed that if the mean time from symptom onset to hospitalization can be halved by surveillance, then the probability that a case leads to transmission is very low (Thompson, 2020). Interestingly, we found associations of clear family-confirmed patient contact history with early hospital admission. This finding makes up for the above result, namely, household transmission caused by imported cases still needs attention. Although we cannot cut off the transmission of the virus among family members by restricting population movement or even lockdown city, it may effectively compensate by reducing the TOH for patients with family-confirmed patient contact history. This reduction may be related to the early detection and isolation of imported cases and their possible close contacts, especially family members (China National Health Commission of the People’s Republic of China, 2020b). In addition, our results also found that the average LOS of 19 days will not decrease by early admission. Perhaps it is related to the characteristics of the viral infectious disease. By contrast, the decrease in LOS in the later phase of the epidemic may be due to the continuous improvement of medical technology for this disease.

This study included a large study cases during an entire epidemic and used a novel methodology. However, there are some limitations. First, as a retrospective study, since the date of symptom onset is self-reported based, there may be recall bias. Second, although we made an effort to collect patient discharge information, we still could not obtain the discharge data of some patients. Fortunately, nearly 90% of patients were discharged from the hospital at the end-point of observation on March 2, which provides an opportunity for the statistical methodology using survival data with left censoring. Third, given the proportion of death cases in the study population was particularly small, which is less than 1%, the impact of death truncation was not considered when analyzing the length of hospitalization. Finally, our study did not include the southeast provinces, but Henan and Zhejiang province were similar to those provinces in intensity of migration and scale of epidemic, and our results are also consistent with several studies conducted in Shenzhen and Hong Kong in epidemiological characteristics during the same period (Bi et al., 2020, Lai et al., 2020a).

In conclusion, through retrospective analysis of the epidemiological characteristics of each phase of this epidemic, this study confirms the effectiveness of policies and provides a reference for the parameters of mathematical modeling. At the beginning of the epidemic, Measures to restrict traffic and even lockdown the city could effectively reduce imported spread. However, household transmission is not yet controlled, particularly for the infection of imported cases to elderly women living in non-provincial capital cities. Fortunately, our results identified patients who clearly contacted with a confirmed case of family members were admitted to the hospital earlier during the entire epidemic. Even so, surveillance and patients’ education about early admission or isolation should still be attached great importance in the future prevention and control, especially for the elderly living in provincial capital cities that were more closely connected with the epicenter.

Authors’ contribution

Feng Zhou: data collection, data analysis, writing. Chong You: data collection, writing. Xiaoyu Zhang: data analysis. Kaihuan Qian: data collection. Yan Hou: data collection. Yanhui Gao: data analysis. Xiao-Hua Zhou: study design.

Not required. Ethical approval

The study was anonymous, and individual information was collected from provincial or municipal health commissions, which is a public data to help control this epidemic.

Conflicts of interest

No potential conflict of interest exits in the submission of this manuscript, and manuscript is approved by all authors for publication.