Introduction

In December 2019, a series of unknown-cause pneumonia cases, later named COVID-19 by the World Health Organization (WHO), were reported in Wuhan, China, and has since spread rapidly, becoming a global pandemic within months. As of July 9th, 2021, COVID-19 has resulted in 185 291 530 confirmed cases and 4 010 834 deaths worldwide1.

Since the COVID-19 outbreak, there are a number of studies that look at the spread of the virus from an epidemiological perspective2,3,4,5,6,7,8,9,10,11,12,13, but there is minimal evidence on how individuals react to the pandemic in terms of COVID-19-related health-seeking behavior and diagnosis capacity14. However, like all other epidemics, the spread of COVID-19 is not only affected by its biological characteristics, such as the incubation period and transition rate but it is also influenced by social and behavioral factors15. Individual protective measures such as social distancing, escalation of care-seeking, and evolution of test and diagnosis protocol and capacity are all factors likely to determine the speed at which the virus spreads16,17,18,19,20. It is, therefore, crucial to understand how COVID-19-related health-seeking and diagnosis vary by socio-demographic factors to understand the progress of the pandemic better, evaluate health equity across groups, and implement effective public policies21,22.

In this study, we use time-to-event models to identify several key factors associated with delays of health-seeking and diagnosis in response to all COVID-19 infections in the province of Shaanxi, China, since the start of the pandemic. We measure the delay of health-seeking as the elapsed time from symptoms onset to the first doctor visit and the delay of diagnosis as the elapsed time from the first doctor visit to the diagnosis of COVID-19 infection. Key factors that we include in our analysis, such as sex, age, prefecture of residence, travel history to Wuhan, days since the first case emerged in the local community, and medical resources, are likely to be associated with both COVID-19-related health-seeking and diagnosis.

This study contributes to the existing literature in several ways. First, in addition to demographic characteristics commonly addressed in epidemiological studies23,24, we explore the importance of the roles played by a series of social factors in explaining the differential access to healthcare for COVID-19 patients. Second, we separately identify the likely determinants of delayed health-seeking and diagnosis, which—as our analysis suggests—are driven by different factors and should, therefore, be targeted differently.

Method

Data

On January 23rd, 2020, three cases, two men and one woman, 42, 22, and 32 years old, respectively, were confirmed and reported to be the first three COVID-19 cases in Shaanxi. Delays from symptom onset to the first medical visit of these cases were one, four, and one day, respectively. Delays from the first medical visit to the diagnosis of COVID-19 were two, three, and two days, respectively. Since then, the epidemic spread rapidly, reaching a maximum of 23 confirmed cases on February 5th and then contracted gradually until no new cases were detected from February 20th onward to the end of the year 2020.

In total, this wave of the COVID-19 pandemic between January 23rd, 2020—the date of the first confirmed case—and February 20th, 2020—the date of the last local case in 2020 resulted in 245 confirmed cases distributed in all ten prefectures with a majority of 119 cases concentrated in ** symptoms (67 cases). We also exclude cases with incomplete information on dates of symptom onset, first medical visit, and diagnosis (9 cases). Additionally, we drop cases younger than 18 years old (4 cases). Applying these criteria generates a working sample of 165 confirmed cases.

Variables

We focus on the elapsed time of two key events concerning health-seeking and diagnosis separately, namely delay from symptom onset to first medical visit and delay from the first medical visit to the diagnosis of COVID-19. The total elapsed time from symptom onset to diagnosis is also examined.

In addition to age and sex, we consider some risk factors including days since the first case emerged in Shaanxi, living or travel history of Wuhan, currently residing in metropolitan areas, the prefecture of residence, prefecture-level medical resources measured as the number of healthcare facilities, health workers and fever clinics per capita as well as the tier of the healthcare facility visited.

After announcing human-to-human transmission of COVID-19 by China CDC and reporting the first local case in Shaanxi Province, the public escalated awareness and vigilance towards the COVID-19 pandemic. Meanwhile, the government began distributing medical resources, including adopting new technology that drastically shortened the diagnosis time. Therefore, the first doctor visit and diagnosis of COVID-19 cases are likely to shorten as the pandemic progresses. We, therefore, include the number of days since the first COVID-19 confirmed case in the province, as suggested in previous studies looking at the effects of shelter-in-place policies on healthcare utilization25.

As COVID-19 is originated from Wuhan, people with a living or travel history to Wuhan were seen with high risk by both the public and surveillance administrators. Individuals with travel history to Wuhan may want to seek or be given medical care in priority relative to other groups of individuals. This observation motivates us to include whether individuals in our sample have traveled to Wuhan since January.

Healthcare services in China are officially classified into three tiers, with the first tier consisting of large hospitals in metropolitan areas, the second tier including intermediary hospitals located in counties and city districts, and the third tier composed of community-based facilities including clinics26. Medical resources are disproportionately concentrated in tier one and tier two hospitals, leaving small facilities under-resourced and understaffed. What is more, the referral system from clinic-based primary care to hospitals is not well-functioning27. As a result, the elapsed time from the first medical visit to the diagnosis of COVID-19 infection may depend on which type of medical facility is used by patients in the first place. Our analysis, therefore, includes the health facility tier to which the medical care used by patients belongs.

To capture unequal access to healthcare across prefectures, we also include prefecture fixed-effects in our benchmark model specification and assess the robustness of our findings by including prefecture-level medical resources such as the per capita number of fever clinics, health facilities, and healthcare workers in an alternative specification.

Time-to-events models

We test a variety of parametric time-to-events models, including Weibull, exponential, Gompertz, log-logistic, log-normal, and generalized gamma distributions, and select the model with the best goodness of fit based on Bayesian Information Criterion (BIC)28. Using the model with the best fit, we explore risk factors associated with delayed time to medical visit and diagnosis from symptom onset and report corresponding hazard ratios and marginal effects of risk factors in median times to event.

Since cases were updated daily with the data granularity based on interval of days, the time interval between events can be interval-censored. For example, patients may see a doctor on the same day of symptom onset. In this case, the time between symptom onset and first doctor visit is interval-censored within one day. We, therefore, also fit our data using alternative parametric models that account for this kind of interval-censoring.

We report associations between risk factors and time-to-event through hazard ratios and average marginal effects of factors to the median time-to-events. We take the average across cases to obtain average marginal effects, holding other factors fixed. Data analysis is conducted using the statistical program Stata (Version 14.1, StataCorp, College Station, Texas).

Results

Descriptive statistics

Table 1 presents the descriptive statistics of our study sample. There were more males than females (60% vs. 40%), and age ranged between 18 and 89 with a mean value of 45.9 years old. 40% of confirmed cases had living or travel history of Wuhan. The mean time of case emergence since the first case in Shaanxi was 16.1 days. After develo** symptoms, among individuals who later tested positive with COVID-19, 40% chose large hospitals (tier 1), 40% chose intermediary hospitals (tier 2), and 20% community-based services (tier 3) for the first doctor visit. Cases were geographically unevenly distributed, with up to 50% cases concentrated in ** with COVID-19 in Shaanxi. Though community-based healthcare and primary care at large is supposed to play a crucial role to ensure an effective COVID-19 response58, community and primary care systems functioned ineffectively in many countries during the height of the pandemic, often because of lack of testing and diagnosis capacities59. In addition to insufficient capacities, China’s primary care system faces many other challenges such as undertrained and understaffed workforce, the fragmentation of clinical care and public health service, and the disintegration between primary care institutions and hospitals60. Lack of diagnostic and testing capacities can potentially hurt the efficiency of primary care services61 and lead to other adverse health outcomes62,63. However, since in China patients are allowed to self-refer to hospitals without paying much more and patients often have a preference for hospitals over primary care institutions, those who do choose community-based healthcare and primary care in the first place are often from lower socioeconomic background, which can be another factor that explains healthcare delay64. Moreover, although the importance of health workforce on health outcomes is well-documented, there is scant evidence on the association between health workforce and diagnosis time65,66. Our findings therefore present some initial and suggestive evidence on the importance of healthcare resources, particularly in terms of healthcare workforce, on the timely diagnosis of COVID-19.

Limitations

We acknowledge several limitations of our study: First, our study takes place in a country with relatively low infection and transmission risks and our results are derived from a small sample size. Our results may therefore not be generalizable to places where risks are higher and strains to health care systems more important.

Second, mild symptoms and asymptomatic cases may be unaware of their conditions and may not seek healthcare. To the extent that only individuals with relatively severe COVID-19 symptoms decided to visit a medical doctor, our analysis is likely to be biased downwards because individuals with only mild symptoms are plausibly more likely to delay doctor visits.

Third, we fail to incorporate in our analysis several important social factors such as education and income due to data availability, though we include the prefecture of residence and whether individuals live in metropolitan areas, which could serve as a proxy for socioeconomic characteristics.

Fourth, our analysis on the geographic variation in time delays for COVID-19 health-seeking behaviors would have benefited from a more formal multilevel analysis. However, given our limited sample size, such analysis could not be performed, and we are limited to consider geographic characteristics as simple fixed effects in multivariate regressions.

Finally, policy recommendations based on our findings may not apply to other contexts and countries. Both the success of disease surveillance and the failure of primary care in dealing with the current COVID-19 crisis is embedded in a more extensive system of social arrangement that is a specific and unique feature of China.

Conclusion

This study presents some suggestive evidence on risk factors associated with a time delay from the onset of symptom to the diagnosis of COVID-19 infection. Based on a relatively small sample in Shaanxi, China (\(N=165\)), we have identified factors like age, surveillance on travel history, types of healthcare facilities visited, size of the health workforce, which are likely to affect either health-seeking or diagnosis or both, contributing to the time delay of proper diagnosis.

Our findings may offer some valuable insights for policymakers. Older adults tend to delay their care, emphasizing the need for closer surveillance and targeted public policies for older adults and vulnerable groups in general. The positive association between the surveillance of high-risk individuals with travel history to Wuhan and a shorter diagnosis delay supports the prioritization of surveillance on groups with higher risks of exposure. The delay caused by the utilization of community-based healthcare has two direct policy implications: First, primary care can only benefit patients if it is empowered and integrated into the extensive healthcare system. Second, policies steering patients towards primary care have to be accompanied by reforms aimed at reinforcing the primary care system.

Our analysis on the factors associated with time delays for COVID-19 diagnosis sheds light on the importance of patients’ health-seeking behaviors and diagnosis capacity of healthcare providers during the pandemic. That being said, the small sample size undermines the external validity of our study and prevents us from a thorough and more conclusive validation of the mechanisms involved. More research is needed to better understand the mechanisms—be they biological, economic, social or a combination thereof– of the age gradient in COVID-19 health seeking behaviors we observe in our data. Moreover, the “social” cost of surveillance should be weighted along with its benefit: the optimal level of surveillance should be based on scientific and factual considerations rather than political preferences and inclinations. Finally, the primary care system should be enhanced in the long run in China. Future research can guide policy makers in striking the right balance between strengthening primary care system and structure and develo** modern hospitals and advanced technologies that can help tackling future public health challenges.