Seroprevalence, spatial distribution, and social determinants of SARS-CoV-2 in three urban centers of Chile

After emerging in Asia, the first wave of COVID-19 took place in the southern hemisphere between March and October 2020 [1]. The first Chilean COVID-19 case was identified on March 3, 2020, becoming by mid-June 2020 one the world’s most per capita affected countries, centered mainly in the capital, Santiago, which accounts for 40% of the country's population [2, 3]. Due to Chilean geography, climate heterogeneity, and other characteristics, each province has presented its own epidemic curve [3].

The SARS-CoV-2 seroprevalence outcomes vary substantially from place to place. Aside from the timeframe, the disease incidence not only depends on the public health response in each country, but also on spatial distribution, population density, and physical layouts of cities where the pathogen is introduced [4].

It has been well established that social, economic, and cultural forces shape the spread and impact of infectious diseases in human populations [5]. Recent studies provided evidence on the relationship between social status and exposure patterns, the likelihood of diagnosis, and infection outcomes for COVID-19 in Santiago. But these studies have an ecological design and are based on administrative registries of morbidity or mortality, and have ignored the effect of social determinants in other urban centers of Chile [6‐8].

Epidemiological surveillance systems for SARS-CoV-2 epidemic have focused, primarily on registries of symptomatic and severe cases and deaths, based on molecular testing capacities using polymerase chain reaction (Rt-PCR). Seroprevalence studies, using antibodies detection, have identified the presence of mild and asymptomatic SARS-CoV-2 infections not registered by surveillance systems relying on symptoms [9‐13]. Thus, seroprevalence studies identify the infection burden in communities leading to more appropriate control measures for local epidemic expansion. They provide information quantifying the susceptibility and the immunity status of the participants, inferring the dynamics epidemic infection in a population. At the same time, these studies may highlight risk factors for infection, severity, and lethality [9, 14, 15] .

The seroprevalence during the first wave in Europe ranged from 0.42% among residential clinical samples in Greece, to 23% in Lodi (blood donors), and 23.6% in Castiglione D'Adda (population based), both located in highly affected places in north of Italy [10, 16, 17]. Latin America was severely affected, with the world’s highest per capita death toll rate in Perú and the third largest toll of cases in Brazil [18], evidencing disparity in pandemic consequences in a deeply unequal sub-continent [19]. Nevertheless, few studies report the seroprevalence in Latin America with a range from 8.3% to 44% in diverse localities [9, 20‐22].

Until this study, there have been no reports of population-based seroprevalence studies in Chile. This article reports a population-based seroprevalence study conducted in three urban centers located in the center-north, metropolitan, and center-south regions of Chile: Coquimbo-La Serena conurbation, Santiago (capital city), and Talca. Using geospatial modelling, the distribution of SARS-CoV-2 infection risk was identified, based on individual and ecological data, in these three locations.

We enrolled 1155 dwellings; refusal to participate was 25.6% at the first contact and less than 1% among residents of the recruited dwellings (Fig. 1). A total of 2493 subjects were enrolled, between 7 and 94 years old (average 41.6 years); 60.3% female; 1441 were residents of Santiago (57.8%); 478 from Coquimbo-La Serena (19.2%) and 574 from Talca (23.0%) (Additional file 1, page 3: Table S1). The ELISA test was performed in 95.2% of the participants; this proportion was lower in Talca (90.1%). All point-of-care tests were negative (n = 120).

Between September 26 and November 25, 2020, the population seroprevalence was significantly higher in Santiago (11.0%, 95% CI 8.2–14.7), followed by Coquimbo-La Serena (5.6%, 95% CI 3.3–9.5) and lower in Talca 2.0% (95% CI 0.8–4.7). Figure 2 shows the geographical distribution of study sites and antibodies results.

The overall prevalence was 10.4% (95% CI 7.8–13.7), representing 734,220 individuals who have been infected with SARS-CoV-2 (Table 1). We found no differences in seroprevalence according to sex, age, or ethnicity. However, Coquimbo-La Serena indicated a significant increase with age up to 59 years (p = 0.0263) (Additional file 1, page 5:Table S2). Regarding nationality, foreigners had higher seroprevalence, reaching statistical significance only in Talca (p = 0.0253).

In terms of socio-economic risk factors (Table 1), people with public health insurance showed a higher seroprevalence, both in Santiago and in the pooled results (OR = 2.5, 95% CI 1.3–5.1). Likewise, a significant association with educational level was observed: people with university education had a lower seroprevalence than those who completed secondary education (OR = 2.3, 95% CI 1.2–4. 5). However, no association was found between seroprevalence and other social-related variables at the individual level, such as overcrowding, number of residents, or other housing conditions. Neither was an association found between the type of occupation or face-to-face work with the presence of antibodies.

Most seropositive cases were symptomatic (153/242; 63. 2%), and presence of symptoms was strongly associated with seropositivity (OR = 5.6, 95% CI 3.3–9.4) (Table 2 and Additional file 1, page 6: Table S3). Likewise, the number of symptoms positively associated with seroprevalence (Mantel–Haenszel = 296; p < 0.0001). Only 38.7% (252/651) of people with symptoms sought medical care. Asymptomatic infection was more frequent in children below 15 years old (p = 0.0304).

The most frequent symptoms in seropositive subjects were: headache, myalgia, and odynophagia. In contrast, the greatest strength of association with antibodies presence was for anosmia and dysgeusia (OR = 30.0, 95% CI 12.1–74.6 and OR = 32.9, 95% CI 11.6–93.9 respectively) (Table 3 and Additional file 1, page 7: Table S4). No association was observed between seroprevalence and self-reported chronic conditions, except for tobacco consumption, which appears to have a negative association with infection (OR = 0.5, 95% CI 0.3–0.9) (Additional file 1, pages 8-9: Tables S5 and S6).

Previous COVID-19 diagnosis was reported by 120 subjects, with 81.7% seropositivity. This antecedent was strongly associated with seropositivity (overall OR = 32.0, 95% CI 10.7–96.3). The hospitalization rate for self-reported COVID-19 subjects was 10% (Table 2). Seroprevalence was also significantly higher in those who reported contact with confirmed cases (31.2% OR = 5.7 95% CI 3.0–10.9). The most frequent contact place was the home, with higher seroprevalence than other contact places (42.1%, OR = 3.9 95% CI 1.1–13.3). More than half of those with close contact confirmed cases were quarantined (173/296; 58.4%); this proportion was higher in Coquimbo-La Serena (63.3%). Coquimbo-La Serena was the only city where quarantine history was associated with seropositivity (p < 0.0001) (Table 2 and Additional file 1, page 6:Table S3). Two COVID-19 related deaths occurred among the residents of the sampled dwellings, both males over 60 years residents of Santiago.

Table 4 shows the estimation of the infection fatality ratio (IFR) for the three cities, the average risk of death for SARS-CoV-2 infected people was 1.3% (95% CI 1.27–1.32) for PCR confirmed deaths and increased to 1.67% (95% CI 1.64–1.70) when considering all COVID-19 attributable deaths.

The proportion of dwellings with at least one seropositive resident reached 12.8% and was higher in Santiago (Table 5). In those dwellings, the average of infected residents was 61.6% (95% CI 56.6–66.4), and the secondary attack rate reached 38.4% (95% CI 32.2–44.8%).

Table 6 shows the bivariate analysis between the presence of SARS-CoV-2 antibodies and ecological variables. Having antibodies against SARS-CoV-2 is significantly associated with high population density (p = 0.004) and with a higher proportion of individuals without university education (p = 0.005).

Multivariate model shows that at the individual level, the risk factors of SARS-CoV-2 infection are contact with confirmed cases and foreign nationality. At the ecological level, a positive association exists between seropositivity, population density, and the proportion of individuals with university education. When comparing the three cities, this model shows significant differences in the risk of SARS-CoV-2 infection only between Santiago and Talca. The selected multivariate logistic model reached a concordance index of 0.80. Final results are provided in Table 7.

The spatial EBK interpolation model using the multivariate prediction (education and population density at the census zone) together with seroprevalence results was representative for the three cities (root mean square standardized error > 0.9). The range of predicted probabilities of SARS-CoV-2 infection were 0.008–0.756, 0.006–0.497, and 0.002–0.265 in Santiago, Coquimbo-La Serena, and Talca, respectively.

In Santiago (Fig. 3), the territorial distribution of the estimated model by census zone showed that in a large part of the city the predicted risk of having antibodies against SARS-CoV-2 varies between 10–15%. In the north-west and south-east areas, it was higher than 15%, while, areas with a lower risk (< 2.5%) were observed in the north-east zone. The conurbation of Coquimbo-La Serena (Fig. 4) showed a similar heterogeneity to Santiago. The highest risk (10–15%) was found in the northern part of La Serena and on a peninsula in southern Coquimbo.,. In contrast, lower risk (< 2.5%) corresponds to the central area of La Serena and the coast urbanized area. The city of Talca (Fig. 5) showed less heterogeneity in the model results and in distribution of analyzed ecological variables.

The correlation analysis showed a significant association between the predicted risk of infection (EBK model) and the MOH reported cumulative incidence rate in Santiago (r = 0.449) and in Coquimbo-La Serena (r = 0.256). While in Talca there was no significant association (r = 0.06) (Fig. 6).

Our study shows the prevalence of SARS-CoV-2 antibodies in Chile at approximately 10% 267 days after the first imported case was detected, varying from 2 to 11% in three different cities, a figure nearly three times higher than the incidence rate reported by the Chilean Ministry of Health (MOH).

This heterogeneous prevalence inside Chile is consistent with other seroprevalence studies; for instance, Spain showed 5% (95% CI 4.7–5.4) with substantial geographic variation, ranging from more than 10% in Madrid to less than 3% in some coastal areas [30]. In Brazil, seroprevalence measured in June 2020 varies among cities from 0% to 25.4% [31].

Our study found similar antibody prevalence between sexes, slightly higher in males but not statistically significant. Comparable results between sexes were reported in several population-based studies [4, 11]. Also, we found no significant differences between age groups. Interestingly, children showed the same seroprevalence as other age groups despite national school closure ten days after the first case was detected [2]. A possible explanation for the high prevalence in children is that, with a relatively high level of population contagion, despite the suspension of school activities, minors were infected within homes, which is consistent with the high rate of attack within the homes.

On the other hand, the prevalence of older people is lower, showing some isolation from people of this age group. A similar phenomenon in children was also observed in Iquitos, Peru [20]. The reported OR for children infection aged < 18 in studies of household contacts was 0.77 (95% CI 0.27–2.17) in Italy and 1.39 (95% CI 0.55–3.53) in Utah, and Wisconsin in the USA [32, 33]. The infection rate in children is still debated around the globe.

In our study, a high proportion (63.9%) of people who developed antibodies reported symptoms. Nevertheless, a third remained asymptomatic, which was significantly higher in people below 15 years old. In Lima, in the first epidemic wave, 56.1% cases were asymptomatic and in Spain 32.7%, which may reflect the variation in expression of COVID-19 worldwide [22, 30].

Nearly 40% of symptomatic cases sought medical care, which may explain that many remained unreported, resulting in undetected ongoing transmission. This behavior could be related to health care access barriers, including cultural barriers associated with interpreting medical signs or symptoms, delaying consulting, and worsening disease outcomes.

Smoking showed a negative association with SARS-CoV-2 infection. This paradoxical effect has been associated with an underrepresentation of active smokers [34]. In that sense, our study detected that people with previous COVID-19 diagnoses reported significantly lower current tobacco consumption than those without the disease. However, the protective effect of tobacco use has also been evidenced in European countries, reporting a significant negative association between smoking prevalence and the prevalence of COVID-19 across the 38 European nations after controlling for confounding factors (p = 0.001) [35]. Some authors propose that nicotine prevents infection by competing with the virus with the ACE2 receptor. In this sense, epidemiological and experimental evidence has been found [36, 37].

In sum, we found that 10.4% of people were infected, almost 5% self-reported COVID-19 diagnosis (120/2493), and 0.4% (11/2493) were hospitalized; practically one in ten self-reported confirmed cases. As expected, the main risk factor was contact with confirmed cases. However, only 60% were quarantined, evidencing failures in contact tracing.

We couldn't find evidence of the association between seroprevalence and overcrowding at the individual level. However, we did find an association at the ecological level between seroprevalence and residential density (people/dwelling). Moreover, the sampling approach enabled us to contact more than one household member, thus facilitating the study of transmission within households. The secondary attack rate (SAR) was 38.4%. A systematic review (43 articles) estimated a pooled SAR of 18.1% (95% CI 15.7–20.6) ranging from 3.9% to 549% [38]. This high SAR is consistent with our finding of higher seroprevalence among people referring contact with confirmed cases at home, compared to contact in other places. Therefore, the critical role of household transmission needs to be considered when confinement measures are implemented, such as facilities to isolate cases away from the household.

The IFR in the three cities ranged from 0.59% to 2.58% for COVID-19 confirmed deaths, with an overall ratio of 1.3%. Talca exhibited a higher lethality related to reduced found seropositivity, giving an unreliable figure. Santiago doubled the Coquimbo-La Serena estimate (IFR 1.32 vs. 0.59), coincident with larger outbreak affecting the capital during 2020. In Santiago, we observed an abrupt rise in mortality and substantial stress on hospital capacity, unlike situations in the other two cities until 2021 [3]. Ioannidis estimated IFR based on seroprevalence data of 61 studies ranging from 0.0 to 1.63%, in the range of our pooled results [39].

Spatial analysis demonstrated the strong association between risk of infection and social determinant distribution, e.g., education and population density, in the territory, which was clearly observed in Santiago and Coquimbo-La Serena (larger and more socially segregated than Talca), with crowded and low-income neighborhoods where the virus transmits more easily [5]. In the case of Santiago, the interpolation shows that census zones with higher levels of socio-educational vulnerability show a seroprevalence six times higher than the wealthiest neighborhoods. Coquimbo-La Serena has four to six times more seroprevalence in sectors with greater socio-educational vulnerability than higher socio-economic levels.

We found a significant correlation between our spatial model of infection risk and the official registered cases by census zone in two of the three cities. Moreover, the residuals (zones with higher deviation) in Santiago pointed out areas that Mena suggested have less access to COVID-19 diagnostic tests [8]. So, the model was able to identify higher infection risk areas within two of the three cities.

The heterogeneity of infection risk distribution is consistent with the structural social inequalities that characterize Chile. Urban segregation is one of the most evident effects, a common phenomenon in larger cities throughout Latin America [40, 41]. Also, at the individual level, higher seroprevalence was associated with lower education and public health insurance (a proxy for middle and low-income levels in Chile). This social disparity was also reported in Lima, where the less advantageous tripled seroprevalence of the wealthier [22].

Thus, both at the individual level and in the spatial analysis, our results provide evidence of social determinants of COVID-19, and we note similarities with other respiratory transmitted diseases considering five levels of analysis [42]. First, social stratification, which affects the probability of contact with a contagious case. Second, the physical and social environment determining the exposure to the virus is inversely related to social position; overcrowding, job insecurity, dwelling conditions, and mobility are related to transmission. Third, vulnerability to infection and serious illness also have social and cultural modulation related to distribution of comorbidities. Fourth, unequal access to diagnostic testing and, late, to therapeutic resources, affect transmission and serious infection. The fifth level of analysis is the difference in consequences, not only of fatalities but also of social impact of deaths, catastrophic expending, loss of future income due to persistent sequelae, which we have not explored in our study.

Our primary limitation was the low number of cases in Talca, perhaps due to the lack of statistical power or the use of a higher proportion of rapid tests (9.9%) with a lower sensibility. Other limitations of our study are potential bias: first, recall bias of self-reported symptoms and diagnosis; second, the natural decrease of antibodies after infection may have biased the results toward lower prevalence and may explain the finding of almost 20% of seronegative results among those who self-reported COVID-19. And third, survivor bias, as we inquired about deaths of household members due to COVID-19, finding two deaths, both males over 60 years old. Finally, as in any model, EBK has limitations; thus, to validate, we compared distribution of our seroprevalence results with incidence of COVID-19 confirmed cases by census zone reported to MOH, finding a significant correlation.

To our knowledge, this is the first population-based SARS-CoV-2 seroprevalence study carried out in three cities of Chile. Strengths of this study include the high response rate and the interdisciplinary approach incorporating geographic modeling. Our results enabled us to estimate the real burden of SARS-CoV-2 infection in three cities and identify the spatial distribution of disease risk and areas with relative lack of access to diagnostic tests. All this data and results can help guide the design of strategies and prioritize intervention areas in urban contexts, not only for SARS-CoV-2 but for other infectious diseases as well.

Our study found the proportion of people with symptomatic infection, those seeking medical care, hospitalizations among confirmed cases, and the infection fatality ratio. Additionally, we evaluated quarantine coverage and as primary and secondary attack rate in households. As expected, antibody presence was higher in people with previous diagnoses of COVID-19, in people in contact with confirmed cases, and in those with COVID-19 compatible symptoms. The most significant infection risk was household contact with confirmed cases. Seroprevalence was significantly associated with lower level of education, public health insurance, and population density.

Our results reveal highly heterogeneous spread of COVID-19 in urban contexts and significant association between seroprevalence and social determinants. The spatial model enabled us to identify high transmission risk areas in different Chilean urban centers, those not detected by surveillance and health care systems and relatively neglected.

In sum, at the end of 2020, besides being very far from any herd immunity level, the seroprevalence shows that the first wave in Chile was remarkable. This is consistent with the severe stress observed throughout the health care system and the excess of mortality, evidencing the impact of COVID-19 on naive populations. In such situations, government authorities must promote interdisciplinary analysis and intersectoral interventions to reduce vulnerability for future waves, especially in highly populated urban contexts.