Social vulnerability during the COVID-19 pandemic in Peru

The COVID-19 pandemic has demanded governments and diverse organizations to work on strategies to prepare and help communities. Increasing recognition of the importance of identifying vulnerable populations has raised a demand for better tools. One of these tools is the Social Vulnerability Index (SVI). The SVI was created in 2011 to identify and plan assistance for socially vulnerable populations during hazardous events, by providing disaster management personnel information to target specific areas. We aimed to evaluate and determine the social vulnerability in different provinces and districts of Peru in the context of the COVID-19 pandemic using an adapted version of the SVI index. Ecological, observational, and cross-sectional study was conducted. We adapted the SVI and collected indicators related to COVID-19. We organized and analyzed the population data of the 196 provinces of Peru, using data from government institutions. We found a distribution of high and very high SVI in the mountainous areas of Peru. High and very high social vulnerability indexes, due to the presence of some or all the variables were predominantly distributed in the provinces located in the southern and highlands of the country. The association between mortality rate and social SVI-COVID19 was inverse, the higher the vulnerability, the lower the mortality. Our results identify that the provinces with high and very high vulnerability indexes are mostly located in rural areas nearby the Andes Mountains, not having a direct correlation with COVID-19 mortality.


I. Introduction
The COVID-19 pandemic has demanded governments and diverse organizations to work on strategies to prepare and help communities to face the pandemic. Peru has been one of the most affected countries in the world.
Certain populations could be more vulnerable to COVID-19 and more likely to manifest with more severe disease. This vulnerability reflects limited capacity of certain populations to mitigate, treat, and delay the transmission of pandemic disease, and reduce its clinical, Potential uses of SVI are: help emergency response planners and public health officials identify and map the communities that will most likely need support before, during, and after a hazardous event and to estimate the number of needed supplies, or where to allocate and prioritize material and human resources, also it could help decide how many emergency personnel are required to assist people and to identify communities that will need continued support to recover following an emergency or natural disaster [20].
The COVID-19 pandemic caused a negative impact on the population's health and on their ability to deal with and recover. However, this vulnerability has not yet been characterized enough. Specifically, any measurement of social vulnerability needs to be considered in the specific context of vulnerability to COVID-19 [21].
This would lead to identifying the geographical regions that, due to their socioeconomic characteristics, infrastructure, population, and access to health, may be more vulnerable and require the greatest support and attention from the national or regional government.

Population
People who reside in Peru as surveyed and recorded in the databases of the Peruvian Ministry of Health, the National Institute of Statistics and Informatics, and the National Census of 2017.

Analysis unit
We analyzed provinces of Peru; provinces are intermediate territorial. Provinces can assume the role of basic unit of territorial development planning due to its intermediate nature. There are 196 provinces of Peru that can be classified by strata according to their number of inhabitants, which correlates directly with their human development index (HDI) [22]. The first stratum corresponds to Metropolitan Lima and Callao with more than 10 million inhabitants. The second stratum, with a population from 300 thousand to 1 million inhabitants, corresponds to 17 provinces. The third stratum with populations of 50,000 to 300,000 inhabitants comprises 102 provinces. Finally, stratum four, with populations of less than 50,000 inhabitants, corresponds to 75 provinces. Except for stratum 1, which only contains coastal provinces, all strata have coastal, mountain and jungle provinces.
INEI is the official government agency of Peru in charge of carrying out national censuses, the last of which was in 2017. Fourteen indicators were extracted from this source corresponding to the topics 'socioeconomic status' (05), 'household composition' and disability' (04), 'minority groups and language' (02), as well as 'type of housing and transportation' (03). SUS-ALUD regulates the collection, transfer, dissemination, and exchange of information generated or obtained by IAFAS, IPRESS and IPRESS Management Units, and publishes aggregated indicators of the country's health services supply capacity. The SINADEF is the computer application that allows the entry of data of the deceased, generation of the death certificate and the statistical report; includes general deaths, fetal deaths, and deaths of unidentified persons.

Instrumentation-Social Vulnerability Index against COVID-19 (SVI-COVID19)
We adapted the SVI [14] to the context of COVID-19 by combining indicators related to the social determinants of health, which measure the expected negative impact of any kind of disasters, and two factors to determine vulnerability against COVID 19: risk factors associated with poor prognosis of the disease and access to health services.
A total of 33 variables distributed in six central themes were evaluated. Four themes correspond to the CDC SVI and refer to 1) socioeconomic status, 2) household composition and disability, 3) belonging to a minority group and language, and 4) type of housing and transportation. The COVID-19 specific factors were: 5) epidemiological factors, based on guidelines from the WHO and the Peruvian Ministry of Health, which identify populations at risk of acquiring severe forms of the disease and, therefore, have higher mortality from COVID-19; and 6) health system factors, including data related to the capacity, strength, and readiness of the health system in relation to the needs posed by COVID-19. We also included the number of hospital beds, total health spending and the percentage of regional health budget spending, number of doctors per 100,000, as well as the number of ICU beds and the number of mechanical ventilators available by region.

Estimation of SVI-COVID19
For the construction of the SVI-COVID-19 indicators, the 33 variables were expressed in population percentages or incidents, which were arranged in ascending order per the 196 provinces of Peru. We ranked negative variables (i.e., poverty) from 1 to 196, and positive variables (i.e., income) whose higher percentage value in the population meant less vulnerability from 196 to 1. Subsequently, we calculated the percentile-ranks with the formula: Percentile-rank = (Rank-1) / 195. We then summed up the indicators by topics of the SVI-COVID19 and performed the calculation of ranks and percentile ranks by topic. Finally, the index was classified into quintiles, obtaining the following groups: very low vulnerability (<20%), low (20-40%), moderate (40-60%), high (60-80%), and very high (> 80%). We then calculated the death rate from COVID-19 per 100,000 inhabitants for each of the provinces of Peru between April 2020 to September 2021. This period includes both two more lethal COVID-19 waves during the pandemic in Peru.
Death from COVID-19 was determined under the standard established by SINADEF, which includes cases confirmed by a positive molecular, antigenic, or serological laboratory test, as well as the diagnoses indicated in the medical certificate.

Statistical analysis
We performed descriptive statistics by calculating the means and medians as appropriate for each SVI-COVID19 indicator. We then created a bivariate map to show the vulnerability index for each province and the COVID-19 mortality rate as a dependent variable. Additionally, we performed a multivariable analysis using a negative binomial regression. Incidence rate ratio (IRR) were calculated and adjusted by population density with their respective 95% confidence interval (95% CI). The bivariate map was generated in ArcGIS 10.7 and the statistical analysis was made in the STATA v.17.0 statistical software package.

Ethical considerations
The research ethics committee of the Universidad Peruana Cayetano Heredia approved the study protocol with certificate No. 501-28-20. Since this was an ecological study with a secondary database, no individuals were evaluated and no personal data were accessed, so there were no risks associated with data confidentiality. In this sense, informed consent was not required for our study.

III. Results
We evaluated all 196 provinces of Perú. The descriptive statistics of the 33 indicators by topic are shown in Table 1. Monthly per capita income is one of the indicators with a bigger range of inequality, averaging S/.913 ($234) in the least vulnerable provinces and S/.392 ($ 100) in the most vulnerable provinces. Likewise, the percentage of the indigenous population that do not speak Spanish in the most vulnerable provinces is three times the percentage of the least vulnerable provinces.
In epidemiological variables, we observed that the median of the immunocompromised rate, respiratory disease, and diabetes registers is higher among the less vulnerable population except for the median obesity rate which is higher in the most vulnerable population. Among the health system province indicators: number of doctors, mechanical ventilators, hospitalizations, and ICU beds; the most remarkable difference was between provinces with higher and lower social vulnerability index. The highest death rate from COVID-19 occurred in the provinces with moderate vulnerability with an average of 429 deaths per 100,000 inhabitants (SD: ± 209.69), the lowest rate was found in the highly vulnerable provinces with 254 deaths per 100,000 inhabitants (SD: ± 87.92). The provinces with very high vulnerability have a smaller population with 39,266 inhabitants on average.
In the regression analysis, we found that for each increase in the percentile of total social vulnerability, the risk of cumulative incidence of death from COVID-19 decreases by 29% (crude Relative Risk: 0.71, 95% CI: 0.55-0.92). When this is adjusted by population density it changes slightly, with the risk of cumulative incidence of death from COVID-19 decreasing by 28% (adjusted Relative Risk: 0.72 95% CI: 0.57-0.94), both statistically significant associations (p = 0.01).
When disaggregating the regression analysis by topic, most often we observed a decrease in the risk of the incidence of deaths from COVID-19 as the percentile of social vulnerability of the province increases. The relative risks are less than one in the other categories, from 0.30 concerning socioeconomic status to 0.73 in healthcare. In all of them, the results in the adjusted model are statistically significant (p = 0.01). Except for categories 2 and 5 ("Household, composition and disability" and "Epidemiological Factors") where we see an opposite situation. For each increase in the percentile of these categories, the risk of cumulative incidence of death from COVID-19 increased by 84% (IRRa: 1.84 95% CI: 1.45-2.35) and 152% (IRRa: 2.52 95% CI: 2.03-3.13) respectively, after adjusting for population density ( Table 2).
In all the regression analyses adjusted for population density, the directionality of the association is maintained.

IV. Discussion
In this study we adapted the SVI tool and included COVID-19 specific epidemiological (Obesity, Smoking, Cardiovascular diseases, Immunosuppression, etc.) and health system factors (ICU beds, number of physicians, per capita expenditure in health, number of COVID19 laboratories, etc.) to categorize the social vulnerability in various provinces of Peru. We then compared these to the mortality data from 2017 to 2020. As expected, we identified the distribution of the high and very high vulnerability index in rural provinces located or bordering the Andes Mountains. These are due to the lower access to health services and other public services, as well as unfavorable socioeconomic conditions in the populations in the mountainous regions of Peru [23]. The Ministry of Development and Social Inclusion of Peru also identifies these high Andean and Amazonian areas as the most vulnerable to any health condition or disaster [24]. Our vulnerability finding coincides and represents the unfavorable conditions of many provinces against COVID-19 and, probably, many other acute and chronic health problems.
The six themes in our instrument varied in distribution across Peru with socioeconomic vulnerability, housing, and transportation are distributed in the Peruvian highlands, the conditions of epidemiological vulnerability in the coastal areas, and the conditions of health system vulnerability in the southern part of the country.
We found that the mortality from COVID-19 increases with factors reflecting the composition of the household or the presence of a disability. Considering that the composition of the household evaluates vulnerable groups such as the elderly, minors, and the presence of disability, this result reflects the known biological vulnerability to COVID-19 [25], or other factors that generate vulnerability in these groups. Similarly, and to a greater extent, we saw increased mortality from COVID-19 and the presence of epidemiological factors such as obesity, cardiometabolic, and chronic respiratory diseases, corroborating that the presence of this group of comorbidities increases the fatal outcome from COVID-19, therefore, placing populations with these comorbidities in a situation of higher mortality [26].
The death rate from COVID-19 was higher in coastal areas of Peru-areas in which the pandemic is known to have started and spread rapidly. However, in most of these provinces, the SVI-COVID19 assessment was low or very low, indicating that with the COVID-19 pandemic there are other variables or factors that should be considered beyond those included in our instrument. We can infer, for example, that coastal areas with low vulnerability have greater urbanization, higher population density that led to closer contact between individuals, greater saturation of hospitalization and emergency services in hospitals, higher mobilization of people within the city and between provinces, and a greater probability of areas of agglomeration thereby increasing the transmission of SARS-CoV-2 [27,28].
The set of the six themes evaluated with SVI-COVID19 present a negative association with the mortality rate from COVID-19, that is, the higher the social vulnerability, the lower the mortality. This result is partially consistent with studies carried out in the United States, where the evaluation of the first months of the pandemic mentions a positive association between mortality from COVID-19 and SVI, which differs from our study. However, as the pandemic progresses, we observed a reverse association, showing that the relationship between both variables is not unidirectional, and is dependent on the time course of the pandemic [20,29]. This may be partly due to public health interventions. For example, since the start of the pandemic in Peru, there have been numerous interventions from the Ministry of Health at the national level to prevent and control the pandemic. Due to the emergency, these measures were multiple, and some were constantly variable, including implementation and relaxation periods, depending on the pandemic epidemiologic situation [30].
Two other elements can explain this factor. COVID-19 disproportionately impacted less vulnerable counties early and later in the pandemic. In the first place, the population with the highest socioeconomic level was affected, later expanding to the rest of the population. This suggests that the impact of COVID-19 is not static but can migrate from less vulnerable counties to more vulnerable ones and back again over time [29].
At the beginning of the pandemic, the communities with the lowest vulnerability index had access to private health services, which to a certain extent were able to absorb the number of patients. However, with the general increase in cases, the health system (public and private) was saturated and unable to respond to the amount of population in need of medical care, affecting the entire population in general, and thus the "advantage" observed at the beginning, could have been diluted, generating in practice a barrier to access that could be comparable to that in which hospital services are non-existent or very scarce. It should be noted that at the beginning of the pandemic (hospital capacity indicators compared to LATAM) and the closing of this gap was partial, late and transitory [31].
On the other hand, social vulnerability is directly related to the existence of gaps in the six topics of our instrument, which can also be analyzed from the point of view of inequities and inequality in health, the latter understood as the different distribution of wealth and access to care according to geographic location. Thus, we consider that the SVI represents a measurement of health inequities, therefore, of the existing gaps in Peru [32,33].
It is necessary to consider the possible influence of social vulnerability with the underreporting of deaths, considering that SINADEF is a computer tool whose registration is more complex in areas without internet access, which has been exacerbated during the first and second waves of the pandemic [34].
The results of the negative association between SVI-COVID19 and the COVID-19 mortality rate could be explained in part by the low registration of mortality, specifically in populations with the greatest social vulnerability. The mortality registration in regions of Amazonas, Ayacucho, Cajamarca, Junín, Loreto, and Pasco, of Peru is less than or equal to 60%. Since these are regions with higher social vulnerability, our results may be skewed due to unregistered deaths.
During 2020 and 2021, registers showed a significant exceed in the number of deaths, according to the reference pattern previously estimated by the INEI for those years. The estimates of deaths expected by the INEI only reach the regional level, but not provinces or districts.
A limitation in our study is the use of mortality data. While the current mode of collecting mortality data by the internet has increased the overall collection of data, it has likely led to underreporting from areas with poor technology and internet connectivity [34].
The registration of deaths with medical certification in Peru-with documentation of the cause of death-in 2019 was 71.6%, so 54,372 deaths in 2019 did not have a medical death certificate or whose medical certificate of death was not entered into SINADEF or the EsSalud mortality database. SINADEF is not the only death registration system with medical certification of causes of death. EsSalud registered 17,483 deaths that are not registered in SINADEF. The registered deaths drop to 62.15% of all deaths if we consider deaths registered by SINADEF [34].
Additionally, other studies report that good socioeconomic conditions and health expenditures, parameters considered in the SVI, can reduce the rate of COVID-19 infection when the incidence is low, but not when the spread has an exponential advance [32]. Among the limitations of our study is the ecological design, since the results can only be interpreted at a population level and cannot address the characteristics that can help to identify the individual factors that can define the personal resilience ability to respond to the COVID19 pandemic. The data obtained from institutional databases could present underreporting bias or misclassification, especially for populations with less accessibility. The crosssectional characteristic of our study limits the understanding of the evolutionary process of the risks or vulnerable conditions in relation to the health measures or policies assumed by the Government. Considering the existing inequalities within the 25 regions that make up Peru, further studies that look at vulnerability analysis with a greater spatial disaggregation, particularly for non COVID19 specific issues, including that of other diseases and conditions that have been unfortunately neglected due to the pandemic.

V. Conclusions
Our results identify the provinces of the country with high and very high vulnerability, most of which are in the rural areas bordering the Andes Mountains. Through this mapping, the index can help government authorities to improve prevention, control, and resource management policies in more detail [20].
It is possible that this negative association found between SVI and the mortality rate could be related to the low reporting coverage of deaths in general and deaths from COVID-19 in the provinces with a higher index of social vulnerability.
Likewise, this study highlights the need for better registration and epidemiological reporting systems. As we have seen, there is a limitation, which mainly affects rural and poor populations, in terms of adequate health data, which can lead to misreporting and potentially increase health inequalities to the detriment of vulnerable people, especially in Andean and Amazonian areas. Likewise, it can orient authorities incorrectly by giving a false perception about the actual health needs of the populations under study and directing the limited resources oriented towards health.
Additional studies are required to determine the causes of the paradoxical results found in this study to assess the importance of other individual and community factors that may have affected the burden of disease due to COVID19.