Risk of death, hospitalization and intensive care unit admission by SARS-CoV-2 variants in Peru: a retrospective study

Abstract

Objectives

Peru has had the highest death toll from the pandemic worldwide; however, it is not clear what the effects of the different variants on these outcomes are. The study aimed to evaluate the risk of death, hospitalization, and intensive care unit (ICU) admission rates of COVID-19 according to the SARS-CoV-2 variants detected in Peru from March 2020-February 2022.

Methods

Retrospective study using open-access databases were published by the Peruvian Ministry of Health. Databases of genomic sequencing, death, COVID-19 cases, hospitalization and ICU, and vaccination were used. Crude and adjusted Cox proportional hazards regressions with clustered variances were modeled to calculate the hazard ratio (HR) of outcomes by variant.

Results

Lambda variant had the highest risk of death (HR 1.92, 95% CI 1.37-2.68), whereas the Delta variant had the lowest risk (HR 0.50, 95% CI 0.31-0.82). Mu variant had the highest risk of hospitalization (HR: 2.39, 95% CI 1.56-3.67), Omicron the lowest (HR 0.45, 95%CI 0.23-0.90), and Gamma had the highest ICU admission rate (HR 1.95, 95%CI 1.40-2.71).

Conclusion

SARS-CoV-2 variants showed distinctive risks of clinical outcomes, which could have implications for the management of infected persons during the pandemic.

Introduction

On March 11, 2020, the World Health Organization declared COVID-19, a disease caused by SARS-CoV-2, a global pandemic, characterized by the development of a series of signs and symptoms in the infected population, including respiratory, psychiatric, and cardiovascular problems, among others [1].

Since the discovery of the Wuhan virus, D614G, or the original strain, different variants and lineages have been identified in such a relatively short time, indicating that SARS-CoV-2 has a higher mutation rate compared to other viruses [2]. While Omicron is the predominant variant, there have been approximately 600 variants of which those currently circulating can be classified as variants of interest and variants of concern [3]. These variations in the virus genome did not only bring structural differences, especially in the spike protein but also in virulence, with diversity in disease severity and transmissibility [4].

The virulence of SARS-CoV-2 may differ between geographic locations because some local strains have differential virulence compared to others. This is because of multiple mutations in its genetic sequence that lead to increased infectivity, pathogenicity, and antigenic capacity [5]. The variants identified to date have presented a heterogeneous geographic distribution, causing mortality and morbidity rates in different proportions depending on the affected country or continent [5]. It is mentioned that ancestry could play an important role in the response of the immune system against SARS-CoV-2, attributing higher infection rates to Latin and African Americans compared with those of European descent. In this sense, the impact that each variant of interest or concern can generate may vary according to the population in which it is distributed [6]. This information will be vital for implementing public health measures to control the spread of variants with local predominance and based on local characteristics.

Worldwide, one of the countries most affected by the pandemic has been Peru, currently reporting 3.29 million cases, positivity of 11.64%, and a case fatality rate of 7.22%. Mortality in Peru is associated with the male sex, older people, and the presence of comorbidities, in particular obesity and diabetes mellitus [7]. It was initially considered that residence at high altitudes, Peru is a country with a wide distribution of altitudes, was a protective factor against SARS-CoV-2 infections and death, but later studies found that this was not correct [8,9].

As part of the pandemic mitigation plan, the Peruvian government developed a nationwide genomic surveillance project with the objective of sequencing positive patients with a cycle threshold (Ct) ≤30, based on the Pan American Health Organization recommendations [10], determining the variant or lineage from which their infection originated for epidemiological control. However, it is not clear what the effects of the different variants were in the Peruvian population.

Understanding how the different variants are associated with severe outcomes such as death, hospitalization, or intensive care unit (ICU) admission could lead to the development of specific mitigation strategies and variant-specific treatments. Therefore, the aim of the present study was to assess the risk of death, hospitalization, and ICU admission by SARS-CoV-2 variants in Peru from March 2020 to February 2022.

Materials and methods

Study design and population

The study was designed as a retrospective cohort, and included people infected with SARS-CoV-2 who were notified through the national platform for disease control and epidemiological surveillance (Netlab 2.0), and whose samples were selected for genome sequencing. The results are reported in the same platform and published in the open-access databases by the Peruvian Ministry of Health.

Data record and merging

The deaths, cases, hospitalization, genome sequencing data, and vaccination information were obtained from the open-access databases of the Peruvian Ministry of Health (Supplementary Material 1).

The open-access databases identify each individual only by means of an encrypted code. Personal data such as name, telephone number, or specific residential address are not included. The merging of the five databases was performed using the encrypted code, which allowed us to construct a timeline of the disease from sample collection to outcome (death, hospitalization, or ICU admission) over a 30-day period. The initial database used for the fusion process was the SARS-CoV-2 sequencing database, which included all patients infected with the identified SARS-CoV-2 variant.

We included 10,733 registries from people aged ≥18 years, with high viral load (Ct ≤30), with or without symptoms whose sample was genome sequenced and reported in the Netlab 2.0 and published in the open-access COVID-19 databases of the Peruvian Ministry of Health.

Variables

The outcomes evaluated were death, hospitalization and ICU admission over a 30-day period, considering day 1 as the sampling date. This period was considered since it is related to the possible acute effects of COVID-19 and not to those of long-COVID.

A total of five categories of SARS-CoV-2 variants were considered: Lambda, Gamma, Delta, Mu, Omicron, and others. The classification was performed based on World Health Organization variants, the lineage list, and Pango nomenclature [11]. Since the Alpha and Zeta variants were of low prevalence in Peru, they were included in the others category.

Other variables considered were the number of doses of the SARS-CoV-2 vaccine taken before the time of infection, age (years), sex, number of hospitals and ICU beds for COVID-19 available within the province of residence up to 7 days before the onset of the outcome. Information on region and province and district of residence was also included.

Statistical analysis

Descriptive statistics were performed showing mean and standard deviation for numerical variables, and absolute and relative frequencies for categorical variables. Pearson's chi-square was used to test for the association between SARS-CoV-2 variants and death, hospitalization or ICU admission. Survival analysis for death, hospitalization and ICU admission according to variant was analyzed graphically with the Kaplan-Meier method.

A time-to-event analysis was performed using the Cox proportional hazards model to assess the association of the different variants with death, hospitalization, and ICU admission. Three separate models were developed, one for each covariate-adjusted outcome:

Equation 1. - Adjusted Cox regression model for death

$$\begin{array}{ccc}\hfill h\left(t\right|X)& =& h0\left(t\right)exp(\beta 1variant+\beta 2sex+\beta 3age+\beta 4doses\hfill \\ & & +\beta 5hospitalization+\beta 6ICU+\beta 7beds+\beta 8ICUbeds+\epsilon )\hfill \end{array}$$

Equation 2. - Adjusted Cox regression model for hospitalization

$$\begin{array}{ccc}\hfill h\left(t\right|X)& =& h0\left(t\right)exp(\beta 1variant+\beta 2sex+\beta 3age+\beta 4doses\hfill \\ & & +\beta 7beds+\epsilon )\hfill \end{array}$$

Equation 3. - Adjusted Cox regression model for ICU admission

$$\begin{array}{ccc}\hfill h\left(t\right|X)& =& h0\left(t\right)exp(\beta 1variant+\beta 2sex+\beta 3age+\beta 4doses\hfill \\ & & +\beta 7beds+\beta 8ICUbeds+\epsilon )\hfill \end{array}$$

where $h\left(t\right|X)$ is the hazard at the time t of the outcome adjusted on covariariates, $h0\left(t\right)$ is the baseline hazard for the outcome, $exp(\beta 1variant$) is the hazard ratio (HR) for the outcome of infection with a given variant compared to the other variants (e.g., Delta infected vs all non-Delta infected), $exp(\beta 2sex$) is the HR of the outcome being a male compared with being a female, $exp(\beta 3age$) is the HR of the outcome for age as a continuous variable, $exp(\beta 4doses$) are the HRs of the outcome of being vaccinated with one, two or three doses compared to zero doses, $exp(\beta 5hospitalization$) is the HR of death when hospitalized compared to not being hospitalized, $exp(\beta 6ICU$) is the HR of death of those who were ICU admitted compared to non-ICU, $exp(\beta 7beds$) is the outcome for number of available hospital beds quintiles up to 7-days prior the death date, $exp(\beta 8ICUbeds$) is the HR of death and ICU for number of available ICU beds quintiles up to 7-days before the date of death, and $\epsilon$ is the error term. The proportional hazards assumption was tested using Schoenfeld residuals, and the linearity of numeric variables was evaluated using Martingale-based residuals. Standard error was adjusted with clustered variances; the region of residence was considered as the cluster.

An established reference group was not set because sequencing was restricted to a sample with a Ct ≤30, limiting the variants detected for this study. So, when calculating the HR for a particular variant with 95% CI, the reference group is considered as the hazard of the outcome for all the other variants (e.g., the hazard of death for Delta vs the hazard of death for all non-Delta). The Omicron variant is generally considered a low-risk variant [12,13], so to assess whether the results were consistent, a sensitivity analysis was performed, leaving the Omicron variant out of the categorization. All statistical analyses were done using STATA 17.0 software, and statistical significance was considered as a P-value <0.05.

Results

In Table 1 it is observed that up to February 2022, Delta was the most prevalent variant of COVID-19, followed by Lambda and Gamma. The other variants represented only 6.8% of the total.

Table 1.

Sample characteristics (N = 10,721).

Variable		N (%)
Age (years)a		39.75 ± 17.28
Sex
	Feminine	5618 (52.4)
	Masculine	5107 (47.6)
Variant
	Gamma	1350 (12.5)
	Delta	4331 (40.4)
	Lambda	3019 (28.1)
	Mu	149 (1.3)
	Omicron	1139 (10.6)
	Othersb	733 (6.8)
Deaths		283 (2.64)
Hospitalizations		554 (5.2)
Instensive care unit admissionsc		139 (25.1)

^aDisplays mean ± standard deviation.

^bIncludes Alpha and Zeta variants.

^cPercentage calculated based on total of hospitalized people.

There was a clear dominance of the Lambda variant from February to mid-August 2021. Delta variant become the most prevalent variant from September until December 2021 during which Omicron began to spread, surpassing Delta by January 2022 (Figure 1 ).

Figure 1.

Variation in number of cases of SARS-CoV-2 variants during the study period. Data before December 01, 2020 is not shown due to the low number of sequencing in those days, and for better data visualization.

Table 2 presents the results of a bivariate analysis of the associations between the variants and outcomes. There were significant associations with death, hospitalization, and ICU admission. Lambda had the highest percentage of deaths, and Mu and Gamma had the highest hospitalization and ICU admission, respectively.

Table 2.

Number of deaths, hospitalizations and ICU admissions by SARS-CoV-2 variants.

Variant	Deaths	P-valuea	Hospitalizations	P-valuea	ICU admissions	P-valuea
	n (%)		n (%)		n (%)
Gamma	51 (3.9)	<0.001	83 (6.2)	<0.001	37 (44.1)	0.001
Delta	69 (1.6)		237 (5.4)		69 (29.2)
Lambda	123 (4.1)		169 (5.6)		36 (20.7)
Mu	4 (2.7)		17 (11.4)		3 (17.7)
Omicron	12 (1.1)		23 (1.9)		3 (13.6)
Others	23 (3.1)		22 (2.9)		8 (38.1)

ICU, intensive care unit.

Deaths and Hospitalizations percentages are calculated based on the total number of infected by each variant.

ICU admissions percentages were calculated based on the total number of hospitalized people by each variant.

Others include Alpha and Zeta variants

^aPearson Chi-squared.

In the survival analysis, it can be seen that all variants except Delta and Omicron showed higher cumulative mortality, although a flatter slope can be observed for Omicron (Figure 2 a). In regard to hospitalization, a similar pattern can be observed (Figure 2 b); for ICU admission, Lambda and Omicron variants showed a lower cumulative incidence of ICU admission, and the Gamma variant was the highest (Figure 2 c).

Figure 2.

Kaplan-Meir survival analysis plots for death (a), hospitalization (b) and ICU Admission (c) by SARS-CoV-2 variant.

ICU, intensive care unit.

The crude and adjusted HRs are shown in Table 3 . For death, infections with Gamma and Lambda variants were risk factors, while Delta and Omicron had a HR below one in the crude analysis, but after adjustment for covariates, only the Delta variant showed a decreased HR for death compared to all other variants (HR: 0.50, 95%CI 0.31-0.82).

Table 3.

Crude and adjusted Cox regression analysis of death, hospitalization and ICU admission by variants^a.

Variant	Death		Hospitalization		ICU admission
	cHR	aHR	cHR	aHR	cHR	aHR
Gamma	1.54 (1.13-2.08)	1.24 (0.85-1.83)	1.24 (0.98-1.56)	1.20 (0.75-1.93)	1.92 (1.33-2.78)	1.95 (1.40-2.71)
Delta	0.47 (0.36-0.62)	0.50 (0.31-0.82)	1.12 (0.94-1.32)	1.00 (0.75-1.33)	1.09 (0.79-1.50)	0.90 (0.59-1.37)
Lambda	1.99 (1.58-2.53)	1.92 (1.37-2.68)	1.13 (0.94-1.35)	1.17 (0.73-1.87)	0.62 (0.43-0.91)	0.69 (0.42-1.13)
Mu	1.02 (0.38-2.74)	1.75 (0.53-5.82)	2.31 (1.42-3.74)	2.39 (1.56-3.67)	0.57 (0.18-1.78)	0.56 (0.19-1.62)
Omicron	0.37 (0.21-0.66)	0.71 (0.35-1.41)	0.36 (0.34-0.55)	0.45 (0.23-0.90)	0.42 (0.13-1.33)	0.46 (0.10-2.02)
Others	1.21 (0.79-1.86)	0.75 (0.30-1.85)	0.56 (0.37-0.86)	0.58 (0.40-0.85)	1.32 (0.65-2.70)	1.43 (0.73-2.81)

aHR, adjusted HR; cHR, crude HR; CI, confidence interval; HR, hazard ratio; ICU, intensive care unit.

^aEach variant model considers the reference group as the combined hazard of all other variants.

cHR and aHR with 95% CI are presented.

HR and 95%CI in bold letters are p<0.05.

Death model controlled for sex, age, number of COVID-19 vaccine doses, hospitalization, ICU admission and availability of hospitalization and ICU beds during the week of admission.

Hospitalization model controlled for sex, age, number of COVID-19 vaccine doses and availability of hospitalization beds during the week of admission.

ICU Model controlled for sex, age, number of COVID-19 vaccine doses and ICU rooms availability during the week of ICU admission.

For hospitalization, in both the crude and adjusted analysis, the Mu variant presented an increased HR compared to all other groups (HR 2.31, 95% CI 1.56-3.67), while Omicron showed a reduction in hospitalization risk of 55% after adjusting for covariates. For ICU admission, the Lambda variant showed a reduction of 38% in the risk of the outcome but only in the crude analysis, while the Gamma variant exhibited a higher risk compared to all other groups (HR 1.95, 95% CI 1.40-2.71).

In the sensitivity analysis, the results obtained were similar to those of the full models, meaning that the effect of the variant was not diluted by Omicron (Supplementary Table 1).

Discussion

This study aimed to determine the impact of SARS-CoV-2 variants on mortality, hospitalization, and ICU admission in Peru. We found that those infected with the Lambda variant had a higher risk of death, while those infected with Delta variant presented the lowest death risk. In the case of hospitalization, the Mu variant had the highest risk.

The Omicron variant was associated with the lowest risk of death and hospitalization in the crude analysis. This is in accordance with its higher transmissibility, but lower severity observed worldwide [14,15]. Regarding ICU admission, we have observed that the Gamma variant had the highest risk.

Different from other countries, especially those in Europe and North America [16], Peru did not have much representability of Alpha variant. It is generally considered to have a higher risk of mortality, hospitalization, and ICU admission compared with the wild strain [5]. Also, even though the number of cases infected with the Alpha variant was small (n = 14) in our study, it was not significantly associated with hospitalization and ICU admission, while the mortality associated with it was lower. An earlier study also reported a lower risk of hospitalization in association with the Alpha variant [17].

Although both the Delta and Lambda variants are capable of immune evasion, the Lambda variant is associated with mutations L4521 And F490S that makes it easier for it to bond with the angiotensin-converting enzyme-2 receptor [18]. This could explain its association with adverse health outcomes even in vaccinated persons [19], and thus the higher associated mortality in this study.

The Gamma variant was associated with a higher risk of ICU admission (HR: 1.95, 95% CI 1.4-2.71). Similar findings have been reported from studies in Canada [20] and France [21], and could be explained by the capability of this variant to provoke severe symptoms in young and middle-aged patients, including those without comorbidities [22,23]. Mu variant was associated with the highest risk for hospitalization in our patients in contrast to the results from a study in Colombia [24], although the Colombian study was done with data from a single state, while we used nationwide data, giving more robust estimates, thus explaining the difference. Also, it is possible that genetic differences between both populations could be involved [25].

It is noteworthy that the Lambda variant was associated with the highest risk of death. Lambda variant was first detected in Peru in August 2020, it presents 19 mutations of which seven are in the spike protein, which is linked to higher chances of evasion of neutralizing antibodies [4,19] and immune resistance [26]. It is possible that, since the variant originated in Peru, its mutations are focused on the host genome, in this case, the Andean population, this being similar to the hospitalization in India for Delta variant infections [27].

Although the study period comprised the first three waves of the COVID-19 pandemic in Peru, it was not possible to fit the regression models by pandemic wave since there were only 15 observations for the first wave. However, we did factor in the availability of beds, which acts as a proxy variable.

The study has some limitations. First, information came from secondary databases, which might be attended by a possible information bias because of misclassification or inadequate data recording, especially in areas with lower technology access and trained personnel. Second, SARS-CoV-2 sequencing was not broadly performed before the second wave, and considering that only samples with specific criteria were considered for sequencing, during subsequent waves it is possible that some variants were not identified. Third, an established reference group was not possible to set because only samples with a Ct≤30 were eligible for sequencing, and considering that not all the samples are sequenced, it is possible that the representativeness of certain variants was lost. The pandemic waves have distinctive traits that could act as confounders in our model, nonetheless, one of the main characteristics was the availability of hospital and ICU beds, which can be considered as a proxy, but some information could not be controlled. Factors such as ambient air pollution [28] were not considered in this study because of a lack of data. These limitations should be considered in further studies.

In conclusion, the different variants had distinctive effects on clinical outcomes which could have implications for the management of infected persons during the pandemic.