Skip to main content

Lethality risk markers by sex and age-group for COVID-19 in Mexico: a cross-sectional study based on machine learning approach

Abstract

Background

Mexico ranks fifth worldwide in the number of deaths due to COVID-19. Identifying risk markers through easily accessible clinical data could help in the initial triage of COVID-19 patients and anticipate a fatal outcome, especially in the most socioeconomically disadvantaged regions. This study aims to identify markers that increase lethality risk in patients diagnosed with COVID-19, based on machine learning (ML) methods. Markers were differentiated by sex and age-group.

Methods

A total of 11,564 cases of COVID-19 in Mexico were extracted from the Epidemiological Surveillance System for Viral Respiratory Disease. Four ML classification methods were trained to predict lethality, and an interpretability approach was used to identify those markers.

Results

Models based on Extreme Gradient Boosting (XGBoost) yielded the best performance in a test set. This model achieved a sensitivity of 0.91, a specificity of 0.69, a positive predictive value of 0.344, and a negative predictive value of 0.965. For female patients, the leading markers are diabetes and arthralgia. For males, the main markers are chronic kidney disease (CKD) and chest pain. Dyspnea, hypertension, and polypnea increased the risk of death in both sexes.

Conclusions

ML-based models using an interpretability approach successfully identified risk markers for lethality by sex and age. Our results indicate that age is the strongest demographic factor for a fatal outcome, while all other markers were consistent with previous clinical trials conducted in a Mexican population. The markers identified here could be used as an initial triage, especially in geographic areas with limited resources.

Peer Review reports

Background

Among middle-income countries, Mexico is one of the most severely affected by COVID-19. According to recent estimates, the burden of disease due to COVID-19 in Mexico amounts to 2,165,424.5 disability-adjusted life years (DALYs) [1]. Mexico has the fifth highest number of COVID-19 deaths in the world, after the United States, Brazil, India and Russia [2]. In October 2022, nearly 6.5 million deaths from COVID-19 were reported worldwide and 330,227 were reported in Mexico, with a cumulative weekly mortality rate of 253 per 100,000 population [3]. Great heterogeneity has been reported among age- and sex-adjusted case fatality rates (CFR) in Mexican adults with a positive diagnosis of SARS-Cov-2, ranging from 4.6% in private hospitals to 18.9% in public facilities [4]. These divergences in mortality rates are largely due to inequalities in the Mexican population [5] and the response to the pandemic by Mexican health authorities [6]. Therefore, markers of risk of death are urgently needed to allow clinicians to perform initial triage in patients with COVID-19 and to anticipate a fatal outcome, especially in the most socioeconomically disadvantaged regions of Mexico.

Several studies around the world have evaluated factors associated with severe disease and death from COVID-19 [7, 8]. Also, predictive models for COVID-19 using laboratory tests or imaging markers have been proposed [9,10,11]. However, their potential use by clinicians, especially in low-resource healthcare settings, is limited [12]. Thus, a model to predict the risk of death using data that are readily available in clinical areas, such as patient signs and symptoms, would be extremely valuable in those areas of Mexico where access to specific tests is limited.

This study is aimed to assess risk markers, stratified by sex and age-group, for COVID-19 lethality in Mexico by a machine learning (ML) approach. Particularly, we evaluated various supervised classification models to predict lethality in patients diagnosed with COVID-19 using clinical data. Then, we applied an interpretability approach to identify those markers that increase lethality risk, by sex and age-group.

Methods

Study design and data source

This is a cross-sectional study using data collected from an open-access database. This database includes symptoms and signs of patients reported to the Epidemiological Surveillance System for Viral Respiratory Disease (SISVER).

SISVER collects data on patients with suspected COVID-19. The cases are reported using the questionnaire “Epidemiologic study of suspected viral respiratory disease”, available on the official site of the Federal Health Secretariat in Mexico.Footnote 1 Typically, the first-contact clinician completes the form with patient demographics, comorbidities, symptoms, signs of illness, and medication prior to hospital admission. If the patient is admitted to an intensive care unit (ICU), the doctor adds this admission to the questionnaire. The patient’s outcome, either death or discharge, is also recorded. Unfortunately, the evolution of symptoms and signs is not recorded.

SISVER data are processed and published in an open-database available on http://covid-19.iimas.unam.mx/. In this database, data on symptoms, signs, comorbidities, and prior medication were recorded as categorical variables. These variables were recorded with three values, “Yes,” “No,” and “Not known.” For instance, if a patient reported chronic obstructive pulmonary disease (COPD), then the value for this comorbidity was recorded as “Yes;” if the patient stated they do not have COPD, the value was recorded as “No;” otherwise, it was coded as “Not known.” All data were collected at hospital admission. User registration providing name, institution, position, and email is required to access data.

Patient selection

The database included information on 5,490,290 suspected COVID-19 cases that were admitted to the hospital from June 9, 2020, to March 1, 2021. Inclusion criteria were, patients with COVID-19 confirmed by RT-PCR living in the State of Morelos who were not undergo intubated and were not admitted to the ICU. A total of 11,564 patients were selected. Baseline characteristics of patients are described in Table 1. Continuous variables are reported as mean and standard deviation, and categorical variables are reported as percentages. From all included patients, 46% were females and 54% were males, with an average age of 49.12 ± 17.53 years. The most common symptoms in both sexes were cough, fever, headache, and myalgia. Lethality rates were 5% for females and 9% for makes. All data used herein are publicly downloadable from this open-access database.

Table 1 Characteristics of patients

The process of training and evaluation of ML methods to identify markers of lethality risk in the population under study is shown in Fig. 1. It included patient selection, preprocessing, model selection, final performance evaluation, and identification of risk markers for lethality.

Fig. 1
figure 1

Process to identify lethality risk markers for COVID-19 patients

Preprocessing

To select the clinical variables for the dataset, those fields in the database with more than 90% of missing data were excluded. Thus, 34 variables with no missing values were included. One-hot encoding was used for all categorial variables. The time of elapsed from the onset of symptoms to the start of medical care was estimated in days. Age and time elapsed were regarded as continuous variables. Mortality, selected as the independent variable, was used to predict lethality (survival or death). All other variables were regarded as dependent. A brief description of all clinical variables used is shown in Additional file 1. Data were transformed as a standard Z distribution. The models were trained with 80% of data, and the remaining 20% was used for testing.

Model selection

Four machine learning classification methods were used in this study: Logistic regression (LR), support vector machine (SVM), random forest (RF), and eXtreme gradient boosting (XGBoost). For hyperparameter selection, we relied on a grid search based on 10 repetitions of stratified tenfold cross-validation on the training set.

The hyperparameters evaluated for each method were set as follows. For LR, saga and liblinear optimizer were used [13]. For weight penalization, ℓ1, ℓ2, and elastic net norms were considered [14]. Ratios of 0.1, 0.2, 0.4, 0.5, 0.6, 0.7, 0.8, and 0.9 were used for ℓ1. Different values for the strength C of penalization were considered on a logarithmic scale, including − 3, 3, and 7. For SVM, a penalization value ranging from − 4 to 4 was considered [14]. For RF, the maximum number of features was set using a base-2 logarithm for each split, and impurity decrease was set to 1e − 4 [15]. For XGBoost, dropout rates of 0.03 and 0.5 were evaluated, and the learning rate was considered as 0.03, 0.05, and 0.1 [16]. For both RF and XGBoost, 200, 250, and 300 estimator trees were used, with a maximum depth of 5 and 6 nodes. All these hyperparameters were evaluated as usually during the training of ML methods.

The selection criterion for training was the area under the receiver operating characteristic curve (ROC AUC). The model with the highest mean of cross-validated AUC was selected. All metrics computed to measure model performance on the test dataset are shown in Table 2.

Table 2 Metrics evaluated on test set

As our dataset is imbalanced, the ROC curve and the Precision–Recall curve (PR) were computed on the test. The ROC curve analyzes the sensitivity and specificity achieved by the model. Meanwhile, the PR curve shows performance based on false-positive and false-negative rates.

Identification of risk markers by the interpretability approach

An interpretability approach was used to identify risk markers. In particular, the Shapley Additive explanations (SHAP) algorithm [17] was used to interpret the output of predictive models. According to Molnar [18], SHAP explains the prediction of an instance \(x\) by computing the contribution of each feature to the prediction. The SHAP algorithm computes Shapley values from coalitional game theory, where the game is the prediction task for a single instance, and the players are the values of the features. This algorithm computes the contribution of each player to the game. The Shapley value \(\phi j(val)\), the payout that a player \(j\) receives for the game, is computed as follows:

$${\phi }_{j}(val)={\sum }_{S\subseteq \{1,...p\}\backslash \{j\}}\frac{|S|!(p-|S|-1)!)}{p!} (val (S\cup \{j\}-val(S))$$

where \(S\) is a subset of the features used in the model, \(val\) is the vector of feature values of the instance to be explained, and \(p\) is the number of features. The Shapley value is the average marginal contribution of a feature value across all possible coalitions. The SHAP algorithm has already been used to calculate markers for diagnosis [19] and mortality by COVID-19 [20], for hypoxemia prevention [17], and for mortality after an infarction [21]. All algorithms were written in Python v.3.8. Scikit-learn,Footnote 2 pandas,Footnote 3 numpy,Footnote 4 and Jupyter NotebookFootnote 5 libraries were also used. Source code is available at https://github.com/rojas-mariano-salvador/Lethality-markers-for-COVID-19.

Results

Performance of lethality prediction models

The dispersion of AUC scores in each cross-validation fold for each model evaluated on the validated set are shown in Fig. 2 and Additional file 2. As shown, XGBoost performed better in terms of discrimination than the rest of the models during training. XGBoost achieved the highest performance, outperforming LR, SVM, and RF. The best parameters found for XGBoost were as follows: maximum depth = 5, estimator = 200, dropout rate = 0.3, and learning rate = 0.03.

Fig. 2
figure 2

Comparison of machine learning classification models using 10 repetitions of stratified tenfold cross-validation

Finally, the metrics shown in Table 3 were computed, and ROC and PR curves for the selected XGBoost model on the test set were plotted (Fig. 3). Overall, the model achieved an ability discriminative AUC score of 0.79 and a precision-recall of 0.503; sensitivity was 0.836 and specificity was 0.74.

Table 3 Performance of XGBoost model on test set
Fig. 3
figure 3

Curves for XGBoost model on test set. ROC curve (left) and PR curve (right)

Lethality risk markers by sex and age-group

The markers found to increase the lethality risk by sex through an interpretability approach are shown in Fig. 4. Bar plots (left) show the features in order of importance. On the other hand, beeswarm plots (right) show the impact of each feature value on the model output. The SHAP value is shown in the x-axis: larger positive SHAP values increase lethality risk, whilst larger negative SHAP values decrease the risk. Each patient is reported as a dot in this plot. Multiple dots in the x-axis shape a density. Colors indicate the value of each feature. Larger values are plotted in red, whilst smaller values are plotted in blue.

Fig. 4
figure 4

Risk markers in the general population and in male and female patients

First, risk markers were evaluated in both sexes. Age was the main marker that increased the risk of lethality. As shown in Fig. 4b, lethality increased with age (larger positive SHAP values on the x-axis). The presence of dyspnea, polypnea, and diabetes also increased the risk.

Then, risk markers were analyzed by sex. Chronic kidney disease (CKD) (Fig. 4c and d) was a good marker for male patients, and arthralgia was good for females (Fig. 4e and f). Common markers for both sexes were dyspnea, hypertension, polypnea, diabetes, and fever.

The markers with the greatest contribution to lethality were dyspnea and hypertension for males and females; however, the third most relevant marker for males was polypnea, while diabetes was the third for females. In both groups, fever was the marker with the lowest contribution to lethality.

Additionally, as shown in Fig. 4d and f, rhinorrhea in males and females, conjunctivitis in males, as well as a sudden onset of the disease and anosmia in females decreased the risk on lethality.

A summary of lethality risk markers by sex and age-group is shown in Table 4. Notably, respiratory compromise signs such as dyspnea and polypnea were the most frequent risk markers in all age groups.

Table 4 Risk markers by sex and age-group

Chest pain was a risk marker for male patients in various age groups. On the other hand, diabetes was the most frequent risk marker for females in various age groups; arthralgia was found to be a marker in females aged 60 years or older.

As shown in the swarm graphs in Additional file 3, the presence of some signs of the disease actually decreases the risk of lethality. Such is the case of odynophagia in male patients younger than 50–59 years old, or rhinorrhea and sudden disease onset in females in some age groups. In addition, the use of antipyretics increased the risk in males and females aged 50 years or older. Unexpectedly, we found that those patients who seek care on time increase the risk on lethality.

The distribution of deaths by sex and age-group is shown in Table 5. As shown, deaths were more frequent in older patients, with 96% of deaths in males and 95% in females in age groups ≥ 40 years. Interestingly, the model assigned a person an initial probability of death as a function of his or her age group, and this initial probability increases with age.

Table 5 Distribution of deaths by sex and age-group

The contribution of clinical history to prediction was also analyzed. The SHAP algorithm was used to analyze data of six patients in the group with significant lethality (50–70 years).

Figures 5, 6, 7 show individual predictions for three patients who died from COVID-19. In particular, these plots show the probability of death or lethality risk computed by the model and denoted as \(f(x)\). The average output probability from all the patients is denoted as \(E[f(x)]\), which represents the baseline risk according to age-group and sex. The clinical variables that increased the risk of lethality ranked in descending order. The contribution of each variable to the prediction is also shown as positive (red) or negative (blue).

Fig. 5
figure 5

Individual prediction for a male and a female patient in the 50–59 age group, who died from COVID-19

Fig. 6
figure 6

Individual prediction for a male and a female in the 60–69 age group who died from COVID-19

Fig. 7
figure 7

Individual prediction for a male and a female in the age group ≥ 70 years who died from COVID-19

The clinical data of a 59-year-old male patient who died from COVID-19 are shown in Fig. 5a. In this case, the risk of lethality was 0.823. The markers that increased the risk were dyspnea, polypnea, chest pain, and diabetes. In contrast, the clinical data of a 50-year-old female patient who also died are shown in Fig. 5b. The risk in this case was 0.779, and the markers that increased the risk were dyspnea, diabetes, and the use of antipyretics.

Similarly, the case of a 67-year-old male patient is shown in the left panel (a) of Fig. 6, and the case of a 63-year-old female patient is shown in the right panel (b). According to our model, the male had a risk of death of 0.863; the risk markers that led to this result were the presence of chest pain, obesity, myalgia, and the use of antipyretics, and to a lesser extent, polypnea, dyspnea, and fever. The risk of death for the female patient was 0.883; the markers that increased the risk were vomiting, dyspnea, diabetes, and a period of 7 days elapsed from the onset of the disease until medical care was given; arthralgia and hypertension also contributed of lethality.

The cases of a 75-year-old male (left panel) and a 70-year-old female (right panel) are shown in Fig. 7. The male had a risk of death of 0.935; because the manifestation of the dyspnea, odynophagia and myalgia, to have hypertension and a time of one days elapsed from the onset of the disease to medical care. The female had a risk of death of 0.951, with a time of 12 days elapsed from the onset of the disease to medical care, to have CKD, hypertension and diabetes as well as manifestation of the polypnea being the major contributors to a fatal outcome.

In all cases, the ground truth of all these patients was that they died. The interest to analysis these cases is to validate the performance of XGBoost model for distinguishing lethality and identifying the risk markers that increase the risk.

Overall, we observed that the variables that contributed to lethality are the symptoms related to respiratory involvement (dyspnea and polypnea). Also, hypertension was identified as a comorbidity that contributes to lethality.

In particular, chest pain was a marker that contributed to lethality in males. The use of antipyretics was also recurrent as a marker in males, in contrast with females.

On the other hand, even though diabetes was identified as a risk marker in both sexes, disaggregated data showed a higher frequency and weight in lethality for females.

Discussion

Summary

Since case lethality estimates the frequency of death among confirmed cases and it depends on factors such as the sensitivity of the case detection system, the responsiveness of health services, and a timely treatment [22], an early detection of risk markers for COVID-19 case fatality could favor a better outcome for infected patients.

The use of machine learning models to predict COVID-19 lethality in a stratified Mexican population has been little explored. This work is aimed to build models based on demographic and clinical variables (symptoms, signs, and comorbidities). These data are often available for practitioners in settings where resources, technology, and access to specialized personnel are limited, a situation common in a middle-income country like Mexico. The results of this study provide valuable information on the relationship between risk markers stratified by sex and age-group and COVID-19 lethality in Mexico.

Comparative with the state of the art

Efforts have been made to develop and validate models for predicting death in Mexican patients with COVID-19 using demographic and patient history factors [23, 24]. The PH-Covid19 scoring system was developed by Mancilla-Galindo et al. using data sets from the Mexico COVID-19 Epidemiological Surveillance Study and validated in cohorts including outpatients and inpatients. The multivariable Cox regression model includes eight predictors and yielded a discrimination of death of 0.8 evaluated by Harrell’s C-statistic [23]. Our study was stratified by sex and age-group, and it is noteworthy that our variables were not limited to comorbidities, sex and age, but we also included signs and symptoms.

Several diagnostic and prognostic models have been proposed for COVID-19, for instance the studies carried out by Aktar et al. [25] and Cini Oliveira et al. [26]. Both approaches evaluated several machine learning models to identify the best predictors of COVID-19 mortality based on demographic data, symptoms and signs, and comorbidities. Like our proposal, both studies use the SHAP approach to distinguish the markers that increase the risk of mortality. On the one hand, Aktar et al., evaluated the algorithms of RF, SVM, XGBoost, Decision Tree, Gradient Boosting Machine, and Light Gradient Boosting Machine (LightGBM) for predicting of mortality. The LightGBM model achieved a discrimination of 0.89, evaluated by the AUC score. On the other hand, Cini Oliveira et al. evaluated the algorithm of XGBoost for predicting of mortality achieved an AUC = 0.94. While our model achieved a lower discrimination power in AUC (0.79), we also evaluated the precision-recovery curve. Additionally, various risk markers identified in their study using the interpretability approach coincided with the ones we found herein.

Machine learning algorithms were used to find prognostic clinical biomarkers in patients with COVID-19 in other populations at the onset of the pandemic; however, small sample sizes in those early efforts limited model robustness and performance [27,28,29,30,31,32]. Subsequent studies included larger sample sizes [33,34,35,36,37]. Other studies have proposed models based on laboratory tests and imaging markers; however these approaches have shown a high risk of bias, have been optimistic in their reported performance [12] and have only limited application in areas where resources are scarce [38,39,40].

Sex and age as risk markers

Concurring with the findings reported by Aktar et al. [25], our results in the general population and stratified by age indicate that older age was the most significant predictor of mortality, as well. Our results confirmed that age is the strongest demographic factor for a fatal outcome in both sexes; however, the risk of COVID-19 lethality increases at a younger age in males than in females. Studies in other populations have reported age as a predictor of COVID-19 mortality [35, 41, 42]; in addition, two previous studies reported that the behavior of COVID-19 mortality and the case fatality rate with respect to age graphically resembles a J-shaped curve, which is interpreted as a minor impact of mortality at an early age and a progressive impact as age increases. This rate is higher in older populations [43, 44].

In a previous study of confirmed COVID-19 cases in Mexico, an age older than 41 years was associated with an increased risk of COVID-19 mortality, in line with our findings that the number of cases of death grew exponentially from the age group 40–49 years onwards [45]. On the other hand, Mesta et al. reported that an age older than 60 years was associated with death in patients hospitalized for COVID-19 in Mexico, which agrees with our results that lethality was concentrated in the age range of 60 years or older and in the male sex [46]. Immune senescence, characterized by progressive lymphopenia with CD4 + T cell depletion and decreased regulatory T cell function in aging, could be a factor that makes older individuals more sensitive to severe COVID-19 [33]. In recent estimates, the burden of COVID-19 in Mexico in DALYs was the highest in patients in the 60–79 age group [1].

The finding that male patients have a higher risk of adverse clinical outcomes for COVID-19 is consistent with previous reports in the literature, and may be related to a weaker immune response; indeed, mechanistic studies in human patients and animal models have shown that immune responses to respiratory virus infections are more effective in females than in males [47].

Comorbidities as risk markers

Considering that diabetes mellitus is a major cause of morbidity in the Mexican population [48], it is crucial to determine its role as a risk marker and its contribution to COVID-19 lethality. In our study, diabetes had a greater weight in case fatality among females in most of the age groups analyzed. However, these results do not rule out the risk posed by diabetes in COVID-19 lethality for males. This could be due to the coexistence of more than one comorbidity; in these cases, hypertension and CKD are better risk markers in males and contribute more to disease lethality [49].

In 2020, Bello-Chavolla et al. developed the MSL-COVID-19 score to predict case fatality rates in patients hospitalized for COVID-19 in Mexico [24]. In that work, CKD was also reported as a comorbidity associated with COVID-19 lethality. In our study, CKD showed a significant contribution to lethality in males. However, those researchers reported that COPD and immunosuppression were risk factors significantly associated with lethal cases of COVID [38]. In the largest cohort study on the disease conducted to date by OpenSAFELY, CKD was reported to be a key risk factor for COVID-19 mortality [50, 51].However, unlike Bello-Chavolla, we stratified our population by age group and sex, and found that CKD had a higher weight in males in the 30–49 and 60–69 age groups, and in females in the 70 and older age group.

Other studies, conducted in a population of the Mexican Social Security Institute (IMSS) [52], in a population of the State of Coahuila, Mexico [53], and in individuals treated in Mexican healthcare units and hospitals [8], all of them with a sample size comparable to ours, reported hypertension, diabetes, CKD, and obesity as comorbidities that increase the risk of mortality from COVID-19 [52, 53]. Our results indicate that arterial hypertension is the comorbidity with the greatest contribution to lethality regardless of sex; interestingly, it was also a risk marker in most age-groups.

The role of diabetes and obesity as risk factors for death from COVID-19 has been consistently reported in several studies in Mexico and worldwide [48, 54,55,56,57,58]; However, obesity was not a prominent risk marker in our work. The combined effect of diabetes and obesity on COVID-19 lethality has been attributed to the proinflammatory state associated with both comorbidities. In addition, SARS-CoV2 infection is known to elicit a dysregulated immune response that promotes hyperinflammation and endothelial dysfunction, which may result in a prothrombotic state [59] and an excessive oxidative stress response [60].

Overall, our results are in agreement with previous studies in the Mexican population, which reported older age, male sex, hypertension, diabetes, and CKD as risk factors for death and severity in patients with COVID-19 [4, 8, 45, 46, 48, 52, 53, 61,62,63].

This is of utmost relevance, given that 76% of the Mexican population over 20 years of age is overweight or obese, and the prevalence of obesity increased by 42%, particularly in persons over 50 years of age [64]. Additionally, 2.8 million Mexicans have diabetes, and a prevalence of 7.6% was reported for this disease in adults aged 30–39 years (5.4,10.7), according to the National Health and Nutrition Survey (ENSANUT) 2020 [65]. Thus, the profile of metabolic risk of the Mexican population makes it extremely vulnerable to a fatal outcome from COVID-19. According to our results cardiovascular diseases were not relevant in COVID-19 lethality, same results with the study by Castelnuovo et al. where they applied machine learning and Cox regression methods in Italian population, they found no association between these diseases with COVID-19 mortality [66]. However, the study´s results by Spinoni where they applied logistic regression models and estimated the survival rate with the Kaplan-Meier method in the Italian population to specifically evaluate the association of atrial fibrillation with COVID-19 mortality, showed that atrial fibrillation increases the risk of dying compared to those who do not have it, even more so when it is an event that occurs during the course of the disease compared to those patients with a history of this cardiovascular disease, possibly due to the their population longevity and the high prevalence of 20% in these patients [67]. 

Symptoms as risk markers

With respect to symptoms as markers of lethality, signs of respiratory distress like dyspnea and polypnea were the most frequent risk markers in all age groups. A study in Brazil reported that respiratory distress was more likely in male patients who died of COVID-19 [35]. Dyspnea, polypnea, chest pain, myalgia, arthralgia, fever, and chills were also the most frequent symptoms. In agreement with those results, dyspnea and fever were also reported by Aktar and Cini Oliveira [25, 26]. However, our results indicate that headache, odynophagia, and cough failed to contribute to lethality. Chest pain was the most important lethality marker for male patients in various age groups, whereas arthralgia was the main risk marker for females.

The interpretability approach allowed us to identify some clinical signs that decrease the risk of lethality, namely rhinorrhea and anosmia in female patients, and conjunctivitis and odynophagia in males; these signs could be related to some benign presentation of the disease. On the other hand, the finding that a more rapid attention increased the risk of lethality could be due to a cultural habit of the Mexican population to seek care immediately when the course of the disease is severe. This contrasts with the results reported by Mancilla-Galindo et al. that the risk of fatal outcome in Mexican patients with COVID-19 increased for each day of delay in receiving medical care after symptom onset, which also correlated with age [23], and by Martos et al. that longer periods elapsed from symptom onset to the start of medical care were associated with more adverse clinical outcomes, highlighting the importance of early medical consultation [62].

The use of antipyretics was a recurrent marker in males and, to a lesser extent, in females. Its risk-increasing effect on lethality could be explained by the fact that these drugs are used by patients as a default response to severe symptoms; therefore, these drugs are not a contributing variable per se, but could be indicating patient-perceived severity.

Limitations and contributions

Our study has strengths and limitations. Its main strength is the stratification of the population and the inclusion of demographic and clinical variables accessible to clinicians in resource-limited settings. This allowed us to identify risk markers and their contribution to COVID-19 lethality, assessing their relative weight according to sex and age group. Male sex is commonly assumed to be a determining condition for disease outcome, which may lead to underestimation of risk in females.

A key contribution of this work is the development of a model for early prediction of increased lethality risk using machine learning algorithms, the use of an interpretability approach to identify markers of lethality risk by sex and age, and of the SHapley Additive exPlanations (SHAP). By analyzing predictors of COVID-19 mortality risk by sex and age group, we were able to identify those segments with the highest vulnerability. In addition, including both outpatients and inpatients allowed us to capture the full spectrum of COVID-19 cases.

Our prediction model based on sociodemographic and clinical parameters could provide a tool to select patients at higher risk of a lethal outcome from COVID-19 in a first-contact setting, especially in geographic areas where laboratory infrastructure and hospital care are limited. However, extensive external validation studies are required to assess its performance in triage prior to its clinical use.

The main limitation of our study is the quality of the information available for some variables, and the fact that some responses are not specific; for example, the variable “cardiovascular disease” encompasses several conditions, often poorly delimited; self-reporting of comorbidities could lead to an underestimation of risk, especially in subclinical patients. Data were not updated as the disease evolved, and all recorded signs and comorbidities were reported by patients when they received medical care; in addition, data on SARS-CoV-2 virus variant in confirmed cases were also missing. Finally, the database was only reviewed and validated by the Mexican Ministry of Health.

We are aware that other general variables, easily accessible to clinicians, such as comorbidity control history, medication, and vital signs, could improve the performance of our prediction model and allow us to identify more accurate risk markers. Another limitation was that our models were not validated with an external cohort. Finally, this study was conducted in a Mexican population, so caution should be taken when generalizing its results to other populations with a different demographic and metabolic profile.

Conclusions

This study showed that age was the strongest demographic factor for a fatal outcome in both sexes. The risk of death from COVID-19 rises with age and begins to increase at an earlier age in males than in females. Age and sex set a baseline risk of death, which is higher in male patients, and this risk could increase or decrease, depending on other risk markers.

Signs of respiratory distress, such as dyspnea and polypnea, were the main predictive symptoms of COVID-19 lethality, showing a greater weight in males than in females.

Hypertension was the comorbidity with the highest contribution to lethality, irrespectively of sex. Chronic kidney disease was a major marker for males, whilst diabetes was a significant risk marker for females.

Some manifestations of the disease decreased the risk of lethality, including rhinorrhea, anosmia, conjunctivitis, and odynophagia, possibly because of their association with some benign presentation of the disease. In addition, the fact that a shorter delay in the provision of care increased the risk of lethality could be an indirect effect of the tendency of the Mexican population to seek immediate care when the course of the disease is more severe.

Herein, we demonstrate that machine learning-based models can identify risk markers for lethality by sex and age, which were consistent with the results of previous statistically based studies in the Mexican population. Finally, individual predictions could help improve our clinical understanding of COVID-19 care by providing an overview of possible outcomes for a patient based on their clinical history. Such markers have the potential to be used in triage and prognosis in the populations that lack access to highly specialized health services.

Currently, the critical stage of the COVID-19 pandemic is over. However, in order to prepare ourselves to respond to possible public health emergency of international concern (PHEIC), and considering the experience of COVID-19, it is crucial to design and implement research tools and methodologies for the next pandemic event. To our knowledge, this work is a first approach to ML-based clinical research in public health issues, and it could be valuable in a country like Mexico, where saturation of health services is the norm and where the use of ML-based approaches in public health is still incipient. We hope that this study will drive the use of machine learning approaches in future research in emergency situations, where the healthcare system requires the support of non-biomedical sciences, emphasizing the need for stratification by sex and age as a first step towards personalized care.

Availability of data and materials

The open-access database used in this article is available at http://covid-19.iimas.unam.mx/. The source code from all experiments is available at https://github.com/rojas-mariano-salvador/Lethality-markers-for-COVID-19.

Notes

  1. https://www.gob.mx/cms/uploads/attachment/file/636504/Estudio_epidemiol_gico_de_caso_sospechosos_de_enfermedad_respiratoria_viral.pdf

  2. https://scikit-learn.org/stable/.

  3. https://pandas.pydata.org/.

  4. https://numpy.org.

  5. https://jupyter.org/.

Abbreviations

DALYs:

Disability-adjusted life years

CFR:

Case fatality rates

COPD:

Chronic obstructive pulmonary disease

CKD:

Chronic kidney disease

ML:

Machine learning

AUC:

Area under the curve, receiver operating characteristics

LR:

Logistic regression

SVM:

Support vector machine

RF:

Random forest

XGBoost:

Extreme gradient boosting

SHAP:

SHapley Additive exPlanations

DT:

Decision tree

GBM:

Gradient boosting machine

LGBM:

Light gradient boosting machine

References

  1. Salinas-Escudero G, Toledano-Toledano F, García-Peña C, Parra-Rodríguez L, Granados-García V, Carrillo-Vega MF. Disability-adjusted life years for the COVID-19 pandemic in the Mexican population. Front Public Health. 2021;9(686700):1–9.

    Google Scholar 

  2. Statista. statista.com. 2022. https://www.statista.com/statistics/1093256/novel-coronavirus-2019ncov-deaths-worldwide-by-country/. Accessed 17 May 2022.

  3. World Health Organization. WHO Coronavirus (COVID-19) Dashboard. 2022. https://covid19.who.int/. Accessed 17 May 2022.

  4. Monterrubio-Flores E, Ramírez-Villalobos M, Espinosa-Montero J, Hernandez B, Barquera S, Villalobos-Daniel VE, et al. Characterizing a two-pronged epidemic in Mexico of non-communicable diseases andSARS-Cov-2: factors associated with increased case-fatality rates. Int J Epidemiol. 2021; 50(2):1–16.

  5. Gutierrez J, Bertozzi SM. Non-communicable diseases and inequalities increase risk of death among COVID-19 patients in Mexico. PLoS ONE. 2020;15(10):1–11.

    Article  Google Scholar 

  6. Knaul FM, Touchton M, Arreola-Ornelas H, Atun R, Calderon Anyosa R, Frenk J, et al. Punt politics as failure of health system stewardship: evidence from the COVID-19 pandemic response in Brazil and Mexico. Lancet Regional Health - Americas. 2021;4(100086):1–11.

    Google Scholar 

  7. Malik P, Patel U, Mehta D, Patel M, Kelkar R, Akrmah M, et al. Biomarkers and outcomes of COVID-19 hospitalisations: systematic review and meta-analysis. BMJ Evid-Based Med. 2021;26(3):107–8.

    Article  Google Scholar 

  8. Prado-Galbarro FJ, Sanchez-Piedra C, Gamiño-Arroyo AE, Cruz-Cruz C. Determinants of survival after severe acute respiratory syndrome coronavirus 2 infection in Mexican outpatients and hospitalised patients. Public Health. 2020;30(189):66–72.

    Article  Google Scholar 

  9. Domínguez-Olmedo JL, Gragera-Martínez , Mata J, Pachón Álvarez V. Machine learning applied to clinical laboratory data in Spain for COVID-19 outcome prediction: model development and validation. J Med Internet Res. 2021; 23(4): 1-11.

  10. Çubukçu HC, Topcu Dİ, Bayraktar N, Gülşen M, Sarı N, Arslan AH. Detection of COVID-19 by machine learning using routine laboratory tests. Am J Clin Pathol. 2021; 157(5): 758–66.

  11. Khuzani AZ, Heidari M, Shariati SA. COVID-Classifier: an automated machine learning model to assist in the diagnosis of COVID-19 infection in chest x-ray images. medRxiv. 2020; 2.

  12. Wynants L, Van Calster B, Collins GS, Riley RD, Heinze G, Schuit E, et al. Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal. BMJ. 2020; 369(1328): 1–16.

  13. Defazio A, Bach F, Lacoste-Julien S. SAGA: a fast incremental gradient method with support for non-strongly convex composite objectives. arXiv. 2014.

  14. Murphy KP. Machine learning: a probabilistic perspective. Cambridge: The MIT Press; 2012.

    Google Scholar 

  15. Scornet E. Tuning parameters in random forests. ESAIM: Procs. 2017; 60: 144–162.

  16. Budholiya K, Shrivastava K, Sharma V. An optimized XGBoost based diagnostic system for effective prediction of heart disease. J King Saud Univ Comput Inform Sci. 2022;34:4514–23.

    Google Scholar 

  17. Lundberg SM, Lee SI. A unified approach to interpreting model predictions. In: Guyon I, Luxburg UV, Bengio S, Wallach H, Fergus R, Vishwanathan S, et al., editors. Long Beach: Curran Associates, Inc.; 2017.

  18. Molnar C. Interpretable Machine Learning A Guide for Making Black Box Models Explainable. 2a ed. München, Germany: Independently published; 2022: 328.

  19. Thimoteo L, Vellasco MM, do Amaral JM, Figueiredo K, Yokoyama CL, Marques E. Interpretable machine learning for COVID-19 diagnosis through clinical variables. Soc Brasil Autom. 2020;2(1):1–8.

    Google Scholar 

  20. Yang R. Who dies from COVID-19? Post-hoc explanations of mortality prediction models using coalitional game theory, surrogate trees, and partial dependence plots. medRixv. 2020; 1–17.

  21. Vázquez B, Fuentes-Pineda G, García F, Borrayo G, Prohías J. Risk markers by sex for in-hospital mortality in patients with acute coronary syndrome: a machine learning approach. Inform Med Unlocke. 2021;27:1–13.

    Article  Google Scholar 

  22. Undela K, Gudi S. Assumptions for disparities in case-fatality rates of coronavirus disease (COVID-19) across the globe. Eur Rev Med Pharmacol Sci. 2020;24(9):5180–2.

    CAS  Google Scholar 

  23. Mancilla-Galindo J, Vera-Zertuche JM, Navarro-Cruz AR, Segura-Badilla O, Reyes-Velázquez G, Tepepa-López J, et al. Development and validation of the patient history COVID-19 (PH-Covid19) scoring system: a multivariable prediction model of death in Mexican patients with COVID-19. Epidemiol Infect. 2020;148:1–8.

    Article  Google Scholar 

  24. Bello-Chavolla OY, Antonio-Villa NE, Ortiz-Brizuela E, Vargas-Vázquez A, González-Lara MF, Ponce de Leon A, et al. Validation and repurposing of the MSLCOVID-19 score for prediction of severe. PLoS ONE. 2020;15(12):1–14.

    Google Scholar 

  25. Aktar S, Talukder A, Ahamad MM, Kamal AHM, Khan RJ, Liaw T, et al. Machine learning approaches to identify patient comorbidities and symptoms that increased risk of mortality in COVID-19. Diagnostics. 2021;11(8):1–18.

    Article  Google Scholar 

  26. Cini Oliveira M, de Araujo Eleuterio T, de Andrade Corrêa AB, Romanoda Silva LD, Coelho Rodrigues R, Andrade de Oliveira B, et al. Factors associated with death in confirmed cases of COVID-19 in the state of Rio de Janeiro. BMC Infect Dis. 2021;21(687):1–16.

    Google Scholar 

  27. Wang K, Zuo P, Liu Y, Zhang M, Zhao X, Xie S, et al. Clinical and laboratory predictors of in-hospital mortality in patients with COVID-19: a cohort study in Wuhan, China. Clin Infect Dis. 2020;71(16):2079–88.

    Article  CAS  Google Scholar 

  28. Chen Y, Ouyang L, Bao FS, Li Q, Han L, Zhu B, et al. An interpretable machine learning framework for accurate severe vs non-severe COVID-19 clinical type classification. medRxiv. 2020.

  29. Gong J, Ou J, Qiu X, Jie Y, Chen Y, Yuan L, et al. A tool to early predict severe corona virus disease 2019 (COVID-19): a multicenter study using the risk nomogram in Wuhan and Guangdong, China. Clin Infect Dis. 2020;71(15):833–40.

    Article  CAS  Google Scholar 

  30. Jiang X, Coffee M, Bari A, Wang J, Jiang X, Huang J, et al. Towards an artificial intelligence framework for data-driven prediction of coronavirus clinical severity. Comput MaterContinua. 2020;63(1):537–51.

    Google Scholar 

  31. Xie J, Hungerford D, Chen H, Abrams ST, Li S, Wang G, et al. Development and external validation of a prognostic multivariable model on admission for hospitalized patients with COVID-19. medRixv. 2020.

  32. Yan L, Zhang T, Goncalves J, Xiao Y, Wang M, Guo Y, et al. An interpretable mortality prediction model for COVID-19 patients. Nat Mach Intell. 2020;2:283–8.

    Article  Google Scholar 

  33. Rahman T, Al-Ishaq FA, Al-Mohannadi FS, Mubarak RS, Al-Hitmi MH, Islam KR, et al. Mortality prediction utilizing blood biomarkers to predict the severity of COVID-19 using machine learning technique. Diagnostics. 2021;11(9):1582.

    Article  CAS  Google Scholar 

  34. Kang J, Chen T, Luo H, Luo Y, Du G, Jiming-Yang M. Machine learning predictive model for severe COVID-19. Infect Genet Evol. 2021;90: 104737.

    Article  CAS  Google Scholar 

  35. De Souza FSH, Hojo-Souza NS, Dos Santos EB, Da Silva CM, Guidoni DL. Predicting the disease outcome in COVID-19 positive patients through machine learning: a retrospective cohort study with Brazilian data. Front Artif Intell. 2021;4:1–13.

    Google Scholar 

  36. Sardar R, Sharma A, Gupta D. Machine learning assisted prediction of prognostic biomarkers associated with COVID-19, using clinical and proteomics data. Front Genet. 2021;12:1–10.

    Article  Google Scholar 

  37. Karthikeyan A, Garg A, Vinod PK, Priyakumar UD. Machine learning based clinical decision support system for early COVID-19 mortality prediction. Front Public Health. 2021;9(626697):1–13.

    Google Scholar 

  38. Collins S, van Smeden M, Riley D. COVID-19 prediction models should adhere to methodological and reporting standards. Eur Respir J. 2020; 56(3): 1-4.

  39. Hooli S, King C. Generalizability of Coronavirus Disease 2019 (COVID-19) clinical prediction models. Clin Infect Dis. 2020;71(15):897.

    Article  CAS  Google Scholar 

  40. Zhang B, Zhou X, Qiu Y, Song Y, Feng F, Feng J, et al. Clinical characteristics of 82 cases of death from COVID-19. PLoS ONE. 2020;15(7):1–13.

    Article  CAS  Google Scholar 

  41. Weng Z, Chen Q, Li S, Li H, Zhang Q, Lu S, et al. ANDC: an early warning score to predict mortality risk for patients with Coronavirus Disease 2019. J Transl Med. 2020;18(328):1–10.

    Google Scholar 

  42. Chen R, Liang W, Jiang M, Guan W, Zhan C, Wang T, et al. Risk factors of fatal outcome in hospitalized subjects with coronavirus disease 2019 from a nationwide analysis in China. Chest. 2020;158(1):97–105.

    Article  CAS  Google Scholar 

  43. O’Driscoll M, Ribeiro Dos Santos G, Wang L, Cummings DAT, Azman AS, Paireau J, et al. Age-specific mortality and immunity patterns of SARS-CoV-2. Nature. 2020;590:140–5.

    Article  Google Scholar 

  44. COVID-19 Forecasting Team. Variation in the COVID-19 infection–fatality ratio by age, time, and geography during the pre-vaccine era: a systematic analysis. Lancet. 2022; 399(10334):1469–88.

  45. Parra-Bracamonte GM, Lopez-Villalobos N, Parra-Bracamonte FE. Clinical characteristics and risk factors for mortality of patients with COVID-19 in a large data set from Mexico. Ann Epidemiol. 2020;52:93–8.

    Article  Google Scholar 

  46. Mesta F, Coll AM, Ramírez MÁ, Delgado-Roche L. Predictors of mortality in hospitalized COVID-19 patients: a Mexican population-based cohort study. Biomedicine. 2021;11(2):1–4.

    Article  Google Scholar 

  47. Ursin RL, Klein SL. Sex differences in respiratory viral pathogenesis and treatments. Annu Rev Virol. 2021;8(1):393–414.

    Article  Google Scholar 

  48. Bello-Chavolla OY, Bahena-López JP, Antonio-Villa NE, Vargas-Vázquez A, González-Díaz A, Márquez-Salinas A, et al. Predicting mortality due to SARS-CoV-2: a mechanistic score relating obesity and diabetes to COVID-19 outcomes in Mexico. J Clin Endocrinol Metab. 2020;105(8):2752–61.

    Article  Google Scholar 

  49. Woolcott OO, Castilla-Bancayán JP. The effect of age on the association between diabetes and mortality in adult patients with COVID-19 in Mexico. Sci Rep. 2021;11(8386):1–10.

    Google Scholar 

  50. Williamson EJ, Walker AJ, Bhaskaran K, Bacon S, Bates C, Morton CE, et al. Factors associated with COVID-19-related death using OpenSAFELY. Nature. 2020;584:430–6.

    Article  CAS  Google Scholar 

  51. Gansevoort RT, Hilbrands LB. CKD is a key risk factor for COVID-19 mortality. Nat Rev Nephrol. 2020;16:705–6.

    Article  CAS  Google Scholar 

  52. Salinas-Aguirre JE, Sánchez-García C, Rodríguez-Sanchez R, Rodríguez-Munoz L, Díaz-Castano A, Bernal-Gómez R. Características clínicas y comorbilidades asociadas a mortalidad en pacientes con COVID-19 en Coahuila (México). Revista Clinica Espanola. 2021;222(5):288–92.

    Article  Google Scholar 

  53. Peña EDL, Rascón-Pacheco RA, Ascencio-Montiel IDJ, González-Figueroa E, Fernández-Gárate JE, Medina-Gómez OS, et al. Hypertension, diabetes and obesity, major risk factors for death in patients with COVID-19 in Mexico. Arch Med Res. 2021;52(4):443–9.

    Article  Google Scholar 

  54. Klonoff DC, Umpierrez GE. Letter to the Editor: COVID-19 in patients with diabetes: risk factors that increase morbidity. Metab, Clin Exp. 2020;108:1–2.

    Article  Google Scholar 

  55. Aghili SMM, Ebrahimpur M, Arjmand B, Shadman Z, Pejman Sani M, Qorbani M, et al. Obesity in COVID-19 era, implications for mechanisms, comorbidities, and prognosis: a review and meta-analysis. Int J Obes. 2021;45(5):998–1016.

    Article  CAS  Google Scholar 

  56. Simonnet A, Chetboun M, Poissy J, Raverdy V, Noulette J, Duhamel A, et al. High prevalence of obesity in severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) requiring invasive mechanical ventilation. Obesity. 2020;28(7):1195–9.

    Article  CAS  Google Scholar 

  57. Caussy C, Wallet F, Laville M, Disse E. Obesity is associated with severe forms of COVID-19. Obesity. 2020;28(7):1175.

    Article  CAS  Google Scholar 

  58. Vera-Zertuche JM, Mancilla-Galindo J, Tlalpa-Prisco M, Aguilar-Alonso P, Aguirre-García MM, Segura-Badilla O, et al. Obesity is a strong risk factor for short-term mortality and adverse outcomes in Mexican patients with COVID-19: a national observational study. Epidemiol Infect. 2021;149(e109):1–11.

    Google Scholar 

  59. Henry BM, Vikse J, Benoit S, Favaloro EJ, Lippi G. Hyperinflammation and derangement of renin-angiotensin-aldosterone system in COVID-19: a novel hypothesis for clinically suspected hypercoagulopathy and microvascular immunothrombosis. Clin Chim Acta. 2020;507:167–73.

    Article  CAS  Google Scholar 

  60. Caci G, Albini A, Malerba M, Noonan DM, Pochetti P, Polosa R. COVID-19 and obesity: dangerous liaisons. J Clin Med. 2020;9(8):2511.

    Article  CAS  Google Scholar 

  61. Martínez-Martínez MU, Alpízar-Rodríguez D, Flores-Ramírez R, Portales-Pérez DP, Soria-Guerra R, Pérez-Vázquez F, et al. An analysis COVID-19 in Mexico: a prediction of severity. J Gen Intern Med. 2022;37:624–31.

    Article  Google Scholar 

  62. Martos-Benítez FD, Soler-Morejón CD, García-del BD. Chronic comorbidities and clinical outcomes in patients with and without COVID-19: a large population-based study using national administrative healthcare open data of Mexico. Intern Emerg Med. 2021;16(6):1507–17.

    Article  Google Scholar 

  63. Hernández-Galdamez DR, González-Block MÁ, Romo-Dueñas DK, Lima-Morales R, Hernández-Vicente IA, Lumbreras-Guzmán M, et al. Increased risk of hospitalization and death in patients with COVID-19 and pre-existing noncommunicable diseases and modifiable risk factors in Mexico. Arch Med Res. 2020;51(7):683–9.

    Article  Google Scholar 

  64. Barquera S, Hernández-Barrera L, Trejo-Valdivia B, Shamah T, Campos-Nonato I, Rivera-Dommarco J. Obesity in Mexico, prevalence and trends in adults. Ensanut 2018–19. Salud Pública México. 2020;62(6):682–92.

    Article  Google Scholar 

  65. Basto-Abreu AC, López-Olmedo N, Rojas-Martínez R, Aguilar-Salinas CA, De la Cruz-Góngora VV, Rivera-Dommarco J, et al. Prevalence of diabetes and glycemic control in Mexico: national results from 2018 and 2020. Salud Publica México. 2021;63(6):725–33.

    Article  Google Scholar 

  66. Di Castelnuovo A, Bonaccio M, Costanzo S, Gialluisi A, Antinori A, Berselli N, Blandi L, Bruno R, Cauda R, Guaraldi G, My I, Menicanti L, Parruti G, Patti G, Perlini S, Santilli F, Signorelli C, Stefanini GG, Vergori A, Abdeddaim A, Ageno W, Agodi A, Agostoni P, Aiello L, Al Moghazi S, Aucella F, Barbieri G, Bartoloni A, Bologna C, Bonfanti P, Brancati S, Cacciatore F, Caiano L, Cannata F, Carrozzi L, Cascio A, Cingolani A, Cipollone F, Colomba C, Crisetti A, Crosta F, Danzi GB, D’Ardes D, de Gaetano Donati K, Di Gennaro F, Di Palma G, Di Tano G, Fantoni M, Filippini T, Fioretto P, Fusco FM, Gentile I, Grisafi L, Guarnieri G, Landi F, Larizza G, Leone A, Maccagni G, Maccarella S, Mapelli M, Maragna R, Marcucci R, Maresca G, Marotta C, Marra L, Mastroianni F, Mengozzi A, Menichetti F, Milic J, Murri R, Montineri A, Mussinelli R, Mussini C, Musso M, Odone A, Olivieri M, Pasi E, Petri F, Pinchera B, Pivato CA, Pizzi R, Poletti V, Raffaelli F, Ravaglia C, Righetti G, Rognoni A, Rossato M, Rossi M, Sabena A, Salinaro F, Sangiovanni V, Sanrocco C, Scarafino A, Scorzolini L, Sgariglia R, Simeone PG, Spinoni E, Torti C, Trecarichi EM, Vezzani F, Veronesi G, Vettor R, Vianello A, Vinceti M, De Caterina R, Iacoviello L. Common cardiovascular risk factors and in-hospital mortality in 3894 patients with COVID-19: survival analysis and machine learning-based findings from the multicentre Italian CORIST Study. Nutr Metab Cardiovasc Dis. 2020;30(11):1899–913. https://doi.org/10.1016/j.numecd.2020.07.031.

    Article  CAS  Google Scholar 

  67. Spinoni EG, Mennuni M, Rognoni A, Grisafi L, Colombo C, Lio V, Renda G, Foglietta M, Petrilli I, D’Ardes D, Sainaghi PP, Aimaretti G, Bellan M, Castello L, Avanzi A, Corte FD, Krengli M, Pirisi M, Malerba M, Capponi A, Gallina S, Pierdomenico SD, Cipollone F, Patti G, Albano E, Dianzani U, Gaidano G, Gennari A, Gramaglia C, Solli M, Giubertoni A, Veia A, Cisari C, Paolo AT, Valletti PA, Adesi FB, Barini M, Ferrante D, De Vecchi S, Santagostino M, Acquaviva A, Calzaducca E, Casciaro FG, Ceruti F, Cittone MG, Di Benedetto D, Gagliardi I, Giacomini GM, Landi IC, Landi R, Manfredi GF, Pedrinelli AR, Rigamonti C, Rizzi E, Smirne C, Vassia V, Arioli R, Danna P, Falaschi Z, Paschè A, Percivale I, Zagaria D, Beltrame M, Bertoli M, Galbiati A, Gardino CA, Gastaldello ML, Via VG, Giolitti F, Inserra I, Labella E, Nerici I, Gironi LC, Cammarata E, Esposto E, Tarantino V, Zavattaro E, Zottarelli F, Daffara T, Ferrero A, Leone I, Nuzzo A, Baldon G, Battistini S, Chirico E, Lorenzini L, Martelli M, Barbero E, Boffano P, Brucoli M, Garzaro M, Pau A, Bertolin S, Marzari L, Avino G, Saraceno M, Morosini U, Baricich A, Invernizzi M, Gallo S, Montabone C, Padelli SA, Boglione L, Patrucco F, Salamina L, Baorda F, Croce E, Giacone I. Contribution of Atrial Fibrillation to In-Hospital Mortality in Patients With COVID-19. Circ Arrhythm Electrophysiol. 2021;14(2):e009375. https://doi.org/10.1161/CIRCEP.120.009375.

    Article  Google Scholar 

Download references

Funding

This work was supported by the Instituto Nacional de Salud Pública, Mexico, and grants from the Consejo Nacional de Ciencia y Tecnología (CONACYT, Mexico) to VMM (Grant F0005-2020-01 COVID-19,#313033).

Author information

Authors and Affiliations

Authors

Contributions

MRG, BV, KTP, and VMM conceived the idea and participated extensively in parameter identification, review of the article, data extraction, quality assessment, analysis, drafting, and revision of the manuscript. All authors also agreed to be equally responsible for all aspects of this research work. All authors read and approved the final version of the manuscript.

Corresponding authors

Correspondence to Kirvis Torres-Poveda or Vicente Madrid-Marina.

Ethics declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1.

Description of clinical variables used in the experimentation.

Additional file 2.

Lethality prediction performance.

Additional file 3.

Risk markers by sex and age group.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Rojas-García, M., Vázquez, B., Torres-Poveda, K. et al. Lethality risk markers by sex and age-group for COVID-19 in Mexico: a cross-sectional study based on machine learning approach. BMC Infect Dis 23, 18 (2023). https://doi.org/10.1186/s12879-022-07951-w

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12879-022-07951-w

Keywords

  • COVID-19
  • Lethality risk markers
  • Machine learning
  • Mexico