Deepsting: Systems medicine and systems epidemiology models

Introduction

The dramatic advances in biotechnology and computational sciences have catalyzed the emergence of interdisciplinary approaches for the study of human health and disease. Two prominent examples are systems medicine and epidemiology. Systems medicine extends the principles of systems biology into the realm of clinical practice, creating models that reflect complex interactions and personalized disease mechanisms (Auffray et al., 2009; Bousquet et al., 2011; System et al., 2005). Systems epidemiology is an interdisciplinary framework that integrates traditional epidemiologic methods with complex systems science in order to understand, model, and intervene in multifactorial health phenomena spanning from the molecular level to populations and ecosystems (Dammann et al., 2014; Joffe et al., 2012). Although both fields address complexity in human health, they operate at distinct scales and complement each other toward improved diagnostic strategies, therapeutic innovations, and health policy development. In systems epidemiology and systems medicine, mathematical modeling captures the inherent temporal evolution and multifactorial risk structures of diseases, thereby supporting decisions on resource allocation and intervention design (Angione, 2019; Dammann et al., 2014; Flores et al., 2013; Joffe et al., 2012).

Modern models of healthcare have grown to be critical of the traditional reactive approach to medical care. Focusing on curative healthcare as the basis for medical systems can lead to preventable deaths, worst outcomes for certain diseases, overloaded medical institutions, and a disaffected population. Systems medicine, and its prediction, prevention, personalized, participatory medicine (4P medicine) has emerged to offer a sustainable, inclusive, and systemic alternative to reactive care (Ayers & Day, 2015; Baiardini & Heffler, 2019; Flores et al., 2013; Slim et al., 2021).

While preventive care has been emphasized as an important component of scorpionism in Mexico, a comprehensive systemic approach has not been fully implemented (Abroug et al., 2020; Ahmadi et al., 2020; Boyer, 2013). This stems from both an ongoing effort to fight sting related deaths that has not been able to fully address more modern approaches, and from existing limitations on the Mexican healthcare system. In recent years, new information technologies have enabled the capture of new epidemiological data, that allows for a broader understating of the determinant of scorpion envenomation. At the same time new efforts for systemic healthcare models for scorpion envenomation have emerged (Chippaux et al., 2020; Hernández-Muñoz et al., 2024; Trinidad-Porfirio et al., 2023). Gaps remain in connecting the necessities of local communities with centralized medical institutions, in ensuring universal coverage of marginalized populations, and in processing the massive amounts of data generated by new epidemiological systems.

The advent of deep learning (DL) has catalyzed transformative changes in medical practice, particularly within the framework of 4P medicine. The utility of DL lies in its unique ability to automatically extract intricate features from vast and diverse datasets, and to learn complex non‐linear relationships that are often inaccessible to traditional statistical methods (Angermueller et al., 2016; Farmer et al., 2019; Piccialli et al., 2021). DL in healthcare has been extensively used for preventive and predictive models focused on chronic diseases, mainly cancer. DL allows both to utilize the large collections of cancer biomarkers to build individualized models of risks based on genomic data from patients, and allows to develop diagnostic aid based on medical imaging (Ching et al., 2018; Mahdi-Esferizi et al., 2023; Piccialli et al., 2021). DL has also been used to model behaviour of infections to decide where to allocate resources and guide public health policy (Ahmadini et al., 2025).

As healthcare systems worldwide confront challenges associated with preventable injuries and accident‐related deaths, the integration of DL techniques enhances clinical decision-making, risk estimation, and real‐time monitoring. These methodologies facilitate proactive management of occupational hazards by tailoring interventions based on personalized risk profiles, and further support participatory care by integrating patient feedback via clinical decision support systems (CDSS) and patient portals (Boffetta & Collatuzzo, 2022; Madden, 2024).

In Mexico, adoption of DL and machine learning (ML) tools for healthcare have not been widely adopted in Mexico. The 4Ps model of systems medicine has been favored in recent years, but challenges remain. Adoption necessitates non only the normalization of new technologies, but fundamental changes in the way medical care is delivered (Rodríguez Weber et al., 2022; Torre-Bouscoulet, 2015). A common criticism for predictive and personalized medicine in Mexico and other developing countries is that the mass data necessary for accurate pre diagnosis and proactive care is materially impossible in the coming years (Baba et al., 2015; Nohara et al., 2015; Teo & Rafiq, 2021). These limitations have moved the objective of 4Ps medicine in developing countries from relying on Omics data, to study risk factors that can be analyzed and gathered at community level. Additionally, focused has also expanded to local prevailing diseases (de Magalhães Brito & Magagna, 2020; Slim et al., 2021; Teo & Rafiq, 2021).

Mexico is the country with the largest number of scorpion stings each year, it accounts for around ¼ of the global 1.2 million cases registered yearly (Hernández-Muñoz et al., 2024; Riano-Umbarila et al., 2025; Trinidad-Porfirio et al., 2023). In the 1990s, Mexico invested significant resources into preventing and treating scorpion stings and successfully was able to reduce mortality caused by these events. This effort was a combination of new, effective, and safe antivenoms derived from immunized horses, and new public measures seeking to prevent scorpion stings and guarantee access to treatment (Abroug et al., 2020; Boyer, 2013; Riano-Umbarila et al., 2025). In recent years, scorpion envenomation has been on the rise, multiple institutions, public and private seek to address the issue. As part of these efforts, new epidemiological data systems have been deployed to better capture the context around diseases and their treatment. Opportunities remain for systems that coordinate centralized production and information with needs of local communities and use new information to integrate into a 4P framework (Chippaux et al., 2020; Hernández-Muñoz et al., 2024; Riano-Umbarila et al., 2025; Trinidad-Porfirio et al., 2023).

In this section we present a DL based analysis of scorpion epidemiological data. We present 3 types of models: an LSTM memory model for a time series analysis to predict scorpion envenomation rates depending on social and environmental factors, a feedforward MLP neural network for classification of at-risk factors, and an autoencoder plus clustering to generate group classifications of at-risk communities.

Scorpion sting epidemiology

Each year around 1.2 million cases of scorpion sting envenomation (SSE) take place, with more than 2 billion humans considered to live in areas at risk of envenomation. Mexico suffers around 25% of all cases of SSE (Hernández-Muñoz et al., 2024; Kumar et al., 2023; Riano-Umbarila et al., 2025). One of the most widely recognized approaches in the epidemiological classification of scorpion sting envenomation is the severity-based system. Multiple studies have elucidated that the severity of envenomation can be categorized into three principal classes: mild, moderate, and severe. Mild envenomation is generally characterized by localized symptoms such as pain, hyperemia, and edema, with minimal systemic involvement. Moderate envenomation involves the emergence of systemic signs: nausea, tachycardia, and other manifestations, that suggest a more diffuse impact of the venom. Severe envenomation, which occurs in a minority of cases, is distinguished by life-threatening symptoms, including significant cardiovascular compromise and neurological disturbances, and is associated with a markedly increased risk of mortality, particularly among vulnerable populations such as children (Hernández-Muñoz et al., 2024; Kumar et al., 2023; Trinidad-Porfirio et al., 2023). From a prevention perspective, stings are regarded similarly to accidents. They primarily happen by contact between animals taking shelter in human spaces (in many cases because of the expansion of human spaces into scorpions’ natural habitats), and unaware humans. Shoes, clothing, dirt-roads, and furniture make frequent accommodation for these animals. In interactions between humans and scorpions, the latter feel threatened, and react by sting (Abd El-Aziz et al., 2019; Hernández-Muñoz et al., 2024; Trinidad-Porfirio et al., 2023).

Methodology

Data retrieval

Mexico has a centralized government office in charge of processing, updating, and publishing epidemiological data of important diseases in the nation, this office is known as the general direction of epidemiology (DGE for its Spanish abbreviation). It has a free and open access weekly publication known as the epidemiological bulletin. This publication, as well as state level publications (which are published by state level offices equivalent to the DGE) derived their data from the national system of epidemiological surveillance (SINAVES). The system is only accessible to medical professionals, and they can only retrieve information about their the local area. For the LSTM model we derived our data of SSE occurrence directly from the bulletin, which contains weekly reports of scorpion envenomation at state levels. Social and environmental data were retrieved from the national institute of statistics and geography and from the national commission of water respectively. Both were organized into a monthly time scale using data from 2019 to 2024.

For the other 2 models we contacted the DGE directly and they gave us access to a new database of information, which contains data for 2025 per municipality. This data set has more detailed information on the specifics of each person that suffered from SSE, like age, weight, gender, activity at the time of accident, among other things.

Models’ construction

All coding was done using MATLAB R2023. Because of data availability the LSTM model was built at the scale of states, and the other models at the scale of municipalities, although these also include internal differentiation by localities.

LSTM model

Long Short-Term Memory (LSTM) models are a specialized class of recurrent neural network (RNN) architectures developed to address the intrinsic limitations of traditional RNNs when modeling sequential and time‐dependent data, in particular the notorious vanishing and exploding gradient problems that impair learning over long sequences (Graves, 2012; Hua et al., 2019; Lindemann et al., 2021; Song et al., 2020). Time series prediction is the process of forecasting future values based on historically observed data points arranged in time order, and it plays a crucial role in diverse disciplines such as finance, weather forecasting, resource planning, and beyond (Hua et al., 2019; Lindemann et al., 2021). It is generally accepted that SSE has great seasonal dependence, since the animals are more common during hot and humid months. At the same time, there is evidence to show that social factors like living in rural areas or poverty also has an impact on the risk of suffering from SSE (Abd El-Aziz et al., 2019; Riano-Umbarila et al., 2025; Trinidad-Porfirio et al., 2023). The LSTM model we built has the objective of predicting the incidence of SSE in 4 states of Mexico by training on social and environmental data and its correlation to SSE.

We built our LSTM model to predict cases of scorpion envenomation based on 2 groups of inputs, social and environmental. The model works at state level for 4 Mexican states with high historic incidence of SSE: Morelos, Guerrero, Guanajuato, and Jalisco. Since the model involves multiples states they are inserted into MATLAB as separate time series samples, within each states time series 80% of entries are used for training and 20% for validation. We had 9 input variables: Average temperature, minimum temperature, maximum temperature, precipitation, education index, health index, income index, human development index, and population. The training options used were as follows: Adam for the optimizer, initial learning rate 0.005, up to 1000 training epochs, and gradient threshold 1.

MLP and autoencoder plus clustering models

Multilayer perceptron (MLP) models are a feedforward artificial neural network (ANN) characterized by its layered architecture that facilitates the approximation of complex nonlinear functions. An MLP is composed of an input layer, one or more hidden layers, and an output layer, with each layer containing interconnected neurons that perform weighted summation and nonlinear transformation via activation function (Ahmed et al., 2020; Angermueller et al., 2016; Ching et al., 2018). Autoencoder plus clustering models are integrated frameworks that combine the unsupervised representation learning capabilities of autoencoders with clustering algorithms to yield robust and discriminative grouping of high‐dimensional data. These models are designed to learn a compact latent-space representation through an encoder–decoder architecture while simultaneously or sequentially performing clustering tasks on the learned embedding, thereby bridging the gap between feature extraction and data grouping (Angermueller et al., 2016; Choi et al., 2020; Heilbroner & Miotto, 2023; Tang et al., 2019).

For supervised classification, we trained multilayer perceptron (MLP) models to predict each of several outcomes separately: activity during sting, municipality of sting, place of sting, and scorpion location. Each MLP consisted of an input layer concatenating numeric feature inputs and embeddings for categorical inputs, followed by two hidden fully connected layers of sizes 128 and 64 with ReLU activations, dropout rate of 0.3, and a softmax output layer. Learning was via the Adam optimizer with learning rate 1e-3, batch size 64, for up to 30 epochs, using early stopping based on validation loss.

For unsupervised representation learning the encoder compresses the input (concatenated numeric features and categorical embeddings) into a latent representation of dimension L (we used L = 20). The decoder reconstructs all input features. To encourage robustness, we trained the autoencoder with mean square error loss; for categorical variables, we reconstructed one-hot encodings (or used cross-entropy loss over categories) where feasible. After training for 50 epochs with an L₂ regularization weight of 0.001, we extracted latent codes for all samples. We then applied k-means clustering with k varying from 2 to 10. The optimal k was selected based on silhouette score and interpretability of clusters. To visualize the latent space, we used t-SNE to reduce to 2 dimensions and colored points by cluster label.

Results & discussion

LSTM model

The LSTM model ran successfully on all 4 states, but results differ by state. The best prediction performance was for the states of Guanajuato and Morelos, with Jalisco closely behind, Guerrero had the worst performance by a large margin. 2 main metrics were used to evaluate performance R2 and mean absolute percentage error (MAPE), for the first one a value closer to 1 is desirable, and closer to 0 for the second one. Scores can be seen in Table 1.

In addition to the scores the model also provides graphs comparing predictions with actual SSE occurrences. In accordance with the scores, while predictions generally follow the trends of occurrence, and have high overlap in some months, they differ significantly in others. Figure 1 has the 4 prediction graphs. Using state level data to generate the model means concentrating a diverse set of local conditions in a single geographic representation. All these states experience different weather conditions depending on the specific region. In a similar manner, social indicators vary greatly by intrastate region. This is particularly exacerbated in Guerrero, the second poorest state in Mexico (after the southern state of Chiapas). While local models may perform better, according to the DGE, the implementation of new epidemiological surveillance systems for SSE is recent, and data at the municipal or colony level is only available for 2025. While these models are limited in the representation of SSE, it shows that machine learning predictions can have adequate performance, even with a less than ideal dataset. It is possible that a more precise future model will be capable of full prediction of SSE at the community level. Even at the state level, prediction was able to mostly reflect the monthly seasonality of SSE, missing on the specific amount. This corresponds with agreed upon consensus among experts that SSE envenomation rates vary by season. While prediction does not reflect the other consensus, that poverty and rural communities are more at risk, this is more likely a miss characterization resulting from the scale of the data used (Abd El-Aziz et al., 2019; Chippaux et al., 2020; Hernández-Muñoz et al., 2024; Riano-Umbarila et al., 2025; Trinidad-Porfirio et al., 2023). As more and better data becomes available, it may lead to better predictive medicine in the future.

Figure 1. Prediction graphs of the LSTM model. Actual stings in solid blue line, predicted strings in red dotted line.

MLP and autoencoder plus clustering models

Both models were run in a single pipeline for the same 4 states used for the LSTM, and at municipal scale for some selected municipalities (Yautepec, Leon, Acapulco, and Zapopan). The objective of this pipeline is to use the MLP model to find how determinant are certain characteristics like the location of the scorpion, and the activity of the patient in SSE eventualities, and to use the clustering model to generate groups with similar profiles that represent general trends in at risk populations. For determinants of sting envenomation, the biggest ones were: Place of the person at the time of sting (97.01±1.2), Locality of sting (96.29±0.97), location of person at the time of sting (91.78±1.7), activity of person at time of sting (90.80±1.6). These are averages for all states and municipalities, and correlate with existing understanding. SSE is localized in certain communities, within those communities, stings happen by accident, as scorpions take place inside human environments like houses or places of work. Humans accidentally rattle the animal causing a sting. These are known factors in existing prevention campaigns, as they focus on checking clothing and shows for potential scorpions (Abroug et al., 2020; Hernández-Muñoz et al., 2024; Riano-Umbarila et al., 2025).

The results for the autoenconding and clustering models were a series of clusters, represented in figures 2 & 3, of groups with similar characteristics that have suffered scorpion envenomation. These groups are specific to each location (state or municipality), but general trends emerge. The most common location for SSE is peoples’ houses, with the second being their jobs, and third roads. The primary group stung at their jobs are middle-aged men, many in rural or semirural communities that work in fields and get stung when handling furniture, tools or wood where scorpions reside. This group, which is differentiated in multiple datasets, is the most easily differentiated group, represented as the blue cluster in Morelos, green in Jalisco, and blue in Guanajuato and orange in Guerrero. This coincides with reports that in rural areas, it is more common for men to be stung working in the fields, while women are stung during domestic work. 2 states show differentiation of women doing domestic work, unlike men, they are not a single age group but vary between young and old this group is represented as orange in Morelos and purple in Jalisco (Chippaux et al., 2020; Trinidad-Porfirio et al., 2023). After this group, cluster become varied. As mentioned, the most common place to suffer SSE are homes, with the most common activities being working and resting. At the municipal level differentiation is mainly based on age differences. Zapopan for example has highly marked clusters, with the orange group representing babies and toddlers (under 4 years old), and the yellow group representing children (around 9 years old). The first resting inside their homes, and the second doing different activities, including studying, playing, and resting, both indoors and outdoors. Kids also appear as a vulnerable group in the other municipalities: orange in Leon, green in Acapulco, and purple in Yautepec. Other 2 groups differentiated at municipal level are middle-aged adults (in this instance not differentiated by gender), and old teenagers/young adults (between 17 and 28 years old). For the first group rest and working are the most common activities, while studying, work, resting, and leisure activities were the most common for the latter.

At state level clustering was more varied, with more instances of difference between areas with historical incidence of SSE and more differences between rural and urban areas. While age did affect clustering at this scale, it was not as high a determinant factor as with municipalities. Morelos was the state with the highest differentiation by age, it follows a similar trend of conditions to that of the municipalities, it also is the one state where municipalities themselves play a lesser role in determining clusters, as most stings occur in Yautepec and Jojutla. Trends across al states and municipalities correspond to historical warnings of health authorities. Scorpion stings are often found inside peoples homes in dark and warm places, like closets, shoes, and drawers. Reinforcing the idea that prevention campaigns need to focus on checking this places, before touching them (Chippaux et al., 2020; Hernández-Muñoz et al., 2024; Trinidad-Porfirio et al., 2023).

Figure 2. Municipal level clustering for selected municipalities with high incidence of SSE.

Figure 3. State level clustering for the 4 states used in the study.

Modeling towards predictive and preventive medicine

The information derived from the models corresponds to general understanding of risk factors in SSE. The advantage of using models like this is that it allows to document local trends, like the differentiation of field and domestic workers in some states. This knowledge can be utilized to design communal based prevention, focusing efforts on local areas where these trends are present (Baba et al., 2015; Hernández-Muñoz et al., 2024). 4Ps systems should focus not only on information derived from these models but also on a bottom-up approach. Epidemiological datasets cannot often capture the reality of some of these communities. Within them there is a lack of understanding of scorpion envenomation across multiple age groups. Many individuals, especially older ones do not know how antivenoms work, some think they are like vaccines (only requiring a single dose for like that makes you immune to future stings, thus they avoid medical care in case of stings in the future), some don’t know the difference between new antivenoms and the older, unsafe serums. Children and youth in general are not instructed on how to deal with a scorpion in case they detect one. Rural communities sometimes don’t have access to medical care and lack access to information on scorpions and antivenoms. This is reported in studies, but also by both the DGE, and Redtox (a local network of experts on animal envenomation) (Chippaux et al., 2020; Hernández-Muñoz et al., 2024).

For the moment we are in discussions with stakeholders (DGE, local communities, Redtox, and general physicians) on how public and private health entities can properly use this data to better the lives of local communities, while allowing them to be decision makers in their own health.

Conclusions

Deep learning models allow for data analysis at different scales in the context of SSE in Mexico. LSTM models are able to predict general seasonal trends of scorpion stings at state level. While MLP and clustering models can revel at risk groups. This data can currently allow for better design of predictive and preventive medicine. However, for maximum effect they should be integrated into models that also consider participatory and personalized medicine within possible constraints. Information like the one derived from these models should be used to actively engage with communities and local health providers and jointly develop strategies on how to address local issues. As more data becomes available the precision and prediction capabilities of these models will grow. Additionally, they are a possible way to implement community level 4Ps medicine without depending on complex molecular data traditionally used for predictive and preventive medicine.

On data and software availability

All code used for this section is available in the software section of the wiki under the appropriate label. Data for the LSTM model is also available in that tab, since it is an aggregate of publicly available information. Data for the other 2 models can be obtained by directly asking the corresponding sub office of the DGE (the office of non-transmissible diseases) https://www.gob.mx/salud/acciones-y-programas/directorio-y-ubicacion-de-la-dge

Bibliography

• Abd El-Aziz, F. E.-Z. A., El Shehaby, D. M., Elghazally, S. A., & Hetta, H. F. (2019). Toxicological and epidemiological studies of scorpion sting cases and morphological characterization of scorpions (Leiurusquin questriatus and Androctonus crassicauda) in Luxor, Egypt. Toxicology Reports, 6, 329–335. https://doi.org/10.1016/j.toxrep.2019.03.004
• Abroug, F., Ouanes-Besbes, L., Tilouche, N., & Elatrous, S. (2020). Scorpion envenomation: State of the art. Intensive Care Medicine, 46(3), 401–410. https://doi.org/10.1007/s00134-020-05924-8
• Ahmadi, S., Knerr, J. M., Argemi, L., Bordon, K. C. F., Pucca, M. B., Cerni, F. A., Arantes, E. C., Çalışkan, F., & Laustsen, A. H. (2020). Scorpion Venom: Detriments and Benefits. Biomedicines, 8(5), 118. https://doi.org/10.3390/biomedicines8050118
• Ahmadini, A. A. H., Raghav, Y. S., Mahnashi, A. M., Islam Rather, K. U., & Ali, I. (2025). Neural networks to model COVID-19 dynamics and allocate healthcare resources. Scientific Reports, 15(1), 15326. https://doi.org/10.1038/s41598-025-00153-9
• Ahmed, Z., Mohamed, K., Zeeshan, S., & Dong, X. (2020). Artificial intelligence with multi-functional machine learning platform development for better healthcare and precision medicine. Database, 2020, baaa010. https://doi.org/10.1093/database/baaa010
• Angermueller, C., Pärnamaa, T., Parts, L., & Stegle, O. (2016). Deep learning for computational biology. Molecular Systems Biology, 12(7), 878. https://doi.org/10.15252/msb.20156651
• Angione, C. (2019). Human Systems Biology and Metabolic Modelling: A Review—From Disease Metabolism to Precision Medicine. BioMed Research International, 2019(1), 8304260. https://doi.org/10.1155/2019/8304260
• Auffray, C., Chen, Z., & Hood, L. (2009). Systems medicine: The future of medical genomics and healthcare. Genome Medicine, 1(1), 2. https://doi.org/10.1186/gm2
• Ayers, D., & Day, P. J. (2015). Systems Medicine: The Application of Systems Biology Approaches for Modern Medical Research and Drug Development. Molecular Biology International, 2015, 1–8. https://doi.org/10.1155/2015/698169
• Baba, Y., Kashima, H., Nohara, Y., Kai, E., Ghosh, P., Islam, R., Ahmed, A., Kuroda, M., Inoue, S., Hiramatsu, T., Kimura, M., Shimizu, S., Kobayashi, K., Tsuda, K., Sugiyama, M., Blondel, M., Ueda, N., Kitsuregawa, M., & Nakashima, N. (2015). Predictive Approaches for Low-Cost Preventive Medicine Program in Developing Countries. Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 1681–1690. https://doi.org/10.1145/2783258.2788587
• Baiardini, I., & Heffler, E. (2019). The patient-centered decision system as per the 4Ps of precision medicine. In Implementing precision medicine in best practices of chronic airway diseases (pp. 147–151). Elsevier. https://www.sciencedirect.com/science/article/pii/B9780128134719000244
• Boffetta, P., & Collatuzzo, G. (2022). Application of P4 (Predictive, Preventive, Personalized, Participatory) Approach to Occupational Medicine. La Medicina Del Lavoro, 113(1), e2022009–e2022009. https://doi.org/10.23749/mdl.v113i1.12622
• Bousquet, J., Anto, J. M., Sterk, P. J., Adcock, I. M., Chung, K., Roca, J., Agusti, A., Brightling, C., Cambon-Thomsen, A., Cesario, A., Abdelhak, S., Antonarakis, S. E., Avignon, A., Ballabio, A., Baraldi, E., Baranov, A., Bieber, T., Bockaert, J., Brahmachari, S., … Auffray, C. (2011). Systems medicine and integrated care to combat chronic noncommunicable diseases. Genome Medicine, 3(7), 43. https://doi.org/10.1186/gm259
• Boyer, L. (2013). History of scorpion antivenom: One Arizonan’s view. Toxicon, 69, 14–20. https://doi.org/10.1016/j.toxicon.2012.12.015
• Ching, T., Himmelstein, D. S., Beaulieu-Jones, B. K., Kalinin, A. A., Do, B. T., Way, G. P., Ferrero, E., Agapow, P.-M., Zietz, M., Hoffman, M. M., Xie, W., Rosen, G. L., Lengerich, B. J., Israeli, J., Lanchantin, J., Woloszynek, S., Carpenter, A. E., Shrikumar, A., Xu, J., … Greene, C. S. (2018). Opportunities and obstacles for deep learning in biology and medicine. Journal of The Royal Society Interface, 15(141), 20170387. https://doi.org/10.1098/rsif.2017.0387
• Chippaux, J.-P., Celis, A., Boyer, L., & Alagón, A. (2020). Factors involved in the resilience of incidence and decrease of mortality from scorpion stings in Mexico. Toxicon, 188, 65–75. https://doi.org/10.1016/j.toxicon.2020.10.011
• Choi, R. Y., Coyner, A. S., Kalpathy-Cramer, J., Chiang, M. F., & Campbell, J. P. (2020). Introduction to machine learning, neural networks, and deep learning. Translational Vision Science & Technology, 9(2), 14–14.
• Dammann, O., Gray, P., Gressens, P., Wolkenhauer, O., & Leviton, A. (2014). Systems Epidemiology: What’s in a Name? Online Journal of Public Health Informatics, 6(3). https://doi.org/10.5210/ojphi.v6i3.5571
• de Magalhães Brito, L. F., & Magagna, D. (2020). Chapter 96—Infection prevention & control—A clinical engineering perspective. In E. Iadanza (Ed.), Clinical Engineering Handbook (Second Edition) (pp. 682–693). Academic Press. https://doi.org/10.1016/B978-0-12-813467-2.00097-3
• Farmer, P. E., Nizeye, B., Stulac, S., & Keshavjee, S. (2019). Structural Violence and Clinical Medicine. In J. Oberlander, M. Buchbinder, L. R. Churchill, S. E. Estroff, N. M. P. King, B. F. Saunders, R. P. Strauss, & R. L. Walker (Eds), The Social Medicine Reader, Volume II, Third Edition: Differences and Inequalities, Volume 2 (pp. 156–169). Duke University Press. https://www.degruyterbrill.com/document/doi/10.1515/9781478004363-023/html
• Flores, M., Glusman, G., Brogaard, K., Price, N. D., & Hood, L. (2013). P4 Medicine: How Systems Medicine Will Transform the Healthcare Sector and Society. Personalized Medicine, 10(6), 565–576. https://doi.org/10.2217/pme.13.57
• Graves, A. (2012). Long Short-Term Memory. In A. Graves, Supervised Sequence Labelling with Recurrent Neural Networks (Vol. 385, pp. 37–45). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-642-24797-2_4
• Heilbroner, S. P., & Miotto, R. (2023). Deep Learning in Medicine. Clinical Journal of the American Society of Nephrology : CJASN, 18(3), 397–399. https://doi.org/10.2215/CJN.0000000000000080
• Hernández-Muñoz, E. A., Zavala-Sánchez, E. V., Rubio-Atondo, A., López-Chávez, J. A., Arredondo-Rojas, V., & Reyes-Reyes, J. S. (2024). Scorpion sting envenomation outbreak in Mexico. Strategies for prevention and control. Toxicon, 237, 107549.
• Hua, Y., Zhao, Z., Li, R., Chen, X., Liu, Z., & Zhang, H. (2019). Deep learning with long short-term memory for time series prediction. IEEE Communications Magazine, 57(6), 114–119.
• Joffe, M., Gambhir, M., Chadeau-Hyam, M., & Vineis, P. (2012). Causal diagrams in systems epidemiology. Emerging Themes in Epidemiology, 9(1), 1. https://doi.org/10.1186/1742-7622-9-1
• Kumar, A., Goyal, S., Garg, M. K., & Gopalakrishnan, M. (2023). Scorpion Sting Envenomation, a Neglected Tropical Disease: A Nationwide Survey Exploring Perspectives and Attitudes of Resident Doctors from India. The American Journal of Tropical Medicine and Hygiene, 109(4), 957–964. https://doi.org/10.4269/ajtmh.23-0194
• Lindemann, B., Müller, T., Vietz, H., Jazdi, N., & Weyrich, M. (2021). A survey on long short-term memory networks for time series prediction. Procedia Cirp, 99, 650–655.
• Madden, K. (2024). Developing a Falls Prediction Tool Using Both Accelerometer and Video Gait Analysis Data in Older Adults (Clinical Trial Registration No. NCT04354623). clinicaltrials.gov. https://clinicaltrials.gov/study/NCT04354623
• Mahdi-Esferizi, R., Haji Molla Hoseyni, B., Mehrpanah, A., Golzade, Y., Najafi, A., Elahian, F., Zadeh Shirazi, A., Gomez, G. A., & Tahmasebian, S. (2023). DeeP4med: Deep learning for P4 medicine to predict normal and cancer transcriptome in multiple human tissues. BMC Bioinformatics, 24(1), 275. https://doi.org/10.1186/s12859-023-05400-2
• Nohara, Y., Kai, E., Ghosh, P. P., Islam, R., Ahmed, A., Kuroda, M., Inoue, S., Hiramatsu, T., Kimura, M., & Shimizu, S. (2015). Health checkup and telemedical intervention program for preventive medicine in developing countries: Verification study. Journal of Medical Internet Research, 17(1), e3705.
• Piccialli, F., Somma, V. D., Giampaolo, F., Cuomo, S., & Fortino, G. (2021). A survey on deep learning in medicine: Why, how and when? Information Fusion, 66, 111–137. https://doi.org/10.1016/j.inffus.2020.09.006
• Riano-Umbarila, L., Romero-Moreno, J. A., Possani, L. D., & Becerril, B. (2025). State of the art on the development of a recombinant antivenom against Mexican scorpion stings. Toxicon, 257, 108306.
• Rodríguez Weber, F. L., Ramírez Arias, J. L., Rodríguez Weber, F. L., & Ramírez Arias, J. L. (2022). Medicina personalizada, individualizada, de precisión y centrada en el paciente; diferencias o sinónimos. Su importancia. Acta médica Grupo Ángeles, 20(1), 111–112. https://doi.org/10.35366/103572
• Slim, K., Selvy, M., & Veziant, J. (2021). Conceptual innovation: 4P Medicine and 4P surgery. Journal of Visceral Surgery, 158(3, Supplement), S12–S17. https://doi.org/10.1016/j.jviscsurg.2021.01.003
• Song, X., Liu, Y., Xue, L., Wang, J., Zhang, J., Wang, J., Jiang, L., & Cheng, Z. (2020). Time-series well performance prediction based on Long Short-Term Memory (LSTM) neural network model. Journal of Petroleum Science and Engineering, 186, 106682.
• System, N. A. of E. (US) and I. of M. (US) C. on E. and the H. C., Reid, P. P., Compton, W. D., Grossman, J. H., & Fanjiang, G. (2005). A Framework for a Systems Approach to Health Care Delivery. In Building a Better Delivery System: A New Engineering/Health Care Partnership. National Academies Press (US). https://www.ncbi.nlm.nih.gov/books/NBK22878/
• Tang, B., Pan, Z., Yin, K., & Khateeb, A. (2019). Recent Advances of Deep Learning in Bioinformatics and Computational Biology. Frontiers in Genetics, 10. https://doi.org/10.3389/fgene.2019.00214
• Teo, K. K., & Rafiq, T. (2021). Cardiovascular Risk Factors and Prevention: A Perspective From Developing Countries. Canadian Journal of Cardiology, 37(5), 733–743. https://doi.org/10.1016/j.cjca.2021.02.009
• Torre-Bouscoulet, L. (2015). Los retos de la medicina personalizada. Neumología y cirugía de tórax, 74(4), 238–239.
• Trinidad-Porfirio, B. E., Morales-Pérez, A., Nava-Aguilera, E., Flores-Moreno, M., Morales-Nava, L., García-Leyva, J., Silva-Domínguez, R., Cortés-Guzmán, A. J., Fernández-Salas, I., & Andersson, N. (2023). Occurrence of scorpion sting and associated factors in a highly marginalized municipality in Guerrero, Mexico: A cross-sectional study. PLOS Neglected Tropical Diseases, 17(5), e0011271. https://doi.org/10.1371/journal.pntd.0011271