Machine learning in drug supply chain management during disease outbreaks: a systematic review

ABSTRACT


INTRODUCTION
The coronavirus disease (COVID-19) pandemic event sparked research on forecasting the spread of disease outbreaks to a greater extent than in the past. The danger of COVID-19 urges the fulfillment of the need for COVID-19 drugs. The challenge for drug manufacturers and supply chain management is to meet the large and sudden demand for drugs [1]. The endpoint of drug supply chain management is on the side of the pharmacy user or patient. According to food and drug administration (FDA) [2], a long drug supply chain will complicate drug supply management and even become one of the triggers for drug shortages. Singh et al. [3] revealed that the pharmaceutical supply chain has a complexity that involves various stakeholders ranging from manufacturers, distributors, customers, information service providers, and the government as a regulatory agency. In addition, supply chains face environmental uncertainty, capacity planning, and inventory management [4]. Responding to the complexity of the drug supply chain, Lozano-Diez et al. [5] suggest taking advantage of technological developments.
The use of technology is not only in production planning, distribution processes, and storage. There must be integral management from upstream to downstream, namely primary manufacturing, secondary manufacturing, distribution centers, wholesalers, and retailers/hospitals [6], [7]. In this industrial era 4.0, all

Supervised learning
The focus of supervised learning is on accurate predictions. Machine learning works so well that previously unseen data reveals new information. Supervised learning begins by labeling the data collected. Learning outcomes produce a certain level of performance accuracy. Supervised learning methods are usually used for classification, regression, object detection, image captions, and semantic segmentation. A simple example, this method can be applied to determine whether the male face is different from the female face. By labeling face examples and gender labels, the machine learns to obey the rules for distinguishing male and female faces [21], [22].

Unsupervised learning
Unsupervised learning classifies the collected data based on comparing the features and characteristics of the unlabeled data. The data classification aims to learn a reasonable and concise description of data. Meanwhile, learning techniques commonly used are clustering and data projection/preprocessing [21], [22]. The ultimate achievement of unsupervised learning focuses on data collection in subgroups, patterns, or clusters.

Reinforcement learning
The learning process allows feedback action to the algorithm. This method uses a dataset with a reward or punishment system. Reinforcement learning is a form of learning to map from situations to actions that maximize scalar rewards or reinforcement signals. Learning from the experience of certain circumstances, the learning agent takes appropriate steps to achieve long-term goals. Reinforcing targeted actions aim at long-term rewards. Trial seeking and delayed reward are two of the most important distinguishing features of reinforcement learning [21], [22].

Drug supply chain
The drug supply chain is a complex process with specialized operations. Strict restrictions and arrangements are made on supply and delivery channels to customers to prevent drug supply errors [18]. Many agents build supply and demand interactions. They participate in the supply process (upstream) and the patient care process (downstream). Franco and Alfonso-Lizarazo [23] stated that the pharmaceutical supply chain has the following two main elements: supply consists of i) drug manufacturers and distributors; demand consists of ii) pharmacies that order drugs from suppliers, maintain drug safety, manage drug supplies, and distributing drugs to hospitals; iii) hospitals that provide drugs to patients and order them from pharmacies or distributors; and iv) patients requiring treatment and medication Figure 1.  Figure 1 shows the echelon and link under consideration in pharmacy and hospital. Decisions about ordering drugs from manufacturers and distributors are in these two units, as are decisions about managing drug storage and distribution. Lozano-Diez et al. [5] show in Figure 2, the five main areas of risk of failure of integrated operations in the case of drug supply in the public sector. Dispensing pharmacy is the meeting point of drug production and drug demand. All five areas have the same potential to contribute to failure in the drug supply chain process. Interrelated relations between areas will affect the resilience of supply. The small gap between demand and supply determines the resilience of the drug supply chain.
Dispensing position Figure 2 is in the middle, allowing you to know the gap between demand and supply from producers. The challenge for dispensing is to convey demand predictions as accurately as possible to producers. Dispensing's success in sharing information with distribution units will also provide  ISSN: 2088-8708 Int J Elec & Comp Eng, Vol. 13, No. 5, October 2023: 5517-5533 5520 readiness for drug supply disruptions. Predictive information helps supply chain operations to respond to unexpected risks. Zahiri et al. [6] stated that the problem must be local and not stop production at the primary producer level. Whereas in the area of demand, the participation of many agents and regulators is an effort to maintain drug supply in joint action [5], [6]. Figure 2. Drug supply process [5] Previously, Anaya-Arenas et al. [24] said that management needs to have a level of preparedness and mitigation, response, and recovery in an emergency. A resilient supply chain is able to identify risks, ameliorate impacts, and return quickly from human or natural events [25]. A risk manager can run a resilient supply chain by designing a pharmaceutical supply chain network and proactively conducting scenario analysis to support the development of risk response contingency plans and reactively analyzing impact and implementing recovery [26]. Management needs to design a strategy so that drug supply can quickly return to its normal position or to a more optimal position. In addition, drug supply management must recognize threats, possible losses, and similar incidents that will recur [27]. Management needs information to make decisions amidst epistemic and stochastic uncertainties [28].

Disease outbreaks
A disease becomes an epidemic when an increase in the number of cases of a disease in a local community, a given geographic area, or a season is high enough to cause an outbreak. Outbreaks occur when an infectious agent spreads directly between people, through exposure to animal reservoirs, insect or animal vectors, or other environmental sources. In addition, human behavior almost always contributes to the spread. Early detection of the spread of epidemics and reporting to the health office is important to prevent the spread of epidemics and minimize negative impacts on public health, society, and the economy [29].
To conclude that there is an outbreak in an area, cases occurring in that population require an epidemiological investigation (example: rabies, plague, and polio). Diseases that are unprecedented, irregular, and rare in a region also require investigation. An increase in the number of cases of the disease may occur in a wider geographic area. Epidemic cases can occur due to [30]: − Increase in the number or virulence of a new agent, − Introduction of new agents into new situations, − High mode of transmission, − Increased susceptibility of the human body's response, − Increased host exposure or new entry points.
Systematic investigation and risk assessment of acute public health events are necessary to prevent a wider spread of the disease. Assessment involves measuring the likelihood of disease occurrence and its eventual impact on human society. For infectious diseases, the risk will relate to the size, characteristics, and living conditions of the person migrating. Migratory human behavior poses serious epidemic risks and health impacts. The risk depends on the previous immune status of the population and the variable threat at the destination [31]. For this reason, it is important to consider resources (human, laboratory, logistics, and materials) to anticipate an outbreak. Without adequate resources, the health system is weakened and the poor are increasingly at risk of outbreaks [32].

METHOD
The literature review methodology proposed by Kitchenham [33] and Kitchenham et al. [34] became the reference for this research methodology. The research stages include planning, execution, and reporting. Figure 3 shows the process of activities in the literature review.

Planning the review
The initial study identified the need for a review study obtained from research articles that attempted to increase the accuracy of drug demand predictions based on historical data and various methods to produce high accuracy. In addition, there are indications of drug shortages (large and small scale) in many − What parameters are considered in the use of prediction methods? − What is the influence (correlation) of the research prediction parameters with the field of drug supply management? This literature study covers more than one field of science (multidisciplinary). The study belongs to the field of computer science because of the machine learning methods and algorithms for prediction. In terms of a subject area, this study discusses diseases and drugs, which are part of health science. Finally, the study talks about the distribution of drugs to consumers, which is in the field of supply chain management. Therefore, the literature sources in this research are related to computer science, information systems, and pharmacy. This research uses the Publish or Perish 8 application from the digital library: Scopus and the National Library of Medicine (PubMed) to ensure the articles used were from accredited sources. The article search categories were from journals and conferences published in 2018 to 2022.

Conducting the review
With the help of the Publish or Perish 8 application, search for articles that will become material for literature review using search terms in the title/keyword. The library search string used is ("machine learning") AND ("predict" OR "prediction") AND ("disease outbreaks" OR "endemic"). In the selection process, articles are subject to the following inclusion dan exclusion criteria Table 1. The selected articles have relevance to answering research questions so that the article criteria address at least one disease outbreak, refer to machine learning methods in the process of solving them, and see the prediction problem as a need in the drug supply chain as much as possible. Assessment of the quality of research articles uses the following questions so that the study articles have library materials with complete data answering research questions. − Was there a clearly stated research objective? − Is the article well designed to achieve its purpose? − Were the variables considered in the study identified? − Is the discipline clearly defined? − Were data collection methods adequately explained? The article management process uses the Mendeley application to facilitate the conducting and data extraction. After reading the abstract and full article content, Table 2 shows the process of selecting articles by searching relevant bibliographic databases, identifying suitable studies, and extracting the results. Searches with keywords in the Scopus and PubMed libraries found 214 study articles. In the conducting process, the study removes articles that do not meet the inclusion and exclusion criteria Table 1. The most common cases of duplication of simultaneous publications are on Scopus and PubMed. At this selection stage, there were 96 articles that had abstract content in accordance with the purpose of this literature review. After reading the contents of the articles in more detail, there are 71 selected articles.

Reporting the review
Researchers used a qualitative study approach. Concern over the issues of drug shortages, drug supply, and the development of drug demand predictions encourage researchers to approach each study material with a line of argument synthesis. The main issues in the research questions are identified and tabulated in a table and descriptions. Finally, this multidisciplinary study selected publications in Scopusindexed international journals.

RESULTS AND DISCUSSION
This section reports the results of a systematic literature review study. The first part will briefly look at the general description of the selected articles. The general description of the selected articles provides a viewpoint of the feasibility of the articles for review. Furthermore, the discovery of outbreaks over the past five years, methods of predicting outbreaks, and prediction parameters are the subject of discussion on how predictive research results support drug supply management.

Overview of selected articles
There are 71 selected articles from the Scopus and PubMed digital library databases. The percentage of the number of articles from the Scopus is greater than that of the PubMed because this literature review study focuses on machine learning Figure 4. During the conducting process, there were many copies of published articles on Scopus and PubMed. The possibility occurs because the topic is multidisciplinary. Any research result working on machine learning methods will publish an article in the science field. The possibility of the same research in medical journals occurs when the research discussion is in the medical journal. Researchers must read all the article to find out the research discussion area. There are 24 author countries gathered in this study, but the number of author countries with more than one (1) article is 12 countries Figure 5. The composition of the author's country representation can also represent various parts of the world. The results of the selection of articles are representative enough to provide a view that the topic of using learning machines in the midst of an outbreak is a topic that is currently receiving the attention of researchers. China, England, and America still dominate research in this field. The number of developing countries that have conducted this research indicates that they are facing problems that are not much different. Figure 5. Composition of research countries with more than one (1) article

Disease outbreaks that have been a research study in the last five years
Of the 32 types of disease detected from the study, an outbreak study with more than two articles only had eight (8) types of disease. Figure 6 shows the eight types of epidemic disease that can be categorized into three disease categories, namely: i) epidemics with disease vectors in the form of mosquitoes (dengue fever, malaria, chikungunya, and zika); ii) outbreaks caused by viruses that attack the respiratory tract (influenza and COVID-19); and iii) outbreaks are caused by bacteria, viruses, and environmental contamination (foodborne food and diarrhea). Each category is marked with the same pattern in Figure 6. The three categories of disease outbreaks identified can occur in one country, several adjacent countries, and a wider area. Each country or continent region has different demographic, socio-cultural, economic community characteristics that are of great concern to researchers. The severity of the outbreak is also a concern for researchers, thus encouraging them to participate in preventing or inhibiting the rate of spread of outbreaks with research on predicting the spread of outbreaks.

Predictive methods used in research to address disease outbreaks
In this literature study, the trend method approach is taken from the methods discussed in more than two (2) articles that conduct studies of certain disease outbreaks. Table 3 (see in Appendix) shows the eight outbreaks identified using machine learning algorithms according to the cases each one faces [35]- [91]. An interesting phenomenon is when more than one outbreak uses the same machine-learning method. There are indications that the use of machine learning is by the objectives to be achieved by research, but also (needs to be investigated more deeply) parameters that are not much different are used as parameters in learning.
Based on the three outbreak categories in this study, the commonly used machine learning techniques are as follows: i) epidemics with disease vectors in the form of mosquitoes (dengue fever, malaria, chikungunya, and zika) using decision trees, regression, support vector regression (SVM), boosting algorithm, neural network; ii) outbreaks caused by viruses that attack the respiratory tract (Influenza and COVID-19) using naïve Bayes, SVM, boosting algorithm, decision tree learning, k-nearest neighbors (KNN), regression, neural network; iii) outbreaks are caused by bacteria, viruses, and environmental contamination using Bayesian Regression models, linear regression models, convolutional neural networks (CNNs), long short-term memory networks (LSTMs) and SVMs. Different approaches can complement the limitations of previous research or improve what is still lacking, such as increasing accuracy in supervised learning.
In the case of technique selection, Benedum et al. [76] developed machine learning (ML), regression, and time series models to predict dengue rates over weekly time periods. The data for the forecast uses weekly dengue fever outbreak monitoring data and available weather data. Meanwhile Jain et al. [74] considers the use of parameters of meteorological data, clinical, disease surveillance lag variables, socioeconomic, and spatial dependence coding on dengue transmission. In one-month intervals, they developed generalized additive models (GAMs) to synchronize the relationship between predictors and clinical data of dengue hemorrhagic fever (DHF). Both of these studies sufficiently show that the use of techniques is always related to the parameters. Comparison between ML techniques to obtain high accuracy needs to consider the level of parameter development.
An interesting study was conducted by Raizada et al. [75], when they combined machine learning algorithms for prediction studies of mosquito vector-borne disease outbreaks for all three diseases (chikungunya, dengue, and malaria). They built the CNN-multimodal disease outbreak prediction (CNN-MDOP) algorithm. Future studies can increase the accuracy by 88% by increasing the accuracy of demographic data, meteorological data, and disease outbreak prediction. Saha et al. [86] found a challenge to elucidate the etiology of meningitis in Bangladesh. At that time, they conducted validation and application studies of unbiased metagenomic next-generation sequencing (mNGS) to elucidate the etiology of meningitis in Bangladesh. They used cerebrospinal fluid (CSF) specimens from patients admitted to the largest children's hospital. In cases of disease of unknown cause (idiopathic), mNGS identified a potential bacterial or viral etiology in 40% of supervised machine learning algorithms. In this case, the discovery of diseasecausing knowledge is useful for determining the production and distribution of drugs for healing.

The parameters to be considered in the use of the prediction method
With the formation of three categories of disease outbreaks, the characteristics of the disease are found in clinical data, patient history, and the spread of disease that is influenced by natural factors. The parameters used for the eight types of outbreaks are also not much different. Parameters are determined for training because they are involved in determining whether the disease can be dangerous or not.
Parameter areas that are widely used are related to disease symptoms, disease history, distribution patterns, climate-meteorological aspects, demographic information about the outbreak environment, mobility behavior of sick people, and real-time data about the rate of disease spread. It needs an in-depth discussion, prediction of the spread of disease is not only done for health prediction or knowledge. Accurate predictions will correlate with drug supply and availability in pharmaceutical units.
Complete parameters in the disease spread prediction model are not guaranteed to improve accuracy. Environmental aspects, the level of socio-economic awareness of the community, the characteristics of the causes of outbreaks, and geographical conditions have dynamic features that will affect the prediction results. For Santosh [56], continuous and unprecedented factors must receive attention in designing complex predictive models. He proposes not to rely solely on random parameter-driven stochastic or discrete indicators. Epidemiological characteristics (of the disease) can form the basis of initial consideration for predictive models. According to Pourghasemi et al. [55], the selection of predictor variables was based on a literature review and expert knowledge related to disease outbreaks. These considerations are relevant when finding several conditions that differ from one area to another.
From Table 4 (see in Appendix) the parameter column shows several indications of data collection from government health institutions and other trusted health organizations/associations. Recent developments and disease outbreak predictions take data from web searches Google Trends and social media such as Twitter and Weibo. Dissemination of information from patients, families, and people affected by the outbreak on the internet media can be a source of information in predicting disease outbreaks. Identification of "sick posts" by social media users reporting symptoms and diagnoses for either themselves or others regarding the COVID-19 outbreak can be classified by machine learning [53]. The prediction results using the social networking site (SNS) model in tracking disease outbreaks can detect COVID-19 accurately, as an initial stage, before official information from the outbreak information center [54]. It is similar to taking data from the main topic of each news article. Subsequent research has led to the identification of news articles and related events that refer to the symptoms of the disease in question [35], [36].

Correlation of research prediction parameters with drug supply management area
Each outbreak has its own characteristics according to the disease vector and its spread. These characteristic differences affect the determination of parameter weights which can improve prediction accuracy. The discussion of determining close predictive parameters on outbreak characteristics requires further research. However, the third column of Table 4 shows the relationship between the parameters used in the drug supply management area. Based on drug supply chain theory and the five integrated operational failure risk areas in the case of public sector drug supply (Figures 1 and 2), we identify the parameters used in the prediction of the eight outbreaks into supply chain management areas: planning, production, distribution, storage, and risk of supply chain. Table 5 shows the correlation of the predicted spread of eight outbreaks using predictive parameters that relate to at least two areas of drug supply management. In influenza outbreaks, predictive parameters showed correlations in determining the amount and type of drug production and distribution, management of drug storage, and planning of type and distribution pattern. However, in this literature study, articles predicting the spread of influenza outbreaks have not found parameters that discuss the risk of drug supply.
In the case of COVID-19, Table 5 shows that the researchers developed the most extensive parameters. The spread of COVID-19 and the high death toll have prompted a large-scale study. The correlation between the parameters used and the area of drug supply management indicates a large study. There is a sizable difference in research between COVID-19 and other outbreak cases. In cases of others outbreaks, there are not many prediction parameters that consider the risk areas of drug supply.
=parameters are in the areas of drug supply management 0=parameters are not in the areas of drug supply management

Practical implications and limitations
The correlation of research prediction parameters to drug supply management has practical implications for predictions by machine learning and drug supply management. In terms of developing machine learning, increasing the accuracy of predicting drug demand is developing by taking data from social media and real-time media for monitoring the spread of outbreaks [35], [36], [43], [45], [46], [54], [57], [90]. The development of drug demand prediction parameters supports the development of drug distribution chain models. Predicting the spread of outbreaks in an area is a consideration for the resilience of the drug supply chain. Parameters of the outbreak spread can be an early indication of the need for a drug supply model on a national and regional scale.
The spread of the COVID-19 outbreaks and the dangers to human health have triggered the use of deep learning machine technology by no longer relying on historical disease and patient data. The basic prediction parameters from disease characteristic data, patient distribution data, data from trusted government/private health institutions, data related to climatological-meteorological conditions, and sociodemographics of the developing community to traffic data related to epidemic diseases. For drug manufacturers, data on developments to predict the spread of disease form the basis for planning drug production and distribution. Predicting drug demand by observing the real-time conditions where an outbreak occurs or is likely to expand will add value to the accuracy of the prediction. In more depth, future research can examine the predictive parameters of types of diseases with high mortality values in a region. The role of machine learning in predicting the spread of disease outbreaks triggers the integrated drug supply management process. We can take the example of demographic parameters in predictions. Mobility and population density will affect the forecasting of disease outbreaks [56], [72]. From the prediction results with demographic parameters, drug supply management cannot ignore the risk factors of the supply chain. In the case of parameters, changes in social and economic characteristics for types of foodborne diseases and diarrhea can be significant in predictions. Dynamic changes in the field can affect prediction accuracy and drug supply management. Practical implications, changes in the direction of research from the smooth distribution of drugs and predictions of drug demand need to find common ground towards integrated drug supply information management.
This study attempts to place predictive parameters as a factor integrating drug demand prediction research in drug supply management. The limitation of this study is that it does not provide specific recommendations or solutions to improve drug supply chain management or predict disease outbreaks due to the extensive discussion on disease outbreaks in general. In future research, predictive parameter research using the machine learning method of a disease outbreak in a region can be one of the factors in developing drug supply management.

CONCLUSION
This literature study found three categories of epidemic causes with eight types of diseases, namely i) mosquitoes as disease vectors (dengue fever, malaria, chikungunya, and zika); ii) viruses that attack the respiratory tract (influenza and COVID-19); and iii) bacteria, viruses, and environmental pollution (foodborne food and diarrhea). The character of the vector carrying the disease also determines the characteristics of the spread of the disease and the prediction method used. The use of machine learning methods to predict the spread of disease during epidemics is still within the technical scope of using combination/hybrid/ensemble methods. No single approach method has high accuracy. Disease characteristics and external (non-clinical) factors determine the prediction method used. In addition to clinical and non-clinical characteristic parameters, public data via social media tends to attract researchers' interest.
This study also reveals that predictive results can support drug planning, production, distribution, and storage. Meanwhile, there are not many studies using predictive results to assess risk factors for drug supply. This research contribution identifies a strong correlation between the learning parameters of the COVID-19 outbreak in machine learning and the field of drug supply management.
This literature review can be a starting point for further studies focusing on the latest methods in prediction for various disease categories that will assist future research. Comparative research on prediction methods and parameter determination tests in the risk area of drug supply to what extent provides increased prediction accuracy. Likewise, the effectiveness and efficiency of the prediction approach will be the attractiveness of its application to various types of diseases. The achievement of effectiveness and efficiency has led to research into integrated drug supply systems and deep-learning hybrid methods. Table 3. Methods of approaches to eight types of diseases   [49] − The augmented regression with general online data (ARGO) model, the net model, the autoregressive model − RF and SVM, multi-layer perceptron (MLP) models, logistic regressor (LR) models, medical statistics − The classification and regression tree-decision tree (CART-DT) algorithm, SVM and backward propagation (BP) neural network, combining CART, XGBoost, and LightGBM − LASSO regression, analysis retrospective COVID-19 (17) [41], [44], [50]- [64] − Distributed machine learning and blockchain, neural network − DT, extra trees, KNN, multilayer perceptron, SVM, RF − Curve-fitting models, agent-based models, susceptible-exposed-infectious-removed or susceptible, infected, and recovered (SEIR/SIR) models − Spatial modeling, risk mapping, and change detection of COVID-19, RF, machine learning technique (MLT), LASSO algorithm − The time series forecasting method, ML approach, WEKA − Mathematical approaches based on susceptible, infected, and recovered (SIR) cases and susceptible, exposed, infected, quarantined, and recovered (SEIQR) cases models, prophet algorithm, Broyden-Fletcher-Goldfarb-Shanno (  − Symptoms of disease − Measures to control epidemic spread, host and environmental susceptibility, pathogen transmission, population density, and health care capacity − Historical disease information in the clinic and new discriminant input patterns − Sentiment analysis and keyword occurrences related to disease − Graph of disease visit surveillance data both from the health department and patient medical records − Records of historical disease symptoms, Google search data, meteorological data -Atmospheric data − Historical/non-real time data from the Epidemic Center and real-time Twitter data − Availability of Regional Datasets on the spread of disease and social media − Word embedding options regarding outbreak, tokenization, and time series conversion data − Google search activity, real-time weather and local information, flu-related Twitter micro-blogs, electronic health record data, and disease activity synchronicity history across regions − Data on symptoms of similar diseases − Data on medical records and data on the spread of epidemics from government health departments − The daily number of influenza incidence and meteorological data: temperature, precipitation, wind speed, relative humidity, air pressure, daily radiation − Keyword search logs − Web search queries − Dynamics of information related to the development of up-to-date and reliable outbreak information data: the current epidemiological, demographic and meteorological situation, and the environment (consideration of the amount and type of drug production and drug distribution) − Update historical data comparing with historical drug supply (drug storage and distribution considerations) − Consideration of the type and quantity of drug supply following the epidemic pattern (consideration of drug production and supply) − External: Twitter, online news articles, and Google Flu Trends (GFT) up-to-date and reliable data: current epidemiological, demographic and environmental situation. Early indication of drug demand dynamics in a particular area. External information accompanies existing data in the health department and hospital/pharmacy (consideration of planning types of drugs and supplies) − External: updating the data base on disease symptoms and pathology to determine the type of drug and the need for similar drugs (consideration of the production of the type of drug needed) COVID-19 (17) − Patient data records − Symptoms of disease − Characteristics of the disease and level of care of the patient, including: i) hospital setting/capacity; ii) test capacity/rate (daily); iii) demographics; iv) population density; v) vulnerable people; and vi) income versus commodities (poverty) − 16 variablesactive case data, died, and correlations between them. Transmission patterns and real demographic, climatological, and distribution cluster situations − Limited to data on confirmed cases, recovered cases and death cases − 16 effective factors: minimum temperature of coldest month, maximum temperature of warmest month, precipitation in wettest month, precipitation of driest month, distance from roads, distance from mosques, distance from hospitals, distance from fuel stations, human footprint, density of cities, distance from bus stations, distance from banks, distance from bakeries, distance from attraction sites, distance from automated teller machines (ATMs) and density of villages − Data on variations, trends and predictions from 2 health institutions − Disease characteristics (Virus) − Data on confirmed cases by time − Genome database and reservoir host disease characteristics − Embedding hybrid words related to epidemic cases − Comparison of patterns/symptoms of similar diseases − Data on medical records and data on the spread of epidemics − Disease data structure − Information on environmental factors and foodborne diseases − Information from case reporting systems, outbreak monitoring systems, molecular tracking networks − Non-real time data center of the outbreak and real-time Twitter − Information on foodborne disease risk factors caused by the spread of water pollution, sick case data on Weibo, geographic location data of the water pollution plants − Updating of the disease symptom and pathology data base for determining the type of drug (consideration of drug production and distribution) − External: Twitter and Weibo, symptom and pathology data base updates for drug determination in certain areas (drug distribution considerationsdrug supply readiness) Dengue fever (14) − Local meteorological data, clinical data, disease surveillance lag variables, socioeconomic data and spatial dependence coding data on DHF transmission − Data on the spread of dengue fever, population density, entomology and environment − Data on Human dengue cases per month and data on basic demographic characteristics (e.g., sex, age, nationality, and residential address) and time of occurrence related to the disease (e.g., date of onset, diagnosis, and death) − Meteorological data: 15 meteorological variables (i.e., extreme wind speed, maximum wind speed, average wind speed, minimum pressure, maximum pressure, average pressure, average water pressure, minimum air temperature, maximum air temperature, temperature average air, highest average daily temperature, lower average daily temperature, average daily precipitation, number of days with precipitation, and average relative humidity maximum pressure, average pressure, pressure average water, minimum air temperature, maximum air temperature, average daily highest temperature, average daily precipitation, number of days with precipitation, average relative humidity, geographical location − Baseline patient data, current data of rainfall, sea surface temperature, and temperature of outbreak sites − Clinical symptoms of dengue, a real-time surveillance system, climate variables such as temperature, wind speed, humidity and rainfall − Data on historic dengue cases, satellite-derived estimates for vegetation, precipitation, and air temperature, as well as population counts, income inequality, and education − Information on laboratory test results, four main variables [age, fever, white blood cell (WBC) count, and platelet count], compatible with clinical and epidemiological knowledge -vector population − Data on symptoms of similar diseases − Demographical data: positive cases; meteorological data: temperature, humidity, rainfall − Provision of drugs based on Medical Record data (planning considerations in pharmacy units) − Consideration of the type and quantity of drug supply following the epidemic pattern (consideration of drug production) − Update on disease symptoms and pathology data base for determining drug types (consideration of drug production and supply in pharmaceutical units) − Update historical data on the spread of epidemics with historical data on drug supply, documentation of pharmaceutical unit warehouse data and information on the need for similar drugs (consideration of drug storage) − Up-to-date and reliable data: current epidemiological, demographic, and environmental situation (considerations of drug planning and distribution)

Chikungunya (3)
− Demographic data: positive cases; meteorological data: temperature, humidity, rainfall − Data on clinical symptoms and variations in phenotypic response − Symptoms of disease and patient test results − Comparison of recent historical data on the spread of epidemics with historical drug supply (consideration of stock storage and distribution of drugs to certain areas) − Update data base on symptoms of disease and pathology for determining the type of drug (drug production considerations) Malaria (4) − Real-time information on transmission dynamics, parasite movements and reservoirs of infection − Data on malaria incidence and climate variability (rainfall, temperature and surface radiation) − Demographical data: positive cases; meteorological data: temperature, humidity, rainfall − Data on disease outbreak characteristics, tropical climate variability and associated sea surface temperatures over the Pacific and Indian Oceans − Consideration of the type and quantity of drug supply following the epidemic pattern (consideration of drug production) − Up-to-date and reliable data: current epidemiological, demographic, and environmental situation (considerations of drug planning and distribution) − Comparison of recent historical data on the spread of epidemics with historical drug supply (consideration of stock storage and distribution of drugs to certain areas) − Meteorological and environmental covariates, demographic distribution of mosquitoes, socioeconomic factors -human movement and urbanization − Data provided by MedIsys. Information on article title, description, publication date, original URL, language code, category indicating disease name, latitude and longitude are displayed as initial data for prediction − Comparison of symptoms of similar diseases − Consideration of the type and quantity of drug supply following the epidemic pattern (consideration of drug production) − External: the news article data. update data base on symptoms of disease and pathology for determining the type of drug (consideration of drug planning and distribution) − Need for similar drugs (consideration of drug production and storage)