Implications of Big Data Analytics, AI, Machine Learning, and Deep Learning in the Health Care System of Bangladesh: Scoping Review

Introduction

Background

In recent years, the global surge in digital technology has ushered in a data-driven decision-making era across diverse sectors, including the health care and public health sectors [1]. The convergence of state-of-the-art technologies, such as big data analytics (BDA), artificial intelligence (AI), machine learning (ML), and deep learning (DL), is collectively reshaping the health care and epidemiological landscape [2]. This paradigm shift promises to revolutionize the comprehension and management of health challenges, offering the potential for more efficient and effective interventions [2].

The health care system fundamentally revolves around safeguarding and enhancing individual and community health by integrating public health and medical care services, using health promotion, disease and injury prevention, infectious disease surveillance, response, and research. Interdisciplinary research now explores various factors influencing individual and community well-being, ranging from microbial infections to social determinants, behavioral patterns, and various facets of lifestyle [3]. In addition, the widespread adoption of emerging technologies such as telemedicine, clinical decision support systems, electronic health records, personal health records, and mobile health care have generated a vast reservoir of health-related data [4].

Harnessing the latent potential of these extensive data, digital technologies, such as BDA, AI, ML, and DL, can surpass conventional data management systems [5]. By applying AI algorithms, ML techniques, and DL capabilities, BDA offers invaluable support to clinicians, health care providers, and policy makers in intervention designing, planning, and execution; expediting disease detection; predicting health outcomes; and advancing personalized medicine [6]. This ultimately results in both cost-effectiveness and quality outcomes [7].

On a global scale, extensive health data can be categorized into 5 primary domains: biological measurements of participants (eg, genomic or metabolomic datasets), participant context measurements (eg, environmental variables), administrative medical records, participant tracking data from devices equipped with GPS, and electronic data from various sources (eg, social media or search history records) [8]. Using BDA, AI, ML, and DL algorithms to extract valuable insights from these datasets can significantly enhance individual health management and enable swift disease trend detection and outbreak responses [9]. Beyond health care, BDA, AI, and ML can play crucial roles in enabling molecular data analysis, synthesis of genetic and clinical information to craft personalized treatments, and development of predictive models based on omics data to identify genetic disease markers and develop effective novel drugs [10].

In Bangladesh, which is a densely populated South Asian nation with >160 million inhabitants, integrating modern technologies with the health care system holds immense potential. Bangladesh grapples with various health-related challenges, spanning from infectious diseases, noncommunicable diseases (NCDs) and maternal and child health issues to environmental health concerns [11]. To tackle these multifaceted challenges comprehensively, the government of Bangladesh (GoB) has actively adopted advanced technologies to elevate the quality of health care services. Initiatives such as District Health Information System 2, the open medical record system, and the open-source smart register platform have revolutionized health data collection and service delivery [12]. These automated applications are instrumental in generating specific health-related evidence across different tiers of health care services and facilitating need-based resource allocations. Although there are potential challenges regarding proper data governance and legal framework, the immense potential to use these data has been capturing the attention of interdisciplinary researchers and practitioners eager to explore the potential of BDA, AI, ML, and DL technologies to use this wealth of datasets [12].

Objectives

In Bangladesh, instances of using BDA, AI, ML, and DL technologies in health care and public health research are sporadic. However, the dearth of clear evidence has led to an inadequate understanding of the proper application of these technologies and their potential contributions to health care and public health research [12]. This scoping review is motivated by the need to address this gap, systematically examining existing research that explores the use of BDA, AI, ML, and DL technologies to analyze health care data in Bangladesh. The primary objective was to compile and synthesize available research findings on these domains in the country’s health care sector. This includes identifying trends, patterns, and methodologies for applying BDA, AI, ML, and DL technologies. The secondary objective was to assess the limitations faced by researchers in integrating these technologies into health care research, aiming to provide insights into the challenges that may hinder their widespread adoption. Overall, the insights derived from this review aimed to furnish evidence-based strategies for policy makers, health care professionals, and researchers, ultimately contributing to enhancing health outcomes and the overall well-being of the Bangladeshi population. Moreover, this exploratory journey is guided by the goal of uncovering valuable insights that can significantly shape the future of health care, not only within Bangladesh but also on a global scale.

Methods

Overview

We have followed the standard guidelines for scoping reviews. This scoping review used the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews; Multimedia Appendix 1) guidelines in line with best practices for research paper selection and reporting [13], and its protocol has been registered with the Open Science Framework [14]. We have provided the PRISMA-ScR checklist.

Ethical Considerations

There were no human participants in this review, and ethics approval is not applicable.

Inclusion and Exclusion Criteria

We included all studies that are primarily focused on health care and public health issues in Bangladesh and used any of the cross-cutting technologies, such as BDA, AI, ML, and DL, in analyzing datasets. The studies that used publicly available global or web-based health care and public health datasets in their research but validated their findings in the context of Bangladesh were also included in this review. All journal articles and conference articles published between January 1, 2000, and September 10, 2023, written in English and publicly accessible, were included regardless of research design and sample sizes.

Studies conducted by Bangladeshi researchers but not focusing on the context of Bangladesh, review articles, book chapters, commentaries, gray literature, and articles published in other languages (other than English) were excluded from the study. Manuscripts that were not accessible through institutional credentials or after contacting the authors were also excluded.

A comparative textbox (Textbox 1) presents the inclusion and exclusion criteria.

Data Sources and Literature Search

We conducted a search across 4 electronic databases: MEDLINE (via PubMed), IEEE Xplore, Scopus, and Embase, on September 10, 2023. Our search was refined by applying a date filter, limiting results from January 1, 2000, to the search date. To identify relevant information, we used a set of key search terms related to the population, concept, and context framework (Textbox 2).

For each database, we meticulously crafted a comprehensive search strategy, tailoring it to the specific databases by incorporating both key terms and database-specific index terms (Multimedia Appendix 2).

The main search keywords included terms such as “Artificial Intelligence,” “AI,” “Machine Intelligence,” and “Cognitive Computing” for AI; “Machine Learning,” “Transfer Learning,” “Predictive Modeling,” and “Pattern Recognition” for ML; “Deep Learning,” “Neural Network,” “Artificial Neural Network,” and “Deep Neural Network” for DL; and “Big Data,” “Data Mining,” “Big Data Analytics,” and “Data Science” for BDA. In addition, we included keywords relevant to the health care context, such as “Healthcare System,” “Health Services,” “Clinical Settings,” and “Public Health,” and geographic context-specific terms, such as “Bangladesh,” “Bangladeshi,” and “Bangladesh Health System.”

Using these targeted keywords, we aimed to capture a comprehensive set of studies relevant to the use of BDA, AI, ML, and DL technologies within the health care system of Bangladesh.

Screening Process

Following the retrieval of articles from the databases, they were imported into EndNote (version 9; Clarivate), and duplicates were removed. Subsequently, the review team was divided into 2 groups: group 1 and group 2. Each group was assigned a subset of articles from the total pool of retrieved articles (group 1: 826/1653, 49.97%; group 2: 827/1653, 50.03%) to independently screen the titles and abstracts in accordance with the established inclusion and exclusion criteria.

Each reviewer independently screened the titles and abstracts of their assigned articles and cross-checked their decisions within groups. After this initial screening, full-text screening was performed by 2 individual reviewers. Any disagreements or uncertainties that arose during any of the screening stages were promptly resolved through discussion and consensus among the reviewers. In cases of disagreements, a third reviewer stepped in to refute the disagreements, considering inclusion or exclusion criteria. This rigorous screening process ensured that the selected studies aligned with the defined objectives and inclusion or exclusion criteria, enhancing the robustness and reliability of this scoping review.

Data Extraction, Analysis, and Quality Assessment

We developed a data extraction sheet to capture relevant data in line with the review objectives. The extraction variables comprised descriptive information of the included studies, such as title, author, published year, sample size, dataset information, study design, the objective of the study, the technology used, the domain of public health, results, limitations, and practical implementation of the study result. Following data extraction, we analyzed and categorized the data according to the dataset information, the similarity of the disease domains, methods used for analysis, and the performance of algorithms.

The quality of all the included studies was assessed following the Joanna Briggs Institute Critical Appraisal tool [15]. In total, 2 individual reviewers assessed the quality of the studies, and disagreement was solved upon discussion. For overall grading using the Critical Appraisal tool, we followed a published systematic review that used a rule of thumb for overall grading [16]. For case-control studies, if 0 to 3 “yes” answers were obtained out of 10 screening questions, the study was marked as poor quality. Consecutively, if 4 to 6 “yes” answers were obtained, the study was marked as moderate quality, and if >6 “yes” answers were obtained, the study was ranked as good quality. For cohort studies, among 11 screening questions, 0 to 3 “yes” answers were marked as poor quality, 4 to 7 “yes” answers as moderate quality, and >6 “yes” answers as good quality. For cross-sectional studies, among 8 screening questions, 0 to 3 “yes” answers were marked as poor quality, 4 to 5 “yes” answers as moderate quality, and >6 “yes” answers as good quality study (Multimedia Appendix 3).

Results

Selection of Sources of Evidence

Of the 1653 articles retrieved from the 4 databases, 78 (4.72%) duplicate publications were removed initially. Next, in the title and abstract screening stage, 83.87% (1321/1575) articles were removed as these did not align with the review objectives or were either workshop articles or not original studies (letters to editors and descriptions of keynotes). A total of 43.7% (111/254) of the publications were also excluded due to inadequate information related to research objectives. The remaining 56.3% (143/254) of the studies were taken for full-text review. Moreover, of the 143 studies, 22 (15.4%) were inaccessible and excluded accordingly. Finally, 121 (84.6%) of the 143 articles were reviewed in full text for eligibility. Subsequently, 44 (30.8%) were excluded due to inadequate information considering research objectives, and finally, 77 (53.8%) articles were selected for data extraction. A detailed description of the article selection process is given in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram (Figure 1).

‎

Key Study Characteristics and Thematic Domains

This scoping review identified 77 research articles that examine how BDA, AI, ML, and DL are used in Bangladesh’s health care system and public health. The characteristics of the studies were grouped according to the study methodology, data types, and different health domains. Researchers mostly applied ML algorithms in their studies, and in 84% (65/77) of the studies, they used cross-sectional study data. The use of cohort study, case-control study, and retrospective study data was lower among researchers. Table 1 shows the characteristics of the selected studies according to different categories.

Although the date limiter was set from January 1, 2000, to September 10, 2023, no published articles according to the research objectives were found before 2012. A small number (4/77, 5%) of published research articles were found between 2012 and 2017, focusing on public health issues using BDA, AI, ML, and DL. After 2017, a substantial surge in interest among researchers in using AI, ML, and DL was observed, which mostly focused on NCDs, infectious diseases, and child health. Among the NCD group, predicting type 2 diabetes mellitus, hypertension, and cancer is mostly focused on, and among the infectious diseases, COVID-19 was mostly prevalent throughout the pandemic from 2020 to 2022. In addition, the researchers highlighted using the target technologies for different vector-borne diseases during this timeframe (Figure 2).

‎

Table 2 summarizes the included articles with respect to their objectives, methods, dataset sizes, and investigated health issues.

Synthesis of Findings

According to the findings, the studies intended to predict various diseases using various risk factor data (33/77, 43%), classify patients in disease-exposed and unexposed groups (21/77, 27%), predict trends of the outbreak (18/77, 23%), formulate intelligent systems for health service (3/77, 4%), and so on. One study explored the regulatory challenges of using big data technology in the health sector, as well as making recommendations for future policy. Another study tried to detect glaucoma with the data obtained from optical coherence tomography and the tonometry test.

The study’s findings indicate that, compared to secondary data, the use of prediction and classification algorithms on primary data has significantly grown (40/77, 52%). In a significant number of studies (37/77, 48%), researchers used data from the Bangladesh Demographic and Health Survey, which is part of the global Demographic and Health Surveys program, designed to collect information on fertility, family planning, and maternal and child health. Furthermore, a substantial portion of studies (12/77, 16%) made use of publicly available data from web-based sources, such as Kaggle, the University of California, Irvine repository, and Worldometer. Researchers validated their prediction algorithms within the context of Bangladesh.

In this review, we observed the adoption of digital interventions, including the use of mobile devices for public health purposes. This review explored articles (4/77, 5%) that leveraged mobile devices for their research, creating intelligent risk prediction tools by analyzing data collected through ML- and DL-based algorithms. The objective was to demonstrate the value of mobile apps in obtaining real-time health data and predicting health status. The primary outcomes of these studies underscored the utility of intelligent mobile apps for self-assessing an individual’s health and offering essential therapy recommendations.

On the basis of the data extracted in this review, 44 distinct AI, ML, and DL algorithms commonly used in public health studies were identified. Table 3 presents the algorithms according to different categories and their respective use frequencies.

Among these, the random forest algorithm emerged as a standout performer, delivering the highest accuracy in 21% (16/77) of the studies. The decision tree classifier was also highly effective, with a track record of accurate predictions in 16% (12/77) of the studies. However, for maternal and child health, logistic regression showed substantial promises, while for outbreak trend prediction, the Facebook Prophet model and long short-term memory (LSTM) demonstrated strong performance. In the mental health domain, random forest excelled, and for NCDs, random forest and support vector machine exhibited strong predictive capabilities. In the context of vector-borne diseases, the decision tree algorithm proved effective.

Several other algorithms showcased their utility in various domains, including support vector machine (9/77, 12%), logistic regression (8/77, 10%), Naïve Bayes (4/77, 5%), k-nearest neighbor (3/77, 4%), convolutional neural network (3/77, 4%), gradient boosting (3/77, 4%), LSTM (3/77, 4%), Facebook Prophet time series model (3/77, 4%), artificial neural network (2/77, 3%), XGBoost (2/77, 3%), a priori algorithm (2/77, 3%), and linear discriminant analysis (1/77, 1%). In addition, support vector regression (1/77, 1%), bagging classifier (1/77, 1%), quadratic discriminant analysis (1/77, 1%), multiple linear regression (1/77, 1%), elastic net regression (1/77, 1%), and bagged classification and regression tree (1/77, 1%) displayed their potential in achieving high prediction accuracy in their respective research studies.

This comprehensive analysis highlights the diverse and effective use of these algorithms in public health research, offering valuable insights for researchers and professionals in the field.

Practical Implication of Studies

This review identified a number (9/77, 12%) [17,24,39,42,47,50,54,87,88] of studies where researchers have made web-based and Android-based solutions by applying AI-, ML- and DL-based models after conducting their research. Of the 9 studies, 6 (67%) studies developed apps that mostly focused on NCDs to allow persons to self-assess their own physical condition. The other 3 (33%) studies developed web-based prediction tools for identifying the trends and hot spots of the COVID-19 pandemic, which were found effective when the world suffered a lot from the pandemic situation.

Limitations Mentioned by the Studies

According to the findings, this scoping review observed a significant number of limitations mentioned by the researchers. In approximately 27% (21/77) studies, researchers have expressed their concerns about the datasets they have used. The articles also expressed concerns about using a large amount of data with different dimensions for risk factor identification and prediction of NCDs. Sociodemographic data (8/77, 10%), clinical feature data (15/77, 19%), and lifestyle data (12/77, 16%) were crucial for effective risk factor identification as well as disease prediction, as mentioned in these 21 studies. However, the researchers were able to work on a small sample size with inadequate features, which overfitted their predictions, as reported by the articles.

For outbreak prediction, in 25% (19/77) of the studies, researchers mentioned the necessity to use diversified data from different dimensions with proper tuning of the AI, ML, and DL models on a regular basis. Environmental data, weather data, population movement data, quarantine data, vaccination information, clinical symptoms of patients, as well as crowd-sourced data were required for effective outbreak prediction. However, due to the lack of these features, they expressed concern that their provided prediction model may not work perfectly in all contexts.

Data source is also vital for an effective predictive model, as mentioned in some of the studies (12/77, 16%). Primary-sourced datasets seemed suitable for predictive modeling as they are usually collected in accordance with the research objective, whereas secondary-sourced datasets have their own research objectives, which was not convenient for effective predictive modeling.

Risk factor identification and prediction modeling were revealed as very sensitive for child health–related diseases. Some researchers mentioned using factors from different domains as well as clinical features (10/77, 13%).

Discussion

Principal Findings

This scoping review is an attempt to explore research focusing on Bangladesh’s perspective on using BDA, AI, ML, and DL technologies across various health care domains. The 77 studies included in this review were mostly quantitative in nature and focused on 7 distinct health care domains, including infectious diseases, NCDs, maternal health, child health, preventive health, vector-borne diseases, and mental health. While most (65/77, 84%) studies adopted a cross-sectional study design, a few used cohort (5/77, 6%), case-control (4/77, 5%), and retrospective (3/77, 4%) designs. The review revealed 44 distinct AI, ML, and DL algorithms for risk factor identification, classification, prediction, and trend analysis. The most used algorithms were decision tree, random forest, logistic regression, Naïve Bayes, artificial neural network, and k-nearest neighbor. Notably, among the studies, the random forest surpassed others in classification and prediction by accuracy and F₁-score, followed by decision trees. However, in trend analysis, the Facebook Prophet model and LSTM performed better.

These technologies have exhibited effectiveness in classification, prediction, and severity analysis across various health care domains. For instance, time series algorithms, such as Facebook Prophet model and LSTM, can be invaluable in potential outbreak prediction, such as the COVID-19 pandemic. Sarkar et al [20] explored better performance in forecasting and severity analysis of COVID-19 outbreaks in Bangladesh using the Facebook Prophet model. Similarly, these algorithms can be of great use in addressing other epidemics and endemics, such as dengue and diarrhea outbreaks in Bangladesh and other South Asian countries. The use of these algorithms may include but is not limited to hot spot detection, disease pattern recognition, disease severity detection, and resource allocation for effective management. Moreover, Islam et al [48] and Asaduzzaman et al [39] incorporated different ML techniques (decision tree, support vector machine, XGBoost, gradient boosting machine, logistic regression, and linear discriminant analysis) to identify individuals at risk of NCDs, such as diabetes, hypertension, and cancer, as well as explored potential factors influencing the severity, considering the growing burden of NCDs in Bangladesh.

Furthermore, ML-driven classification algorithms can play a pivotal role in categorizing patients based on disease severity, facilitating efficient resource allocation in health care facilities. For example, Rahman et al [17] developed a web application incorporating ML algorithms to identify the severity of COVID-19 among patients in Bangladesh, enabling health care providers to prioritize treatment and intervention strategies effectively. Such categorization technology can reduce human error and overall stress if incorporated into high mobility wards (eg, diarrhea and casualty) and emergency rooms. However, future studies that focus on efficient fine-tuning and validation strategies are needed to ensure higher specificity and prevent false alarms.

Apart from these, several clinical decision support systems have been developed to assist depression diagnosis, considering the recent conversations on mental health, anxiety, and depression [77,79]. Although the models were proven successful in predicting and classifying information to aid public health experts, their real-world validation remains somewhat limited. Few studies (0/77, 13%) tested algorithms in controlled clinical settings with smaller sample sizes [17,32,36,41,49,70,73,76,83,88,89], while others (9/77, 12%) developed AI-, ML-, and DL-based applications without any real-world effectiveness testing [17,24,39,42,47,50,54,87,88].

In Bangladesh, the application of AI, ML, and DL in health care and public health has primarily been confined to risk factors identification, classification of exposed and unexposed disease groups, early danger-sign prediction, trend analysis, and early forecasting of outbreaks on a smaller scale. However, studies from high-income countries have presented the successful implementation of AI, ML, and DL interventions in diverse health care settings as well as mentioned stronger policy evidence [93-96]. For example, scoping reviews mentioned the successful use of BDA for mental health treatment [97], decision support management [98], patient classification, emergency department triage, and hospital resource allocation [99] as well as effective management of virus transmissions such as influenza, Zika virus, AIDS, chikungunya, and Ebola [100]. In Bangladesh, except for 1 policy article on big data, no studies mentioned any call for policies or actions on BDA for the public health domain [11].

While AI and ML have been used to forecast potential outbreaks such as COVID-19, influenza, and dengue in Bangladesh, BDA could be used in such domains for more comprehensive outcomes. Moreover, despite remarkable performances in screening, diagnosis, and prediction of cardiovascular diseases, the use of BDA in clinical settings remains limited [101]. Nonetheless, AI and ML have the potential to improve public health efforts and promote quality health care for all individuals across diverse communities, as mentioned by 1 review article [102].

Research Implications

Our scoping review unveiled that integrating ML algorithms for diagnosis improves accuracy, saving time and effort. Moreover, AI-based decision support systems enhance prescription recommendations, aiding health care professionals in resource-constrained settings such as Bangladesh. Although ML and AI algorithms showed potential regarding outbreak trend prediction and hot spot detection, most studies were conducted on a smaller scale with a smaller sample size and within controlled environments. To gain a better understanding of the effectiveness of AI, ML, and DL algorithm–based prediction models and decision support systems, it is imperative to conduct large-scale studies in diverse environments and real clinical settings. Furthermore, in countries with robust health care systems, ML-based prediction systems tend to be successful in analyzing patient data from electronic health records and generating illness scores for predicting future health care needs [6]. For example, the Cambridge Adjutorium system, developed by a team at the Cambridge Centre for AI in Medicine in the United Kingdom, uses cutting-edge ML techniques to accurately forecast death rates, the necessity for intensive care unit admission, and the need for ventilation in patients with COVID-19 [103]. AI-powered physicians can offer health advice for common medical conditions, reducing hospital visits and staying through digital health coaching. Yet, applying these approaches in lower-income countries such as Bangladesh is limited due to structural, regulatory, and legislative constraints. Further studies and future research are essential to assess the effectiveness of AI-, ML-, and DL-based solutions in genuine clinical and larger population-based settings.

Policy Implications

Technology has become integral to Bangladesh’s health care system, with platforms such as District Health Information System 2, electronic management information system, and open medical record system collecting vast amounts of patient data [12]. Despite challenges in structured data acquisition, data security, and processing capacity, research and implementation of AI, ML, and DL initiatives are expanding, promising substantial benefits for resource-limited countries such as Bangladesh. The government of Bangladesh is actively integrating AI, ML, and knowledge-based systems into health care operations as part of its strategic plan [104]. They are supporting technology industries and small teams in developing such solutions. Moreover, the government of Bangladesh is hosting national and international events and offering small grants for AI-driven health care projects, reflecting a strong commitment to incorporating these technologies despite financial limitations.

Training health care workers and ensuring consistent data entry are crucial to enhancing data integration and interoperability. Implementing AI, ML, and DL pilot projects can help optimize resources and gather insights for incremental and broader implementation. Moreover, collaborating with private technology companies, universities, and international organizations through public-private partnerships can codevelop and fund these initiatives, easing the financial burden on the government.

Furthermore, it is essential to develop robust data privacy and security frameworks suited to local needs and establish standards for the ethical use of AI, ML, and DL technologies. Investing in training health care workers and IT professionals, as well as encouraging local universities to include AI and ML in their curricula, will build a skilled workforce. In addition, targeted funding should prioritize high-return investments such as predictive analytics and AI-assisted diagnostic tools supported by thorough cost-benefit analyses.

Despite budget constraints, these strategic initiatives can potentially promote AI- and ML-based solutions in health care settings. By focusing on these actionable steps, Bangladesh can effectively integrate these technologies, ensuring sustainable advancements in digital health and improved overall health outcomes.

Strengths and Limitations

Our study’s primary strength lies in the rigorous adherence to the scoping review methodology at every stage. This encompassed thorough screening and quality assessment of the studies included in our analysis. To our knowledge, this is the first initiative to outline the extension and utility of AI, ML, DL, and BDA in health care, along with illustrating the potential scope for interventions, research, and policies around such burgeoning domains in Bangladesh, which is also applicable for similar resource-limited settings. However, further studies are needed to investigate and contrast findings from other settings.

Nonetheless, our study does have other limitations. First, we exclusively considered literature in the English language, potentially excluding studies in other languages that could have offered valuable insights into the applications of DL, AI, ML, and BDA. However, it is worth noting that previous research has suggested that language restrictions in scoping reviews do not substantially introduce bias [105]. Moreover, we only included MEDLINE, IEEE Xplore, Scopus, and Embase databases to search the relevant articles, whereas we acknowledge that there might be potential useful insights from gray literature, including government of Bangladesh reports and nongovernmental organization publications. However, our included databases thoroughly cover peer-reviewed literature in medicine, engineering, and public health, aligning closely with our scoping review’s subject matter and ensuring the reliability and credibility of the evidence synthesized. Furthermore, several studies in our analysis involved small population sizes, and the overall quality of many included studies left room for doubt. Consequently, caution is warranted when interpreting our findings. Another limitation pertained to including studies that relied on secondary datasets and global data sources, primarily due to the scarcity of datasets specific to Bangladesh. We excluded 22 (15.4%) of the 143 articles from our review due to inaccessibility, which may have contributed valuable insights to our review. Finally, some studies did not mention explicit information about their ML models. Because our primary objective did not involve assessing the quality of modeling, these studies were included as part of our analysis.

Conclusions

This scoping review outlines available research on BDA, AI, ML, and DL in the health care system sector of Bangladesh. The increasing demand for these cross-cutting technologies for analyzing and processing health data is steadily expanding. The use of ML for disease prediction and classification can aid health care professionals in enhancing diagnosis and treatments. Advanced algorithmic scoring systems based on multiple health determinants have immense potential to predict diagnosis, readmission, and mortality rates in the near future, potentially saving resources and efforts. Therefore, it is crucial to use BDA, AI, ML, and DL technology in designing future health interventions and policies to maximize the use of health resources and maintain optimum competitiveness. Finally, it is crucial to identify priority areas for the application of these technologies and develop a national strategy for the integration of AI and data science. This strategy should include but not be limited to components such as data monitoring, privacy and security, data availability and validation for research, academic and industry collaboration, e-infrastructure establishment, fostering a supportive legal framework, and training data scientists to ensure a successful transition to a future digital health era.