Keywords

1 Introduction

Digital health is an umbrella term describing electronically captured data, along with the technical infrastructure and the applications connecting data producers and consumers. Continual technological advances are transforming healthcare delivery, clinical research, and biomedical science, as well as impacting on the design and delivery of the digital tools within those sectors (Abernethy et al. 2022). Specifically, some of those advances include cloud computing, remote patient management, artificial intelligence-enabled diagnostics, and consumer-facing mobile health applications (Abernethy et al. 2022). Though most implementations are designed for resource-abundant settings, some examples do exist in low- and middle-income countries (LMICs). In Malawi, for example, VillageReach and the Ministry of Health created and utilize a national health information line, called Chipatala cha pa Foni (CCPF). In less than a decade, CCPF developed from providing Reproductive, Maternal, Newborn, Child Health, and Nutrition (RMNCHN) services locally, to the national level and addresses all health topics, including also COVID-19 relevant information (Mitgang et al. 2021; Viamo 2020). Babyl Rwanda is another such example of an integrated digital health solution, as it delivers virtual triage and primary care services over the phone, and the ability to also provide post-consultation prescription and other downstream services. The latter are referred via a text message and are valid at both public and private designated facilities. These developments promise to transform healthcare delivery, by improving accuracy and/or outcomes, and through a more personalized interaction, increase patient engagement (Abernethy et al. 2022; McGinnis et al. 2021).

The digitization of healthcare represents the foundational precondition for enabling the downstream data analyses for quality of care and operational efficiency and effectiveness. However, the advancement of clinical knowledge and diffusion of digital innovations are integrally linked to establishing and maintaining data standards. For example, understanding the medication lists prescribed to each patient and centralizing this information offers the potential to identify adverse reaction between the offered prescribed medications for each patient (Garfield et al. 2020). However, this knowledge needs to be translated into actionable pathways. Information infrastructure is required to capture such information in detail (e.g., pharmacies must add new drugs as they become available, and new prescriptions to individual patient lists), and digital decision-support systems must be adapted to alert the clinician and/or patient if a pre-defined, unexpected event takes place. For example, if the patient’s record does not include entries on potential adverse reactions to particular pharmaceutical susbtances (https://www.measureevaluation.org/resources/publications/wp-18-211/at_download/document). Currently using telemedicine, patients are anticipated to benefit from round-the-clock remote monitoring, in particular for long-term critical illnesses and/or post-operative care. Efforts to enhance communication and information technologies with appropriate health data, are actively being made with high hopes that these can make a significant leap forward toward safer care, although little has been achieved in LMICs. The application of available digital solutions is more likely to be used in the USA, in Europe and other digitally advanced regions for continuous healthcare improvement.

The COVID-19 pandemic has exposed the fragility points of healthcare systems, additional to persistent and deepening inequities (Mitgang et al. 2021). In particular, the limited capacity of LMICs to respond to an evolving pandemic, such as COVID-19, and its impact on the most vulnerable populations presents a marked challenge. Digital health can mitigate some of those challenges, as an alternative communication tool; that is scalable and able to incorporate/combine information service delivery models; thus, empowering healthcare delivery.

2 Methodology

This is a narrative review of publicly accessible information in scientific journals, from the last five years, to identify the opportunities and gaps in the proliferation, ingestion, and interpretation of digital health data in LMICs. To this end, the most highly cited articles identified on the Web of Science and PubMed were used, identified by the keywords: LMIC; ingestion; proliferation; healthcare data; interpretation. The date of search was in the last 5 years (2018–present). Having identified those starting articles, additional manuscripts were identified through ‘snowballing’, i.e., using reverse citation tracking to find articles that cited articles already deemed relevant to the review (Callahan 2014). For this topic, a narrative review approach was preferred, as the aim of this chapter was to provide a broad perspective and explore the general debates and developments. By contrast a systematic review focuses on unique and specific queries, using explicit methodology and a typically a narrower perpsective (Rother 2007).

3 The Proliferation of Digital Health Data in LMICs

The proliferation of health is one of the most important developments in the digitalization of health (Chowdhury and Pick 2019). Improving and scaling-up data collection remain fundamental to all these activity domains: process optimization such as digitalization of medical records, training physicians, and improving the quality of care given to patients (Chowdhury and Pick 2019); preclinical research like reducing the time taken for new drugs to reach patients; clinical pathways; including improving access to healthcare information (Aisyah et al. 2021); patient-facing applications like making applications that are user-friendly for patients (Abusanad 2021); including population-level applications (Chowdhury and Pick 2019; Cheong and Wang 2022).

Data-driven research conducted over time and interpreted within a local context can increase the capability to undertake population-level planning. For example, mobile phone data was used to model the spread of cholera in Haiti (2010) and of dengue fever in Pakistan (2013) (Bengtsson et al. 2015; Wesolowski et al. 2015). The Global Health Monitor is another such example: locating and analyzing English-language news stories as a proxy for monitoring infectious diseases outbreaks (https://www.researchgate.net/publication/239813129_Global_Health_Monitor_-_A_Web-based_System_for_Detecting_and_Mapping_Infectious_Diseases). With machine learning techniques there is the potential to identify, map, and track complex diseases quicker. This can reduce response time and attenuate their impact. Another example of data-driven actions is the continuous development of the Global Antimicrobial Resistance and Use Surveillance System (GLASS) by the World Health Organization (WHO), the former launched in 2015 with the main aim “to promote, enhance and harmonize the surveillance of antimicrobial resistance (AMR)” and inform relevant policy decision. Since its launch, GLASS expanded in multiple directions: in its scope, in the data volumes received, and in its global coverage, as over 100 countries and territories worldwide are no enrolled (World Health Organization 2021a). While this is a great proliferation of data producing and sharing, the financial model is entirely based on the WHO funding to promote the scaling-up. A similar model of data generation is followed by the WHO FluNet, originally established in 1952 as the Global Influenza Surveillance Network (GISN) (Monto 2018). In the past two decades it has incorporated genomic data and was transformed into FluNet, a system of over 100 National and international Influenza Centers (NICs), all consistently recording population-level influenza globally (Brammer et al. 2009). The proliferation of data within this platform, where data and Standard Operating Procedures (SOPs) are publicly available, has been significant, and inclusive of many LMIC-generated data.

The availability of healthcare data (at either a global or local level) is anticipated to improve predictions about the changing demands and the effectiveness of any initiatives taken in response (Altmann-Richer 2018). Health inequalities including access to care and life expectancy amongst others can be difficult to resolve at present due to a lack of reliable data on underrepresented populations in research, such as those with low income and educational levels (Chowdhury and Pick 2019). Adding new technologies to such a context, while desirable in terms of providing new data and identifying potential opportunities to resolve existing challenges, can be problematic, as these technologies need to operate consistently within challenging environments. Therefore, evaluation frameworks are needed in LMICs to determine the impact of introducing data-generating new technologies, such as genomic-based pathways, in these settings both clinically and financially (Roberts et al. 2019). The data from such evaluation frameworks is critical for creating the evidence-base for policymakers and practitioners in LMICs to make better data-based decisions to improve healthcare (https://genomemedicine.biomedcentral.com/articles/10.1186/s13073-021-00911-0). It is also important to note that data-driven systems can mitigate or alternatively entrench inequalities, depending on their implementation (https://www.adalovelaceinstitute.org/report/knotted-pipeline-health-data-inequalities/).

The rapidly expanding quantity of healthcare data poses a significant question in terms of governance of this information across all settings. Information governance presents both risks and opportunities. The weak regulatory frameworks in place in many LMICs are also a challenge (Vodosin et al. 2021) but more than that is the lack of effective governance (Hoxha et al. 2022). The effective governance includes operational aspects of security and safety, as well as the longer-term strategic planning with multiple stakeholders. Inevitably, the latter also requires multi-sectoral coordination, including multiple governmental departments, and a population-level strategy (Chowdhury and Pick 2019).

With the growth of digital health tools and their integration into daily reporting dashboards, data-based decision-making may become the norm. These may confer benefits such as more accurate diagnosis, more efficient resource allocation, and even improved patient outcomes. However, associated risks should also be considered. A potential over-reliance on data-based decision-making may lead to the de-skilling of decision-makers (e.g., clinicians), who may follow algorithms’ outputs uncritically (Duran 2021), sometimes unable to explain the automated pathway underpinning them due to the inability of developers to communicate their methodologies in a clear and non-technical manner. Incomplete and/or inaccurate data input may exacerbate these problems further, introducing bias into the decision-making that is difficult to remove. For example, in an article on the ‘global measles crisis’, the WHO Director-General highlighted the vaccination uncertainty “fueled by the proliferation of confusing and contradictory information online” (Gu et al. 2018; Stahl et al. 2016). Thus, the next focus, following proliferation, is the ingestion of digital health data.

4 Ingestion of Digital Health Data in LMIC

There is a need for digital healthcare information to become findable, interoperable and shareable across organizations, to avoid duplication and magnify the collaborative impact. For example, the Ugandan Ministry of Health in 2012 issued a moratorium on m-Health projects that were unable to share data nor integrated into the health system and were not making lasting contributions to the overall healthcare (World Health Organization 2016). Specialist and secure data warehouses may need to be created to facilitate the ingestion of digital health data on a routine basis (Rudniy 2022). From such silos, data can be used to develop downstream applications, however, the completeness and quality of the collected healthcare data are or paramount importance. If incomplete training data were to be utilized for a downstream artificial intelligence (AI) application and those data do not accurately reflect an entire population across specific parameters (e.g., gender or race), this would result in skewed outcomes for any health application, as recommended treatments may be beneficial only for subsets of the population. In LMICs, healthcare data are often incomplete and of mediocre quality, and these can be ingested into different applications. For example, the real-time generation of evidence in a learning health system (i.e., a system that would integrate the lessons learned at regular intervals), that links datasets, integrates and analyzes those datasets using AI and machine learning remains nascent and limited to a few pilots (Amruthlal et al. 2022; Miguel 2022).

Accurately analyzing the spread of infectious diseases, identifying patient risk factors, and distributing resources effectively necessitate complete and quality data that is currently difficult to obtain in most LMICs through a systemic approach. Having said that, certain parts of the healthcare ecosystem can develop from the pioneering examples. Specifically, electronic medical records, when interoperable, have the potential to support evidence-based decisions at a system level. For example, a Ghana-based telemedicine initiative supported by the Novartis foundation is frequently highlighted as a positive example of scaling up digital healthcare provision in LMICs, from a single district in 2011 to nation-wide coverage in 2016, incorporating end-user insights, such as front-line community healthcare staff (Novartis Foundation 2010). Another case study is the Open Medical Record System (OpenMRS), created in 2004 as an open-source, electronic health records platform. OpenMRS and one of its better-known implementations was during the Ebola epidemic (2014–2016), when it was deployed successfully in a treatment center in Sierra Leone (Chowdhury and Pick 2019). OpenMRS is currently being used in a number of locations globally (India, Haiti, South Africa, Zimbabwe, etc.), with successful examples reported (Uwamariya 2015; Jawhari 2016), and a third release of the updated product (i.e., OpenMRS 3.0) taking place in 2023. However, it has not become as ubiquitous in its LMIC use as originally anticipated.

The data produced and contained within medical records, beyond their immediate clinical use, can also have a secondary use by researchers and/or commercial parties, for example to create new tools, processes or treatments. This adds a second layer of data ingestion requirements. Hence the need to improve the quality of data collected in LMICs; ensuring that end-users, are trained, digitally literate, and have access to use such information. For example, the use of telemedicine in Indonesian hospitals proved successful, especially during the COVID-19 pandemic (Aisyah et al. 2021; Aisyah et al. 2023).

5 Interpretation of Digital Health Data in LMICs

Data interpretation is defined as “the process of reviewing data and arriving at relevant conclusions using various analytical methods” (Spiggle 1994). It goes beyond the identification of patterns that an analytical process will result to, and highlights the reasons behind those observed patterns. A good data interpretation process typically involves integration which is collecting and merging data from multiple sources to create unified sets of information for downstream applications, resulting to findings, conclusions and/or recommendations. Devices that collect data but do not integrate with other extant databases (e.g., at local or national level) are unlikely to be utilized outside of the program for which the data is collected. However, the interpretation process presumes that a good level of understanding exists throughout this cycle of data collection-analysis-interpretation-recommendation. In LMICs the lack of digital literacy has been highlighted on many healthcare reports, in particular as lack of relevant staff training can lead to misinterpretation. This can be prevented, at least in part, by providing clear, standardized operations to facilitate data utilization. For example, in Nigeria, MEASURE Evaluation provided AIDSRelief with a standardized data checklist, this very tight standardization facilitated the integration of family planning and HIV treatment data, and supported evidence-based decision making at the facility level (Chabikuli et al. 2009).

The relatively low level of digital literacy in LMICs, highlights a need for national governments to adapt their educational systems to be more inclusive of digital applications, to lead in the information and sensitization of people about digital health and the importance of collecting accurate and precise data, while providing successful examples so that there is a direct understanding of the benefits to the individual and to society. The digital tools developed need to be adjusted to best fit the population in which they are to be used, and whenever possible with the input of the local end-user groups (Labrique et al. 2018). The 2018 WHO resolution on digital health and the ‘Global Strategy on Digital Health 2020–2025’, urge member states to develop, as appropriate, legislation and/or data protection policies (World Health Organization 2018, 2021b). As individuals are both consumers and producers of data, simultaneously at a personal and a community level, the data interpretation has the potential to affect several facets of their lives. Therefore, developing relevant frameworks for privacy, security, data access and ownership, and consent are essential, if interpretation of digital health data in LMICs is to progress.

6 Discussion: The Way Forward

While technology does not intend to entirely replace human decision-making in healthcare, the vision of precision medicine may become realizable because of the proliferation, ingestion, and interpretation of data. In an ideal scenario, with rich, accessible, high-quality datasets on patient diagnoses, disease treatments, and drug effectiveness, a sustained global growth of personalized treatments could be anticipated, leading to tailored therapies and improved health outcomes (Vogenberg et al. 2010). However, two aspects need to be taken into account. Firstly, precision medicine carries an increased procedural (and perhaps pharmaceutical) cost, as well as the infrastructural burden, and those cost demands could divert funds from elsewhere. The second risk is that precision medicine could become an inequality driver by implementing a barrier to entry for those healthcare systems who cannot afford it.

One initiative addressing the latter risk is the USA-based Digital Square, a partnership between PATH, USAID, the Gates Foundation, and others, that works together with local ministries of health to align digital technologies with local health needs, aiming to improve how healthcare tools are designed, used, and paid for (Novillo-Ortiz et al. 2018). In Europe, the European Open Science Cloud (EOSC) is a European Commission infrastructure providing its users with services for open science practices and digital interoperable environments, including for healthcare research (Budroni et al. 2019). These are high-profile and impactful initiatives but would be slow to produce change in LMICs. Perhaps educational initiatives on core competencies in data analysis, interpretation, synthesis, and presentation would be more applicable for LMICs, especially if staff at all levels of a health system are included in such educational initiatives (https://www.measureevaluation.org/resources/publications/wp-18-211/at_download/document; Amaro et al. 2005; Nutley and Reynolds 2013). Such education activities can be complemented by data quality assessment tools/evaluation framework tools, that are to be used in iterative assessments, including the data quality review (DQR) and the routine data quality assessment (RDQA) (Chen et al. 2014).

The process from data collection and storage to the final interpretation and dissemination to the end-user overgoes different stages, while addressing the need for patients’ rights protection through data sharing and access. Successful integration of the plurality of patient related data, from environmental to metadata, and of many digital systems, within national Electronic Data Capture Systems, requires digital ontology engines and protocols for harmonization, standardization, and homogenization of data (Kush et al. 2020). Thus, the investment required in both technology and trained staff is considerable. Data redundancy, as well as legislative and governance issues can be overcome by establishing robust Digital Rights Management (DRMs) systems, enforcing role-based access, manipulation, and control of data (Hu et al. 2014; Alahmar et al. 2022). As the personalized medicine notion, becomes more prevalent, health related data collection systems implemented for interoperability, secure data migration, mining, and interpretation, should have patients as their focal point and be patient-centric. These can include consent status management, predefined data ownership and accredited security options and create a dynamic electronic framework for future uses, without making it obsolete when faced with emerging technologies (Kaye et al. 2015; Ivanova and Katsaounis 2021). Successful interoperability implementation systems will most definitely lead to scalability; thus, regional and national organizations should be established overlooking and synchronizing actions, adoption of technologies and effective non-overlapping adoption and integration (Austin et al. 2021).

Within this need for creating appropriate frameworks, Sensitive information covered by data privacy and biosecurity must be identified, classified, and protected. There is confidential and exclusive data that can only be accessed by limited individuals (e.g., high-risk pathogen research, and/or personal details). The existence of a reliable infrastructure (e.g., electricity and internet accessibility) is a precondition to the existence of a digital infrastructure and the latter underlies the implementation of diagnosis tools, data analytics, or drug discovery technologies. For example, the internet can lead to greater numbers of trained doctors, nurses, and community health workers, by lowering barriers to access education. Given the low healthcare staff coverage in LMICs, the ability to train healthcare practitioners effectively is vital—and digital solutions can be incorporated within the training courses.

In summary, some recommendations would be closely aligned with those made in the Riyadh Declaration on Digital Health (Al Knawy et al. 2020, 2022).

  • Adoption of standards that encourage data harmonization and eventual exchange and interoperability.

  • Investment in foundational infrastructures, e.g. electricity, to support digital access.

  • Investment in nationwide digital literacy initiatives.

  • Development and implementation of clear, national frameworks for data protection, for sensitive and confidential information.

  • Establishment of governing bodies to oversee the implementation of digital health strategies.

  • Implementation of a digital patient consent

7 Conclusions

In LMICs there is a marked lag in the implementation of digital healthcare applications even though digital healthcare carries the potential to improve overall healthcare provision. Healthcare challenges in LMICs have attracted digital initiatives, which typically are operating siloed, though they all aim to improve access to quality healthcare delivery. Some of these digital initiatives have moved beyond the initial piloting and experimentation phases, and now focus on effective scaling and/or integration with other existing healthcare system operations.

Digital initiatives typically take advantage of existing platforms such as mobile phone networks and devices, combined with health information systems, automated processing and information exchanges. Their focus has been shifting towards data proliferation, ingestion, and interpretation, as critical steps in the development of digitally-enabled healthcare systems. However, digital health interventions are bound by high-quality data, which is not always forthcoming in LMIC settings. Therefore, any investment in infrastructure improvement would need to be complemented by digital literacy training programs, within amenable regulatory frameworks.