Keywords

1 Introduction

The transformation of healthcare is both a necessity and inevitability. A necessity due to the increasing rates of global disease burden and limitations in terms of availability of affordable and trained staff, especially so in Low- and Middle-Income Countries (LMICs) (Ngwa et al. 2022; Kozlakidis et al. 2020); an inevitability because of the rapid technological developments which eventually integrate with and influence all aspects of healthcare (Nadhamuni et al. 2021; Peiris et al. 2014; Ospina-Pinillos et al. 2021; Gurung et al. 2019). For example, the expansion of remote healthcare coverage, already existing before the COVID-19 pandemic (Greenwood et al. 2014; Aisyah et al. 2020; De San et al. 2013) and multiplied during its duration (Caetano et al. 2020; Liu et al. 2020; Aisyah et al. 2023), has been a major step in this direction. It is anticipated that many of these remote healthcare coverage services will continue post-pandemic (Jazieh and Kozlakidis 2020), albeit with further context-driven customizations, adapted to the local milieu of patients and professionals (Abusanad 2021).

However, for the sustainability of such transformation in the long-term, a number of parameters would need to be considered, such as the scale-up of existing technologies; the security aspects of big and integrated data; and the personal, social and market acceptability. The transformation is most likely to occur upon the confluence of those factors rather than each one acting separately. This chapter, in the form of a perspective article, presents an overview of current advancements and their likely future impact on the transformation of digital healthcare services, with particular focus on LMIC settings.

2 Scaling Up Digital Health

The speed by which avian influenza (Uyeki 2008), zika virus (Garcia et al. 2016) and SARS-CoV-2 (Koelle et al. 2022) spread -and the potential of other infectious diseases exhibiting a similar, rapid global impact in the future- highlights the need for reliable and agile surveillance systems. Such systems, vertical by necessity as they are initially of a single-disease focus, can be established for surveillance of one disease and then be expanded to incorporate others (Zinsstag et al. 2020; Njuguna et al. 2019). The surveillance of infectious diseases is defined as the aggregate of reported positive results from designated clinical laboratories or laboratory networks for specific microorganisms that constitute a public health threat (Zeng et al. 2011). Thus, the routine use of surveillance data would need to be coupled with the ability to scale up the generation, ingestion and interpretation of such data during times of need (presented in detail in a different Chapter of this book). Moreover, as a preamble to such a scale up during times of emergency, this capacity would need to be tested through the ‘exercise’ of model crises, i.e., iterated emergency simulations with the aim of identifying and understanding the critical bottlenecks of existing systems. Identifying any such weak points would likely aid their addressing, thus strengthening existing structures over time. Therefore, there is a need for transparent, standards-based assessment of digital health systems that (as a possible solution) will be guided by a formal assessment across the main activity domains in the field, encompassing technical, clinical/operational and financial aspects (Mathews et al. 2019).

As surveillance technologies can now be mobile-enabled, and deployed at scale to monitor and to flag potential healthcare needs as they emerge in individuals and/or populations, one needs to consider that the generation and ingestion of data will be bi-directional. Specifically, the scaling up of digital health will not only be based on healthcare facilities outputting greater volumes of data through high-throughput analytical platforms (e.g., genomics, metabolomics). It would also involve the patients themselves, generating data (e.g., input for remote monitoring applications) and utilising this data (e.g., using clinical details to input into wellness and lifestyle guidance applications) (Moore 2020). Thus, new digital tools are likely to continue being introduced and integrated within public healthcare, supporting our understanding in an increasingly connected and challenging global environment. The ways in which such tools can be adapted and customised to LMIC settings at a population level, and beyond individual cases, is only now starting to emerge (Labrique et al. 2018).

3 Predictive Algorithms and Synthetic Data

The increasing volume of healthcare data underlies an ever-growing need to develop predictive algorithms that can ingest and translate that data. This will help clinicians treat patients based on their individualised response(s) to care rather than on generalised risk scores. Thus, in order to improve performance of existing algorithms and support the creation of new ones, access to large amounts of diverse and high-quality clinical data is needed. This is especially true for LMIC populations, which are under-represented in existing healthcare data sets, or where collected data can often be of low quality (Curado et al. 2009). Unfortunately, in most settings clinical data is also siloed due to privacy restrictions, and access to them is often limited only to the treating clinician or to clinicians within the same department/institution.

The reasoning behind such stringent data accessibility regulations is based on the premise that algorithms that consume and learn from large amounts of personal data can leak private details pertaining to individuals, which can then be used to discriminate against them specifically. Potential data breaches cannot be entirely prevented with current systems, because of the constant need to access, distribute and utilize information, that provides the opportunity for a deliberate data breach or a spontaneous mistake. According to datalossdb (2015), a platform from the Open Security Foundation (2005–2016), in 2014, approximately 50% of recorded data leakages were in private businesses, ca. 20% in government functions and about 30% in the education and health sectors (Alneyadi et al. 2016). There is the possibility of linking healthcare data to blockchain technology, so as when data actually leaks, it should be fairly straightforward to identify the source of the leak and address it appropriately. However, this technology has not been tested widely (Jayabalan and Jeyanthi 2022). In healthcare this is particularly dangerous, as this data can have irreversible consequences for an individual, or if the data is damaged/deleted, can never be replaced. Thus, a methodology is needed to mitigate the aim to harness data on the one hand, with the requirement to protect patient privacy on the other hand. Possibly, one of the most promising solutions to this need lies in synthetic data.

Synthetic data allows researchers to explore data independently of data protection constraints while maintaining patient privacy, enabling them to potentially share data worldwide. Synthetic data do not contain any of the original data sets (Chen et al. 2021). They have the same format as the original data, and they have identical statistical characteristics as the original individuals/population, across parameters and within subgroups in the population. All of this makes synthetic data similarly suitable for analysis, while at the same time overcoming privacy concerns.

Some of the uses of synthetic data in healthcare are:

  • They can simplify the collaborative and regulatory efforts when trying to share raw data.

  • Synthetic data platforms can facilitate hypothesis testing and model validation without intermediaries (Foraker et al. 2020)

  • Synthetic data can be used to train students and staff on new platforms (prior to using such platforms on real data) and be used to host hackathons and competitions improving existing platforms (Gonzales et al. 2023)

  • And finally, synthetic data can liberate data publicly, allowing access for scientists, citizen scientists and clinicians (even from different locations globally) to use those data freely in order to develop better care pathways (Benaim et al. 2020; Foraker et al. 2021)

For example, MDClone, a self-service data analytics environment, has developed a platform for querying and synthesising patient cohorts in a self-service manner. Specifically, a user can query an organisation’s data lake while being sequestered from it at all times, and subsequently create synthetic derivatives of the cohort and its corresponding characteristics. This new technology has been used on a great number of recent clinical studies (Masarweh 2019; Inbar and Dann 2019; Hochberg 2018; Meilik et al. 2022; Hod et al. 2023; Masarweh et al. 2021; Isenberg et al. 2022) and has the potential to model/‘re-create’ LMIC-specific data sets, while maintaining the data security requirements for the real data.

4 A Sustainable Path Forward

As with many other technologies introduced to the healthcare field, the operational advantages for any technology by themselves are unable to guarantee a long-term adoption. Instead, the operational aspects need to be complemented by the social and market acceptability of any new technologies. This definition of sustainability along three axes, the operational, financial, and social, has been successfully applied previously on other large, data-heavy infrastructures, such as biobanking, and can be extrapolated for the digital healthcare data needs, as a useful planning model (Table 1) (Watson et al. 2014; Henderson et al. 2015). For example, in terms of operational sustainability, the ever-growing need to produce and consume data, will introduce additional infrastructure requirements in LMICs in terms of data storage and security, staff training and integrations of systems (Kumar and Mostafa 2019; Labrique et al. 2018). Thus, future data infrastructure approaches in healthcare would need to be evaluated so that they align with global healthcare data requirements (to maintain a global connectivity and interaction) (Al Knawy et al. 2020), but also to design new approaches/data architectures that would be more appropriate for local needs/capacities.

Table 1 Summary of the future challenges and developments along three sustainability axes
Full size table

An example for the successful re-design of data architecture, customised for LMICs, comes from the field of construction (which is also ‘data heavy’). In those examples, data flows were adapted to local needs with the aim of reducing costs and infrastructure pressures, while maintaining data output (Raes et al. 2021; Liu et al. 2021). This approach for transforming infrastructure costs was based on the concept of digital twins, i.e., a digital model of a physical entity that results in measurable outputs. The digital twins of existing models were used for example in the re-design of modular construction systems, allowing for a more context-adaptable output, in this particular case a quicker on-site assembly of the construction (Jiang et al. 2022). In terms of healthcare, digital twins can relate both to the physical infrastructure (i.e., a new methodology to enhance the infrastructure creation in LMICs), as well as the digital infrastructure (i.e., allow for the creation of alternative data pathways to identify the optimal one for a particular context).

The financial sustainability aspect, inevitably would align to and reflect market-driven needs. The potential structures and needs of financial incentives were presented in detail in chapter “Universal Internet Access Supporting Healthcare Provision: The Example of Indonesia” of this book. While the hard digital infrastructure (i.e., hardware) has reduced in costs considerably over the last two decades, the soft digital infrastructure (i.e., software) follows a different pricing structure, often developed as a Software as a Service (SaaS) model (Berndt et al. 2012; Oh et al. 2015). There have been a few individual SaaS implementations within LMICs (Ogwel et al. 2022; Karthikeyan and Sukanesh 2012), however, a more universal understanding or model has not emerged as yet. As the investment incentives have been discussed in chapter “Universal Internet Access Supporting Healthcare Provision: The Example of Indonesia”, a repetition of the information would be avoided here. The only additional aspect that would come into consideration however, in terms of investment, is the necessary investment in trained staff, that is necessary for all of the described infrastructure to be maintained as operational and impactful (Curioso 2019; Long et al. 2018).

Finally, in terms of social and market sustainability, the collection and potential distribution of immense amounts of information regarding individuals (e.g., even if self-reported via social networks) raises ethical concerns, as complete data anonymization is rendered ineffective in concealing the original data source, becoming harded, yet still feasible, to (re)identify individuals via the use of advanced systems and triangulation (Cecaj et al. 2016). However, if systems are entirely designed to use anonymized data, as an effort to protect individuals or population groups, this approach might not work optimally either, as the elements of information accountability and, hence, transparency may be affected. Regarding infectious disease outbreaks the use of anonymous data at source, is considered as current best practice, however, it is not a definitive solution for all situations that might arise within a healthcare ecosystem (Coltart et al. 2018). Therefore, challenges still remain in terms of ethics, as well as in terms of the legal framework for handling large healthcare datasets, including for example “credentialing, licensing, reimbursement, and issues related to technology, security, privacy, safety, and litigations” (Jazieh and Kozlakidis 2020). At this point a distinction would need to be made between public health ethics and clinical ethics: the former prioritizes common good; the latter prioritizes individual autonomy and ways to safeguard it (Chia and Oyeniran 2020). These nuances in ethical views/priorities may come into sharp focus within LMICs, during the implementation of digital healthcare technologies, where the pressures on availability of staff and funding are consistently acute.

Finally, data protection regulations are emerging within LMICs, albeit at a slow pace (Vodosin et al. 2021). The emergence of such frameworks is desirable from a market perspective, as they delineate the extent to which digital health can be implemented, systems integrated and reports provided to competent authorities. A question that is currently often discussed revolves around the regulation of algorithms that evolve (e.g., artificial intelligence (AI)-based applications) (Reddy et al. 2020; Amann et al. 2020). The approach currently proposed, including for LMICs, is that of a ‘reasonable explainability’ for regulating AI in healthcare, i.e. addressing explainability requirements based on the risks involved and providing explanations based on input, process and output norms (Sharma et al. 2020). Even at this level, the training of experts would be a necessity, in particular of regulatory authorities, as well as users of the AI-based solution (i.e., medical practitioners, nurses) on the limitations of explainability, and the risks that may not be explainable, which need to be communicated with patients. Thus, future developments within LMICs are anticipated to incorporate the development and implementation of healthcare digitization applications.

5 Conclusion

The current SARS-CoV-2 pandemic highlighted limitations and vulnerabilities of health systems and has driven a review of many healthcare systems, so that lessons are learned. The need for an increased capacity of healthcare systems to respond to iterative infectious disease emergencies, as well as systemic pressures, creates a driving force for the transformation of healthcare systems, based on new digital technologies. The examples over the last few decades of new technologies that were introduced, integrated and worked well within healthcare, including within LMICs, constitute the benchmark for an even greater integration of digital technologies. Hopefully this can be achieved as part of routine healthcare services design and procurement.

However, current challenges pertain to the scaling up of digital healthcare technologies, post-introduction in the field, and the use of predictive algorithms. Solutions to these challenges have already emerged, such as synthetic data, which allows the use of high-quality datasets without compromising the security of the original datasets. However, for that to be achieved, the sustainability of digitalization of healthcare in LMICs needs to be considered through the lens of infrastructural, financial, ethical and regulatory concerns.