Date Published: January 06, 2018
Publisher: The American Society of Tropical Medicine and Hygiene
Author(s): David Bell, Noni Gachuhi, Nassim Assefi.
Most health care in low-income countries is delivered at a primary care level by health workers who lack quality training and supervision, often distant from more experienced support. Lack of knowledge and poor communication result in a poor quality of care and inefficient delivery of health services. Although bringing great benefits in sectors such as finance and telecommunication in recent years, the Digital Revolution has lightly and inconsistently affected the health sector. These advances offer an opportunity to dramatically transform health care by increasing the availability and timeliness of information to augment clinical decision-making, based on improved access to patient histories, current information on disease epidemiology, and improved incorporation of data from point-of-care and centralized diagnostic testing. A comprehensive approach is needed to more effectively incorporate current digital technologies into health systems, bringing external and patient-derived data into the clinical decision-making process in real time, irrespective of health worker training or location. Such dynamic clinical algorithms could provide a more effective framework within which to design and integrate new digital health technologies and deliver improved patient care by primary care health workers.
Two of the greatest obstacles to high-quality primary health care in low-income countries are a lack of skilled health workers and the limited access to reliable, actionable health information. Even where clinicians exist with sufficient skills, health information systems are rarely organized to collate and deliver feedback on health trends to the provider. Medical records have minimal content and may be difficult to access, and diagnostic results may arrive too late to influence the therapy.1–3 Communication gaps between health providers caring for the same patient result in fractured care. Barriers, such as limited capacity to collect and use data or accountability for performance, can severely limit the ability of health workers to integrate local epidemiology and real-time data on diseases.4 Although systems are moving toward digital transfer and collation of data centrally, processes that feed back to influence management are often cumbersome and restricted by rigid policies and guidelines. These barriers, which within the context of an already inefficient paper-based system, reduce the benefit patients might gain from data on current conditions (such as seasonal prevalence, disease outbreaks, and drifts in pathogen prevalence and drug susceptibility).
To transform medical care in low-income countries into a data-driven, logical, and optimized decision-making process, a DCA system will require, at minimum, an accurate means of patient identification, digitized health care information, and a connection with referral, specialized and supervisory levels of the health care system. A DCA system would link patient data to the decision-making process through three broad mechanisms (Figure 1). First, it requires a universal, de-identified clinical database that integrates local epidemiologic data (outbreaks, antimicrobial susceptibility, seasonality, and other historic and geographic trends), with health policy changes, using a search engine in near real time. A second level retains pertinent patient-specific background data (medical history, current medications, demographics, etc., known as EMR) linked to a biometric signature. Monitoring the data of patients who are mobile could be linked to such a system.
In low-income countries, infectious diseases, often in conjunction with underlying nutritional and physical vulnerabilities, are a major cause of morbidity and mortality. The risk of disease often depends on local prevalence, which is a factor of many elements including season, living conditions, environmental change, animal vectors, and population immunity. Diseases such as dengue or malaria, for instance, should only be considered a high probability during certain seasons, and health worker’s responses, including diagnostic testing, should be informed by epidemiology and logically reflect this (Figure 2). Current generations of real-time surveillance systems, which often use mobile health technology, offer the possibility of improving clinical decision-making by incorporating the most up-to-date information about local disease incidence into diagnostic algorithms.25 Although a deterministic human-driven approach could initially be necessary for the system to be acceptable, it is reasonable to assume that future improvements in data quality at input and acquired confidence in electronic systems could enable AI-driven algorithms to recognize patterns and automate changes that would appear as recommendations on a health workers’ screen without direct human intervention.
Responsive AI-based management algorithms will require the expertise of data inputters, data architects, data mining specialists, predictive modeling, and machine learning experts, and user interface designers to create engaging, intuitive software and robust novel algorithms that meet the needs of a changing health environment. Clinical, public health, and engineering teams will need to collaborate to generate algorithms that consider patient data matched against the outcome of interest, as well as the relative value of interacting data in the match process. These algorithms should be modifiable to reflect the changing nature of medicine and the dynamic nature of the data and use only the requisite patient data as to not overburden the health worker’s workflow. Of critical importance to the effectiveness of computerized algorithms is an automated function, which automatically prompts the user to use the system, and in addition, suggests the clinical action that the health worker should take. The algorithms and design would require localization—at a minimum to each region, language, facility and cadre of health worker. Finally, the system will need to work within certain parameters of variability set by the central health authority and be open to direction from an experienced on-site or remote clinician as health authorities, experienced clinicians, and patients will not readily accept handing over full decision-making power to a machine.
Accurate patient identification and consistent linkage of patient data into the population health database are fundamental to the underlying operating assumptions of a successful DCA system. Biometric recognition technology is a promising way of identifying individuals in countries with inconsistent record keeping and patient tracking, including mobile populations of internally displaced people and migrants. The source of biometric identification is the patient’s unique anatomical, physiological, and/or behavioral features, such as fingerprint, face, iris, retina, palm, ear geometry or acoustics, vein pattern recognition, gait, odor, electrocardiogram, signature, and voice.27 Identification occurs by comparing a biometric sample obtained from the subject with a set of records stored in a database (Figure 3). Unlike usual identification methods based on patient documents (a passport, identification card, bank card, etc.) or memory (personal identification number, Social Security Number, password), biometrics allows for rapid identification of a person independent of patient or health worker recall and can be consistent across time.
The wave of rapidly evolving digital technology sweeping other industries presents an opportunity to create evidence-based, health care algorithms, where decisions are made with the input of solid data, predictive analytics, and documented outcomes, rather than on individual experience and inconsistent use of applicable informational resources and norms. DCAs could eventually empower the frontline health worker, providing an augmented ability to diagnose and treat diseases based on evidence and changing conditions and contribute to bridging the huge disparities in health care that currently exist. Although such a transformation will take time, the necessary components already exist and have already penetrated other sectors of society. The global health community has an opportunity to shape that transformation today and speed its delivery through a thoughtful and comprehensive approach to designing and delivering innovative health management systems.
1.DCAs based on real-time data and changing epidemiological information could transform health care in low-income countries by using existing and emerging data sources and data-management technologies to drive logical and potentially self-learning decision-making algorithms.2.A DCA system will require an accurate means of unique patient identification (ideally, through biometrics), digitized medical and health data, and a connection with supervisory, referral, and policy levels of the health care system.3.DCA software development will require the expertise of a multidisciplinary team of clinical, policy, public health, communication, and data engineering experts. However, there is no technology gap preventing such an evidence-based system from impacting health worker practice.