Portal de Periódicos da CAPES

Scientia Potentia Est: Striving for Data Equity in Clinical Medicine for Low- and Middle-Income Countries

2022; Lippincott Williams & Wilkins; Volume: 135; Issue: 1 Linguagem: Inglês

10.1213/ane.0000000000005993

1526-7598

Marcel E. Durieux, Bhiken I. Naik,

Ethics in Clinical Research

Thomas Hobbes, who, in 1668, coined the phrase "scientia potentia est"1—knowledge is power—never could have foreseen the astounding amount of knowledge that the world would acquire over the subsequent centuries. Even less could he have imagined that much of this knowledge, and its associated power, would be in the form of digital data. The world of modern medicine, from scheduling systems to reporting of laboratory results to digitized imaging and physiologic data, is as good an example as any. We gain our power over disease to a large extent from a continuous, deep flow of digital knowledge. And it is not an overstatement to say that our present health systems would cease to function without this digital information stream. Hobbes would have been first in line to point out that the rules of logic do not allow us to conclude from the phrase "knowledge is power" that lack of knowledge equates to weakness. But to any physician, it is completely clear that without our streams of digital clinical data, we would not be able to provide high-level patient care; we could not generate up-to-date, actionable quality improvement information; we could not perform large-scale clinical research (certainly not the newer, big data approaches2); and the emerging systems providing real-time clinical decision support3 would fall silent. In other words, we would not, by any stretch of the imagination, be able to take good care of our patients, in the short term or the long term. Health care providers in low- and middle-income countries (LMICs) do not have access to such streams of digital clinical data. Most record keeping is on paper, which makes it difficult, if not impossible, to access and act on the information that the records contain. The data are there, but are largely out of reach. Those countries have been described as "data rich but information poor."4 They lack accessible knowledge about their own patients and systems. And because of this absence of knowledge, they lack the power to improve their clinical outcomes. Even if other barriers to providing care (training of clinicians, availability of drugs, etc) are addressed, outcomes in LMICs likely cannot be improved to levels seen in high-income countries (HICs) without adequate, timely access to digital clinical data. This discrepancy between having readily accessible, curated, and reproducible data between HICs and LMICs constitutes data inequity. As with most cases of inequity, it worsens as the 2 settings diverge. Health care facilities with fewer resources also have less access to data. A central university hospital in an LMIC may have some computers available in clinical settings and may even install an open-source electronic medical record system in its outpatient facilities. In contrast, a small peripheral district hospital may be without any clinical computing systems and possibly without a single monitor in its operating rooms. Data inequity compared with HICs is greater in the latter setting. Inequity between LMICs and HICs also becomes worse as one moves to areas of clinical care that are more data-dependent. Therefore, the data inequity compared with HIC counterparts is much greater in, say, an LMIC procedural area than in its psychiatry ward. Based on this premise, the greatest level of clinical data inequity between LMICs and HICs likely exists in the perioperative arena, as data density (data generated per patient per time unit) is probably among the highest within an HIC health system. The amounts of data collected during a surgical procedure are remarkable. The Multicenter Perioperative Outcomes Group currently reports having 17 million surgical cases in its database and 36 billion physiologic observations.5 This means that each surgical case generates around 2000 physiologic data points. If a typical procedure lasts 2 hours, that equals 1000 measurements per hour, and this does not even count the enormous number of other perioperative pieces of information collected (preoperative assessments, nursing information, records of equipment and disposables used, etc). All this information is digitized and available for analysis and application to guide care and to assess processes and outcomes. In most LMIC settings, the workload of manual collection forces a much lower acquisition rate (about 50 data points per hour), and little of this is available for digital analysis. This lack of access to actionable perioperative data explains in part why surgical mortality is at least twice as high in sub-Saharan Africa as the global average—despite an overall younger and healthier surgical population.6 One striking example is the African Surgical Outcomes Study, which found that maternal mortality was 0.5%, with complications occurring in 17.4% of mothers, which were predominantly severe intraoperative and postoperative bleeding.7 Of note, the majority of these events were considered avoidable by improved monitoring. As anesthesiologists are the primary producers of the extremely high data density in HIC perioperative areas, global data inequity is likely greatest for our specialty. This provides us with an incentive to take on the challenge of improving data equity in medicine. But how do we even start approaching this worldwide problem? At first glance, it may seem a purely technical issue: installing the necessary hardware and software may seem to be a straightforward solution. But it is not. For example, automated data capture from monitors and other devices, while routine in HICs, is currently not a feasible option in many LMIC settings for a variety of reasons. In addition to the very high cost, it requires a solid information technology infrastructure on site, with dedicated, well-trained personnel. And it requires some standardization of equipment in patient care locations. Those prerequisites typically are not met. Additionally, data capture equipment will need to be specifically designed for the local environment (eg, able to deal with frequent, prolonged power failures). Yes, we should strive for eventual implementation of data capture approaches similar to those used in HICs. But in the setting of a typical LMIC small hospital, with old, donated monitors from various brands (or with no monitors at all), such systems cannot be expected to be in use for decades to come; thus, a conventional solution (implementing in low-resourced environments what has worked in higher-resourced settings) is often not feasible, at least not now. Innovative thinking will be required to address this pervasive problem. One robust digital data system that does exist in LMICs is the cellular phone network. Penetration of mobile phones (eg, in sub-Saharan Africa) is high. For example, Rwanda has >80 mobile cellular subscriptions per 100 inhabitants.8 This suggests a role for mobile technology in addressing the problem. At the same time, proposed solutions cannot impose an additional work burden on already overworked clinical providers in LMICs. For example, a model in which clinical data are manually entered into a tablet app has been used in the research setting,9 but this is unlikely to work in clinical practice, particularly in the clinical areas with high data density and, therefore, greatest data inequity. We are part of a team at the University of Rwanda and the University of Virginia that is developing a system leveraging mobile phone technology to address medical data inequity. Anesthesia providers in Rwanda continue to use their existing method of hand-charting the perioperative record. But immediately after the case, a smartphone image is taken of that record. Deidentified and encrypted, this image is uploaded to cloud storage, where the information is extracted using computer vision, machine learning techniques.10 The resulting digital data can be returned to the provider in near real time ( 5 minutes at a systolic blood pressure of <70 mm Hg might be flagged as at risk for increased 30-day mortality.11 Or parturients on a labor ward can have Modified Early Obstetric Warning Scores12 calculated, and this information can be used to alert the provider in near real time of patients at risk (Figure).Figure.: An approach to improving data equity in low-/middle-income countries. Hand-written medical records in Rwanda are photographed using a smartphone app, and deidentified, and encrypted images are transmitted to a cloud structure (EC2 and S3) accessible by the UVA. Digital data are extracted from the images using computer vision, machine learning AI techniques. Algorithms can be applied to these data and used for multiple purposes. Shown here are 2 examples: (1) population of a quality dashboard that tracks hypothermia, completed MEOWS assessments, and adherence to ASA monitoring standards, and (2) providing early warning to clinical providers when MEOWS scores in a parturient or prevalence of hypotension during a surgical procedure indicate increased risk for a patient. These clinical support results can be returned to the users through the same app. All data shown are fictive and created solely for this illustration. AI indicates artificial intelligence; ASA, American Society of Anesthesiologists; CSV, comma-separated values; MEOWS, Modified Early Obstetric Warning Score; UVA, University of Virginia.This is one approach, but other innovative solutions to improve medical data equity should be investigated. These potentially include development of low-cost, stakeholder-centric electronic health records and creation of open-access data sets for perioperative quality improvement and research. The importance of data equity and the need to harmonize disparate data sources have been recognized by federal funding agencies. Recently, the National Institutes of Health funded multiple projects through the Harnessing Data Science for Health Discovery and Innovation in Africa (DS-I Africa) program.13 This program focuses on 3 key foundational pillars: (1) develop data tools and applications that can be shared and adopted, (2) creation of research hubs to advance population-relevant, affordable, acceptable, and scalable data science solutions to improve health, and (3) advance an open data science platform to develop and maintain a data-sharing gateway for existing resources and new data generated. Funding programs such as these are critical to catalyze research for investigators willing to search for novel solutions to this multidimensional problem. By resolving data inequity, clinicians around the world can improve the health of their patients, supported and buoyed by, as Thomas Hobbes said elsewhere, "delight in the continual and indefatigable generation of knowledge."1 ACKNOWLEDGMENTS We wish to acknowledge our main collaborators on the data digitization project: Dr Paulin Banguti (anesthesiology) at the University of Rwanda and Dr Don Brown (data science) at the University of Virginia. DISCLOSURES Name: Marcel E. Durieux, MD, PhD. Contribution: This author helped draft the article. Name: Bhiken I. Naik, MBBCh. Contribution: This author helped revise the article. This manuscript was handled by: Angela Enright, MB, FRCPC.