AI doesn’t save lives, people do
Digest #74
A partner once jokingly summarized the hype around artificial intelligence (AI) in the global health sector as: “Create AI models, save lives!”. Really though, it’s the people who use AI and data to inform their actions that save and improve the lives of those around them.
The tendency to exaggerate the benefits of technology has led to a narrative that AI will easily solve many of humanity’s biggest problems. Whilst AI can play a role in addressing challenges such as ensuring good health for everyone, it is in a supporting role and only when people have the training and resources to take advantage of it. An AI model might be able to diagnose a disease in patients, or forecast how many patients will attend a hospital each month. But it cannot play the role of the health workers, medication, and facilities that are needed to provide the treatment that will improve patients’ health. In order to unlock their potential to improve people’s lives, data and AI-enabled interventions in healthcare need to be purposefully designed collaboratively with the patients that benefit from them, the people who use them, and the needs of the healthcare system in mind.
Technology alone cannot fix data quality, much less save lives
Since moving from the world of national statistics into global health, I’ve seen a number of data-related problems that, on the surface, appear to be solvable with technology. But the problems are far more challenging than they first seem. Data quality is one prominent example. Firstly, it is important to note that unlike in a national data system where enumerators are trained to be specialists in data collection, in a national health system, doctors and nurses specialize in caring for patients and saving lives. However, in many low and middle income countries (LMICs), healthcare workers are also expected to function as data collectors which poses complex challenges to improving data quality.
Data quality can be improved by upgrading data collection and data management tools alongside providing specialized training to data collectors. In a health system, this is further complicated by it being also necessary to ensure that workers have access to essential medical equipment and supplies, such as blood pressure monitors, to collect the data that is needed. Plus, there must be enough staff to enable health workers to fulfill their primary duty of caring for patients and their secondary duty of collecting and reporting data. Technology can ensure standardized reporting and the efficient compilation of data, but technology cannot solve the problem of incomplete or inaccurate reporting caused by a lack of equipment, training, or staff. That’s why it is essential to focus on the basics of investing in data skills for health care workers and the right equipment to collect healthcare data - which are anyway critical for providing patients with the right healthcare.
An example that frequently crops up in my area of work is that the correct classification of fetal death (miscarriage versus early stillbirth versus late stillbirth) requires that an ultrasound machine and fetal heart rate monitor were used to record key data throughout a woman’s pregnancy. Whilst those machines and trained staff are widely available in high income countries, that is not always the case in LMICs. A lack of reliable data about patients’ health outcomes makes it impossible to ascertain what interventions are or are not effective.
Lack of equipment is not the only problem facing quality healthcare data, human resources also place a huge constraint. High quality data is the basis of useful AI models, and if data quality is compromised, the systems that use them will be too. But even if we are able to magically fix the data collection and quality issues and could train an AI model to identify which babies are at risk of life-threatening illness, if there aren’t enough healthcare professionals or medical equipment and supplies to provide the necessary treatment, what helpful action can be informed by the AI model? Who benefits? We can spend time thinking about how to adapt an AI-powered tool to assist nurses in monitoring babies, by enabling offline functionality and the use of solar-powered chargers, but no technology in the world can replace nurses in an intensive care unit. I once visited a clinic in Zanzibar where a doctor showed me a shiny new tablet with a beautiful data collection and reporting app. She explained that she didn’t have time to use the tablet and fill in reports, gesturing to a crowded waiting room full of patients that she had to attend to that morning, alone, as she was the only clinician on duty (again). Research indicates that her situation is not uncommon.
No individual actor can fix both a health data system and healthcare staffing issues. The task is too vast and complex. But if things remain as they are, with disconnected silos of practitioners focused on technology, data, and AI on one side, and the development of workforce capacity on the other side, the health systems that are already strong enough to benefit from the use of AI will benefit, while the remainder will not.
It does not have to be this way.
Focusing on enabling AI and data to provide equal benefit to all healthcare systems
If we maintain the status quo where complex solutions, such as AI, are promoted among health systems that are not strong enough to reap the benefits, the gap between the most and least developed health systems increases, and resources are wasted on implementing solutions that will not work. This is especially harmful if it causes resources to be diverted away from essential foundational work; in fact the World Health Organization provided guidance that cautions against overestimating the benefits of AI for health, especially when this occurs at the expense of core investments and strategies required to achieve universal health coverage.
What is important is to acknowledge that a health system needs to be sufficiently mature before solutions like AI can be effectively leveraged and judiciously choose to implement those solutions in only the systems where most of the requisite foundations for health AI are in place.
There are several actions the AI and data for health communities can focus on to try to reduce these inequalities. We can contribute to the foundational strengthening of health systems by shifting focus towards the development of solutions that support the training and supportive supervision of staff, or the management of equipment and supplies, rather than just focusing on using AI to directly impact health outcomes. By focusing on what it is these health systems actually need in order to take full advantage of the opportunities offered by AI, we follow through on the commitment to leave no one behind and contribute to closing, rather than widening, the gap between healthcare systems.
Unlocking the potential of data and AI for healthcare will require a major rewrite of the narratives that surround data and AI, as well as the formation of diverse interdisciplinary partnerships that bring the ‘technology, data and AI’ and ‘health systems strengthening’ silos together. We need to shift the emphasis from technology as a silver bullet solution and instead focus on human action and collaboration. Investing in the democratization of data and digital skills for healthcare workers, and for the people managing health systems, is crucial to unlocking the potential. These skills will also improve understanding of the limitations of data and AI so that there is better recognition of what data and AI can not do as well as what they can do.
Finally, I will emphasize the #DataValues Campaign’s call for the need for greater participation in the use of data and AI for healthcare systems. It is only by including the people who are supposed to benefit from health AI (patients) as well as the people who will use the AI (health system staff), that we will ensure that decisions about the use of AI are effective, not just driving the creation of a plethora of unusable AI models. The potential of data and AI to improve lives is great, but realizing that potential requires effective utilization by skilled people, in whom we must focus our investment and action.
 | A guest post by
|