Public Health, that is, population-based medicine, has a pivotal role in the development of Artificial Intelligence for health and well-being.
I have witnessed many blank faces as I explain to new friends about the field of Public Health. Countless times people have confused Public Health with Family Medicine (also known as general practice), and with ample enthusiasm I explain to them the crucial difference. I point out that a family doctor will see a few thousand patients over the span of a career, whereas a public health specialist considers the health needs of an entire population at the same time.
This usually takes my friends by surprise and they ask, “How can you be a doctor of 450,000+ people all at the same time?”. The answer deserves a separate conversation, but it is inevitable that Artificial Intelligence must be part of that equation in the very near future.
Artificial intelligence in public health and epidemiology
One of the main ingredients of Artificial Intelligence is machine learning. In China, where the infectious disease dengue fever remains an important public health issue, public health researchers and infectious disease experts have figured out a way to forecast disease spread in the province of Guangdong. They used a combination of case reports, Baidu (China’s top search engine) search queries, and climate factors. Then they compared their working models with existing results from five other provinces, and discovered that their data models showed stronger statistical significance. This is just the beginning.
An epidemiology startup in from Malaysia called A.I.M.E has a single focus on beating the spread of vector-borne diseases. This kind of disease is spread by a vector, which in this particular case is the infamous mosquito. A.I.M.E works with government agencies and non-governmental organizations with their existing data and uses specific algorithms which take into account weather variability, land use, and population density. With this information, they are able to predict the next outbreak with an accuracy of 88.62%.
Is this only happening with infectious diseases spread by mosquitoes? No! Tuberculosis is one of the top ten causes of death worldwide, with more than a million deaths per year. But there is hope: a scientific team in Peru is implementing a system using deep learning and mobile technologies to improve the diagnostic process and detect tuberculosis earlier, at a stage where it is curable, and thus reduce the risk of death to practically zero. They achieved this by using mobile devices to store x-ray images. These images were then sent to a server-based image database which was optimized for automated TB screening and for computation models for categorizations of manifestations. This is known as digital epidemiology, and it is playing a major role in filling in the gaps which are not covered by traditional methods.
Epidemiology and infectious disease specialists work together to define diseases, which helps clinicians working on the ground to diagnose a specific infectious disease. This process is usually quite long and labor-intensive, and requires a large number of cases. In the case of the infamous Zika virus, scientists made use of an incredibly interesting disease detection system, known as GUARDIAN, which is automated and operates in real time. From a practical point of view, as soon as the cases come to the Emergency Department, real-time analysis is performed on various aspects of electronic health records (chief complaints, vital parameters, etc.) and laboratory results. This means that a physician is informed and alerted of a case as soon as it is identified. The main benefit is avoiding delays and preventing further spread of contagious infectious diseases such as influenza, plague and anthrax. Even better, comparison with traditional systems using rigorous scientific protocols has shown it to be the most accurate prediction model.
Healthcare management and policy
Public Health is not just about dealing with infectious diseases outbreaks, but much more. One important aspect of public health, believe it or not, is argumentation. This is not the first time we have tussled with politicians to put forward our public health arguments and ensure the highest quality health for our population.
Cereal products, specifically bread, are used by decision makers as an instrument to fight diseases such as obesity or diabetes. Scientists have succeeded in creating models that output new recommendations based on stakeholders’ arguments by targeting specific audiences. Imagine a consensus that must be reached on risk/benefit evaluation between stakeholders with differing views, such as the massive food industry, government regulators and public health professionals. This workflow allows the consensus to be achieved in a much shorter time and with positive effects, in contrast to a lengthy and mostly bureaucratic process.
Policy has a crucial role in management, and specific policies often influence the way matters are handled within a healthcare system. A change in policy could bring forth various forms of feedback — negative, positive and constructive. What if there was a way to bring all of this together through existing social networks? Sentiment analysis, especially when done across different social networks, can offer insight into the experience of patients making use of various healthcare institutions and organizations.
Way back in 2010, a research team from the UK succeeded in analyzing approximately 6,412 free-text online comments using machine learning. This can be quite useful, in comparison to relying solely on surveys or tedious questionnaires, which take time to fill out and could be biased or not truly reflective of the situation, especially if done in a hurry. The research team developed their viewpoint further and explored patient experience and engagement through Twitter in the US, and has even proposed updates to the way we report on hospital quality.
The vast public health field is ripe with potential; there is so much more to cover and learn. We will be covering disease prevention, behavior change, evidence synthesis and more in Part 2 of this series.