The Impact Of Machine Learning On Public Health

Machine Learning (ML) can be valuable for public health beyond just being a novel technology. ML can assist in analysing the massive quantities of health-related data as it has the ability to recognises patterns that would be difficult (or near impossible) for people to identify.

The JMIR Public Health and Surveillance recently published a significant paper that provides guidelines for the use of ML in Public Health. The guidelines provide insight on the current implementations of ML within the NHS and other UK health-related agencies.

What is Machine Learning?

Artificial intelligence has numerous branches, one of which is called machine learning. Computer systems that use machine learning are able to learn and grow from analysing their own from data and therefore improve their performance. These advanced systems allow machine learning models to recognise patterns in data sets and use them to make better-informed predictions.

Machine Learning in Public Health Settings

Disease surveillance and health trend predictions, as well as policy support, have all greatly benefitted from machine learning in public health. The analysation of population data, which is created from electronic health records, testing results and social indicators, is invaluable. This data allows machine learning models to find and identify emerging health risks and forecast the demand for health services, as well as having the capability of evaluating the effectiveness of health interventions.

The UK’s Rapid Expansion into Public Health Machine Learning

The UK is embracing the use of machine learning within public health, with one example being a machine learning algorithm designed to prevent strokes by locating unknown atrial fibrillation in patient records. This system is predicted to potentially reduce tens of thousands of serious events annually.

There are also government initiatives underway in the country with a newly announced artificial intelligence specialist  team and ethical AI infrastructure to enhance transport, public safety and health with funding from Meta.

Why is ML Important to Public Health?

ML uses algorithms to analyse data and predict future trends in order to assist public health officials in:

• Recognising potential disease outbreak sites

• Projecting the future prevalence of chronic diseases, and;

• Assessing the effects of public health initiatives.

Five Key Principles for Ethical and Effective Use

Machine learning is a powerful tool and should be used responsibly, especially when it comes to the health of a population.

1. Conduct Bias Risk Assessments

Models must be evaluated rigorously for bias throughout the life cycle of the model, from design to development and to deployment and beyond. UK AI Regulatory Principles also emphasise fairness and transparency in the use of AI in public sector services.

2. Use ML Responsibly in Fast-Moving Situations

While ML is valuable in rapidly evolving situations such as public health emergencies, including epidemics, where timely insights are needed, speed should not compromise ethical and privacy protections.

3. Share Data Sources and Methods Transparently

Model users are more likely to trust the data the models are trained on if it is publicly available. This is particularly the case with clinicians, policymakers and the general public. Therefore, to increase reproducibility and trust, developers must be transparent and explicit about the logic of their models.

4. Prioritise Equity and Underserved Populations

ML must assist, not harm, already vulnerable populations. If a model is trained on biased data, it can reproduce health inequities. This serves as a warning to all organisations, including NHS teams, to assess model fairness before it is deployed.

5. Encourage Multidisciplinary Teams

Due to the challenges in public health being multi-faceted,  the application of ML requires multi-disciplinary methodologies from fields such as, statistics, computing, sociology and ethics.

The Challenges Machine Learning in Health

The most notable challenges in the implementation of machine learning in public health seem to be ethical and structural that are helping shape the use of these systems within the public health care framework:

Fragmentation and Quality of Data

The effectiveness of machine learning in public health is determined by multiple factors, including the quality of data being used. Health data can be insufficient, inconsistent or dispersed among various organisations and areas. Differences in health data recording can diminish model accuracy and reduce scalability, especially in implementations at national level use.

Inequalities in Health and Bias

Reinforcement of current health inequalities is one of the greatest dangers cited by researchers. If a machine learning model’s training data is incomplete and in particular, certain groups such as ethnic minorities or people from lower socioeconomic backgrounds are under-represented, the algorithms could cause more harm than good.

The Intersection of Privacy, Public Trust and Consent

The use of population-level data to inform public health machine learning applications raises concerns regarding the privacy and consent of the individuals whose data is being used. Access and use of health data, including anonymised data, can impact public confidence regardless of the intended use of the data.

From Pilots to Scale

Many machine learning tools demonstrate strong performance in pilot studies, however, they often experience challenges when implemented across larger populations. Variability in data, demographic characteristics and infrastructure can all negatively impact performance over time.

The Readiness of Skills and Workforce

The rapid advancement of technology can also be a barrier when it is disconnected from practical applications. It is not uncommon for public health practitioners to lack the skills to analyse the outputs of ML systems or fully assess the shortcomings of the models. Educating and fostering collaboration across various disciplines is critical to ensuring that machine learning simplifies decision-making, rather than complicating it.