Machine learning (ML) is not just a new technology; it also has value for public health. ML helps analyze large amounts of health-related data because of its ability to recognize patterns that are difficult (or nearly impossible) for humans to identify.
JMIR Public Health and Surveillance recently published an important paper providing guidelines for the use of ML in public health. This guideline provides insight into the current implementation of ML in the NHS and other UK healthcare organizations.
What is machine learning?
Artificial intelligence has many branches, one of which is called machine learning. Computer systems that use machine learning can learn and grow by performing their own analysis on data, improving performance. These advanced systems allow machine learning models to recognize patterns in datasets and use them to make better-informed predictions.
Machine learning in public health settings
Disease surveillance, health trend prediction, and policy support all benefit greatly from machine learning in public health. Analysis of population data generated from electronic medical records, test results, and social indicators is invaluable. This data enables machine learning models to discover and identify new health risks, predict demand for health services, and even evaluate the effectiveness of health interventions.
Rapid expansion into public health machine learning in the UK
The UK has embraced the use of machine learning in public health, with one example being a machine learning algorithm designed to prevent strokes by identifying unknown cases of atrial fibrillation in patient records. It is estimated that this system could potentially reduce tens of thousands of critical events annually.
The country also has a government initiative underway with a newly announced team of artificial intelligence experts and ethical AI infrastructure to strengthen transportation, public safety and health with funding from Meta.
Why is ML important for public health?
ML helps public health authorities by using algorithms to analyze data and predict future trends.
• Recognize where potential disease outbreaks occur.
• Predict future prevalence of chronic diseases.
• Evaluate the effectiveness of public health efforts.
Five key principles for ethical and effective use
Machine learning is a powerful tool and must be used responsibly, especially when it comes to population health.
1. Conducting bias risk assessment
Models must be rigorously evaluated for bias throughout their lifecycle, from design to development to deployment and beyond. The UK AI Regulatory Principles also focus on fairness and transparency in the use of AI in public sector services.
2. Use ML responsibly in fast-moving situations
ML can be valuable in rapidly evolving situations that require timely insights, such as public health emergencies, including epidemics, but speed must not come at the expense of protecting ethics and privacy.
3. Transparently share data sources and methods
Users of a model are more likely to trust the data on which the model was trained if it is publicly available. This is especially true for clinicians, policy makers, and the general public. Therefore, developers need to be transparent and explicit about the logic of their models to increase reproducibility and reliability.
4. Prioritize equity and underserved populations
ML needs to help. do no harman already vulnerable population. Training models based on biased data can reproduce health inequalities. This is a warning to all organizations, including NHS teams, to assess the fairness of their models before rolling them out.
5. Encourage interdisciplinary teams
Because challenges in public health are multifaceted, applying ML requires interdisciplinary methodologies from fields such as statistics, computing, sociology, and ethics.
Machine learning challenges in health
The most notable challenges in implementing machine learning in public health appear to be the ethical and structural ones that are helping to shape the use of these systems within the public health framework.
Data fragmentation and quality
The effectiveness of machine learning in public health depends on multiple factors, including the quality of the data used. Health data can be insufficient, inconsistent, or dispersed across different organizations and disciplines. Differences in health data records can reduce model accuracy and reduce scalability, especially in implementation for use at the national level.
health inequalities and stigma
The current widening of health disparities is one of the biggest dangers cited by researchers. When training data for machine learning models is incomplete and certain groups are underrepresented, especially ethnic minorities or people from low socio-economic backgrounds, algorithms can do 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 about the privacy and consent of the individuals whose data is used. Access to and use of health data, including de-identified data, can impact public trust, regardless of how the data is used.
From pilot to scale
Although many machine learning tools have shown good performance in pilot studies, challenges often arise when implementing them in larger populations. Variations in data, demographic characteristics, and infrastructure can all negatively impact performance over time.
Skills and workforce readiness
Rapid advances in technology can also create barriers when disconnected from real-world applications. Public health workers often lack the skills to analyze the output of ML systems or fully evaluate model shortcomings. To ensure that machine learning simplifies decision-making rather than complicates it, it is important to educate and encourage collaboration across disciplines.
