Introduction

Artificial Intelligence (AI) is a topic that always seems to draw mixed reactions from everyone. Some people love AI, whilst others dislike it and believe that it is taking over everything. Whatever individual opinions may be, there is no denying that AI is becoming more prevalent, and that there is a keen interest in research on how it can be utilised more in healthcare. Similarly, Machine Learning (ML) can also be subject to varied perceptions due to its association with AI. In this article, we will look in depth at predictive analytics, a subset of ML, in healthcare and discuss its applications, potential risks and limitations associated with it.

Machine learning in healthcare

Artificial intelligence (AI) is a technology that allows computers to imitate human thinking. There are two key areas within AI: Machine Learning (ML) and Deep Learning (DL). 

ML is like teaching computers to learn from examples and make decisions based on patterns they find in data. For instance, if you show a computer many pictures of cats and dogs, it can learn to recognise whether a new picture is of a cat or a dog. It's all about helping computers make smart guesses, even when they don't have all the information.^1,2

DL is a more advanced type of machine learning. It focuses on creating models that can understand and process complex information in a way that's even closer to how humans think. This involves using special kinds of algorithms, called Neural Networks (NN), which are inspired by the way the human brain works. These NN can handle large amounts of data and learn to perform tasks like recognising speech, identifying images, and even playing games at a very high level.^1,2

ML in healthcare thus seeks to enhance patient outcomes by analysing various forms of data. This data may contain genetic information, medical imaging, notes from doctor visits, and basic information such as age, gender, and medical history.^2,3

Researchers are particularly interested in applying ML to treat nervous system disorders, heart diseases, and cancer. However, ML has the potential to benefit many other aspects of healthcare. For example, ML can help doctors diagnose diseases more correctly, forecast patient outcomes, personalise treatments, and even manage healthcare systems more effectively.³

Simply said, ML enables doctors and healthcare professionals to make better decisions and deliver better treatment to patients by leveraging the massive quantity of information accessible.

What are predictive analytics?

Predictive analytics, a subset of ML, uses statistical techniques, machine learning algorithms and data mining to analyse past and current data to predict future events or outcomes. A notable example of this is the COVID-19 case number prediction.^2,7

Besides managing infectious disease outbreaks, some key benefits of using predictive analytics in healthcare include predicting which patients are at risk of developing certain diseases, tracking the progression of the disease, and responding to treatments, which can help improve recovery rates.²

Applications of predictive analytics

In public health

Using data to predict trends in widespread diseases can be used for improving public health. As mentioned earlier, a good example of the use of predictive analytics is the COVID-19 case number prediction.⁷ Predictive analytics can use data from various sources to forecast outbreaks of infectious diseases and also their spread. This information is used to implement preventive measures, allocate resources, inform the public and plan to reduce the outbreak's social and economic impacts.^2,4

In hospitals

Predictive analytics was used for forecasting the impacts of COVID-19 on hospitals during the pandemic.⁸ This allowed the hospitals to know what to expect and adapt accordingly so that they were better equipped to deal with patients. This is done by analysing patient flow, optimising scheduling and reducing wait times by predicting when demand will be at the highest.⁴

Analysing the rate of hospital-acquired infections and predicting which patients are most at risk of getting infected is also another use of predictive analytics in hospitals as it contributes towards the effective management of quality patient care in hospitals.⁴

Pharmacometrics

Pharmacometrics is an area of study that uses mathematical models and computer simulations to better understand how medicines act in the body. It involves studying how medicines travel through the body, how they impact people, and how variables such as age, weight, and genetics affect drug efficacy and safety. Thus, it assists in determining the appropriate dose for each individual, as well as researchers in developing better and safer medicines.

Hence, another application of predictive analytics is using models of diseases, drugs, and clinical trials to support decision-making in drug development and regulation. If a drug is believed to be potentially effective, it is more cost-effective to analyse how likely it is to work using simulations on computers first. This helps save lab resources and time. If the results are promising, the drug is then further tested in clinical trials.⁵

Clinical decision making

Predictive analytics may assist doctors in clinical decision-making and reducing human errors. Patients with common traits can be grouped together to aid doctors in making decisions on the best course of treatment to improve patient safety. It is also used to make informed decisions by forecasting the likelihood of adverse drug reactions (ADR) based on patient data. For example, it is used to predict and analyse the results of stroke treatment in individuals.³

Personalised medicine and predicting disease progression

Traditional risk assessment methods for predicting patient outcomes may lack accuracy. Instead, ML can be used as a more accurate method to predict patient outcomes. Using data such as diagnostic tests, demographics and medical history as inputs, patterns can be identified using advanced algorithms and ML. This would allow patients with common patterns to be grouped and have treatment plans that are more tailored to them. This is known as personalised medicine. All this leads to an increased rate of recovery from disease.²

Hypertension is an example

Hypertension, commonly known as high blood pressure, is very common and can be caused by many factors, such as cardiovascular diseases and stress from personal relationships. It occurs due to consistently high blood pressure.¹

Although it affects approximately 1.13 billion people globally, with an annual healthcare cost of $370 billion, many individuals with hypertension are unaware that they have it.¹ 

Research suggests that using predictive analytics can help to predict hypertension in individuals and also help in early diagnoses.¹ 

However, there are still improvements to be made to this technology to make it more accurate, which will help reduce the number of individuals who go undetected with hypertension. It needs to be evaluated in a real clinical setting with actual patients before it can be used more extensively.¹ 

Other examples of predictive analytics being used in healthcare include image processing from breast cancer screenings to anticipate patients and exploring genomic data to identify new gene types in cardiovascular conditions.¹

Potential risks and drawbacks

Added responsibilities

• Data Privacy: Predictive analytics involves the handling of sensitive patient information. Robust security measures need to be in place to safeguard patient information and prevent privacy breaches.²
• Ethics: Predictive analytics raises ethical issues on the potential for discrimination. Hence, ethical guidelines and transparent decision-making need to be strictly enforced to address these concerns.² 

Limitations 

High expectations but being unable to deliver on them is an issue that is quite common with ML technologies like predictive analytics. For example, accurate and up-to-date data is essential for reliable predictions. The data also needs to be diverse and representative to avoid bias.⁶

There are many predictive models being developed these days. However, it can be difficult to understand the value a model has. There needs to be ways in which this value can be assessed for safe, sustainable and effective use. Additionally, these models should support, and not replace, clinical judgement.⁶

There is always a risk of spending lots of money on developing and maintaining predictive models without having the benefits in care and outcome for patients. They are also complex and costly to implement on a grassroots level in healthcare.⁶ 

Summary

Overall, predictive analytics have many useful applications in healthcare. It has many benefits like forecasting disease outbreaks, drug development and clinical decision-making. Though it has many benefits, ethical considerations about patient data privacy persist. More ways need to be devised by which predictive analytical models can be assessed to ensure they meet the purpose for which they have been designed.