Healthcare has traditionally relied on broad clinical guidelines, population-level evidence and a clinician's professional judgement to determine how patients should be diagnosed and treated. These foundations remain essential, but modern healthcare now produces an extraordinary amount of individual-level information. Medical histories, laboratory results, imaging, genomic data, medication records and information from wearable devices can all contribute to a more detailed picture of a person's health. Artificial intelligence is increasingly being used to make sense of this information and support more individualised approaches to care.
The growing adoption of Artificial Intelligence in Healthcare reflects a wider shift towards data-informed and personalised medicine. The World Health Organization describes AI as having applications across diagnosis, treatment, research, drug development and public-health functions, while its recent work on precision medicine emphasises the use of clinical, molecular, genomic and other health data to account for differences between individuals. The potential is considerable, but so are the questions surrounding accuracy, privacy, fairness, transparency and clinical responsibility.
What personalised healthcare actually means
Personalised healthcare is often described as care tailored to an individual rather than applying exactly the same approach to everyone with a particular condition.
That does not necessarily mean that every patient receives a completely different treatment. Established clinical guidelines remain important because they are based on evidence gathered from large populations. Personalisation instead involves using additional information about an individual to help determine which available approach may be most appropriate.
For example, two people may have the same diagnosis but differ in age, medical history, genetics, existing medications, lifestyle or other health characteristics. Those differences can affect how a person responds to a treatment or how likely they are to experience certain complications.
Precision medicine takes this concept further by incorporating different forms of health information when making decisions. The World Health Assembly endorsed a resolution on precision medicine in 2026, describing it as an approach that can use clinical, molecular, genomic and other health data to inform prevention, diagnosis and treatment while taking ethical and legal safeguards into account.
AI can help process this complex information and identify patterns that would be difficult to assess manually.
How AI supports individual treatment decisions
A major strength of AI is its ability to work with large and complex datasets.
Machine-learning systems can be trained to recognise relationships between patient characteristics and clinical outcomes. When properly developed and validated, such systems can provide clinicians with predictions or decision-support information.
Consider a patient being treated for a chronic disease. A clinician may need to consider previous treatments, laboratory measurements, other conditions and changes in symptoms before deciding whether the current treatment remains appropriate.
An AI system could analyse those variables alongside information from comparable cases and identify patterns associated with particular outcomes.
The result is not necessarily an automatic treatment decision. In a responsible clinical setting, the information can instead form one part of the clinician's assessment.
This distinction is important. AI can support clinical reasoning, but it should not be treated as an infallible substitute for medical expertise.
The US Food and Drug Administration notes that AI and machine learning can produce diagnostic, therapeutic and prognostic recommendations, including applications involving personalised diagnostics and monitoring of treatment response.
Genomic information and more precise treatment
Genomics is one of the areas where personalised healthcare has particularly strong potential.
A person's genetic characteristics can influence susceptibility to certain diseases and, in some circumstances, how they respond to particular medicines. Cancer care provides a prominent example because tumours can contain molecular characteristics that help clinicians classify disease and consider treatment options.
AI can assist researchers and clinicians by analysing large genomic datasets and identifying relationships between genetic variations, disease characteristics and treatment responses.
The challenge is that genomic information is highly complex. A genetic difference does not automatically mean that a particular treatment will work or fail. Clinical context still matters.
AI therefore works best as part of a broader evidence-based process in which genomic findings are interpreted alongside medical history, pathology, imaging and other relevant information.
Medical imaging can contribute to personalised care
Medical imaging is another area where AI has developed rapidly.
Algorithms can analyse images such as X-rays, CT scans, MRI examinations and mammograms to identify patterns that may be relevant to diagnosis or disease assessment. Some AI-enabled medical devices are designed to assist with detection, classification or measurement.
The FDA's current database of AI-enabled medical devices includes authorised technologies across areas such as radiology and cardiovascular care. The agency also notes that its published list is not comprehensive, but is intended to provide transparency around identified AI-enabled devices.
For personalised treatment, imaging analysis can be useful beyond simply identifying whether a disease is present.
AI may help quantify characteristics such as tumour size, tissue changes or disease progression. Repeated measurements can then provide clinicians with additional information when evaluating whether a treatment appears to be working.
This is particularly relevant when small changes over time are clinically meaningful.
Monitoring treatment response
Treatment plans are rarely static.
A patient may respond well to a therapy initially and later require a change. Another patient may experience side effects that make an alternative approach more appropriate.
AI can support this ongoing process by analysing information collected over time.
Electronic health records, laboratory results, imaging, medication histories and data from monitoring devices can provide a series of observations rather than a single snapshot of a patient's condition.
Machine-learning models can examine these observations to identify patterns associated with improvement, deterioration or other outcomes.
The FDA specifically identifies therapeutic treatment-response monitoring as an application of AI-enabled medical technology.
The potential advantage is that clinicians may receive additional evidence when deciding whether a treatment should continue, change or receive closer monitoring.
However, the quality of this process depends on the reliability and timing of the underlying data. A model cannot compensate for missing observations or inaccurate records.
Wearable devices and continuous health information
Wearable technology has introduced another source of patient information.
Smartwatches and other connected devices can collect measurements related to activity, heart rate and other physiological signals. Depending on the device and intended use, these measurements can provide additional information between clinical appointments.
AI can analyse streams of data to identify unusual patterns or changes.
For someone managing a long-term condition, this could potentially help healthcare professionals understand how health indicators vary over time rather than relying solely on occasional appointments.
There are limits, however. Consumer devices are not automatically equivalent to clinically validated medical equipment. Measurements can be affected by device limitations, user behaviour and environmental conditions.
Healthcare professionals therefore need to understand what a particular measurement actually represents before using it to influence a treatment decision.
AI can help identify patients at higher risk
Another important application is risk prediction.
Healthcare providers routinely assess whether a patient is at increased risk of complications, hospitalisation or disease progression. AI models can analyse combinations of variables and estimate the likelihood of particular outcomes.
Such predictions can potentially help clinicians decide which patients may benefit from closer monitoring or earlier intervention.
For example, a model might identify a combination of clinical indicators associated with an increased risk of deterioration. A healthcare team could then review the patient's situation more closely.
Risk prediction should not be confused with certainty.
A predicted risk is an estimate based on patterns observed in data. It does not mean that an outcome will definitely occur. Communicating this distinction is important because patients and clinicians may otherwise interpret statistical predictions as definitive conclusions.
The importance of data quality
Personalised healthcare depends heavily on data.
AI systems need relevant, representative and sufficiently accurate information to produce useful results. If the training data exclude certain populations, contain systematic errors or reflect historical inequalities, the resulting model may reproduce those problems.
This is particularly important in healthcare because patient populations are diverse.
A model that performs well in one hospital or population may perform differently elsewhere. Differences in demographics, clinical practice, equipment, disease prevalence and data collection methods can all affect performance.
The FDA identifies limited data, difficulties obtaining reliable reference information, bias and uncertainty as significant challenges in evaluating AI-enabled medical technologies.
This is why validation needs to continue beyond the initial development stage. Systems should be evaluated in settings that reflect the environments in which they will actually be used.
Privacy becomes more complicated as data use expands
Personalised medicine requires information about individuals, and health information is among the most sensitive forms of personal data.
AI can increase the value of health datasets, but it can also increase the consequences of inappropriate access or use. Combining information from multiple sources can create additional privacy concerns, particularly when data that appear anonymous can potentially be linked with other information.
A recent review of AI governance in healthcare highlights privacy, security, consent, patient autonomy and algorithmic bias as continuing concerns. It also discusses approaches such as privacy-enhancing technologies and federated learning that can reduce some risks associated with centralising sensitive information.
Healthcare organisations therefore need clear rules around data collection, access, retention and secondary use.
Patients should also be able to understand, as far as reasonably possible, how their information is being used and what role automated systems play in their care.
The challenge of explaining AI decisions
One of the more difficult questions concerns explainability.
Some AI models can produce highly accurate predictions while making it difficult for users to understand precisely why a particular output was generated.
In healthcare, this creates a practical problem.
A clinician may receive an AI-generated risk score or recommendation, but still need to understand whether the result makes clinical sense before acting on it. A system that simply provides an unexplained answer may be less useful than one that provides relevant supporting information.
Transparency is therefore becoming an important part of AI development.
The FDA, Health Canada and the UK's Medicines and Healthcare products Regulatory Agency have established guiding principles for transparency in machine-learning-enabled medical devices. These principles emphasise communicating information that could affect risks and patient outcomes to the people who interact with the technology, including healthcare professionals and patients.
Explainability does not necessarily require revealing every technical detail of an algorithm. It means providing enough meaningful information for users to understand its purpose, limitations, performance and appropriate role in decision-making.
Keeping clinicians in the decision-making process
Personalised treatment does not mean handing control of healthcare decisions to an algorithm.
Clinical decisions involve factors that may not be captured completely in structured datasets. A patient's preferences, circumstances, concerns and priorities can all influence the most appropriate course of action.
For example, two treatment options may have similar predicted outcomes but very different effects on daily life. One patient may prioritise reducing treatment frequency, while another may place greater importance on avoiding a particular side effect.
An AI model may help estimate clinical outcomes, but it cannot independently determine a person's values.
The concept of the human-AI team is therefore increasingly important. The FDA's transparency guidance specifically highlights the performance of human-AI teams rather than viewing the technology in isolation.
The goal is not simply to make a prediction more accurate. It is to ensure that the combined decision-making process is useful, safe and understandable.
Regulation must keep pace with changing technology
Healthcare AI presents a regulatory challenge because software can change more quickly than traditional medical technologies.
Some machine-learning systems may be updated as new data become available. This creates questions about when an update represents a minor improvement and when it changes the system sufficiently to require further evaluation.
The FDA has developed guidance around predetermined change control plans for AI-enabled medical devices, allowing developers to describe anticipated modifications and how those changes will be managed. The broader principle is that AI systems need appropriate oversight throughout their lifecycle rather than only at the point of initial deployment.
The FDA's current guidance landscape also includes recommendations concerning lifecycle management for AI-enabled device software and clinical decision-support software.
This continuing oversight matters because a model that performs well when first introduced may behave differently as patient populations, clinical practices or data patterns change.
What the future of personalised treatment may look like
The next phase of healthcare AI is likely to involve greater integration rather than isolated tools.
Patient information currently sits across many systems, including laboratory platforms, imaging systems, electronic health records and specialist databases. Better integration could allow clinicians to consider a broader picture of an individual's health when evaluating treatment options.
AI may also become more useful in anticipating changes rather than simply analysing existing information.
For example, future systems could combine longitudinal clinical records, molecular information and other appropriate health data to identify patterns associated with disease progression or treatment response.
Generative AI may also contribute to clinical workflows by helping summarise patient histories, explain complex medical information in accessible language or support communication between healthcare professionals and patients. At the same time, generative systems introduce their own risks, including inaccurate or fabricated information.
The FDA is actively examining regulatory considerations for generative-AI-enabled medical devices, including risk assessment, pre-market evaluation and post-market monitoring.
This indicates that the technology is developing faster than simple assumptions about how healthcare software traditionally works.
A more individual approach, with appropriate safeguards
AI has the potential to make personalised healthcare more practical by helping clinicians interpret large and complex datasets. It can contribute to diagnosis, risk assessment, treatment-response monitoring, medical imaging and the interpretation of information relevant to precision medicine.
But personalisation is not simply a technical exercise.
The quality of healthcare decisions depends on reliable evidence, representative data, clinical expertise and an understanding of individual patient circumstances. Privacy, security, transparency and fairness must also be considered throughout the development and use of AI systems.
The most useful role for AI may therefore be as an additional layer of intelligence within healthcare rather than an independent decision-maker. It can help clinicians identify patterns, organise information and consider possibilities that might otherwise be difficult to evaluate at scale.
Ultimately, personalised treatment remains centred on the individual patient. AI can process information and generate predictions, but responsible healthcare still requires human judgement, clinical accountability and meaningful communication with the people receiving care. As these technologies mature, maintaining that balance will be central to ensuring that more personalised healthcare is not only technically possible, but also safe, equitable and genuinely useful.
