Use Case for Personalised Medicine – Key Insights for Responsible AI Governance

Healthcare is complex as every patient responds to treatments differently.
Personalised medicine - key insights for responsible ai governance

Every patient responds to treatments differently. Personalised medicine has excellent potential for providing targeted care and treatment. Machine learning helps identify people at risk of disease and find the best treatment plans based on patient history, a probable response to current therapy, account for drug interactions and side effects and cross-referencing similar patient data and outcomes. 

Key benefits

Personalised medicines provide significant benefits for individual patients, healthcare systems, clinicians, biologists, academics and the broader economy, providing: 

  • Early detection and prevention of disease
  • Disease diagnosis to enable individualised treatment
  • Improved treatment outcomes and reduced side effects 

Real-world Use Case Examples

The UK 100,000 Genomes Project is a significant initiative that will promote personalised medicine in the UK by providing a mechanism for developing new diagnostics and treatments and explicitly linking these to clinical care.

GNS Healthcare, a leading precision medicine company ( – uses AI to mine patient clinical data to match them with the most effective treatment and care management plans for multiple myeloma, prostate cancer, Alzheimer’s, and Parkinson’s, with other conditions. They have developed a virtual patient to simulate disease progression and drug response to better select patients and rapidly generate comparative effectiveness evidence.

For clinical prediction of overall survival (OS) in patients with metastatic colorectal cancer(mCRC), GNS uses the Bayesian machine learning approach and clinical trials datasets to identify gender and primary side specific predictors of OS in mCRC and patient subpopulations with better response to a given treatment.

GNS uses more than 2,000 patient data with mCRC with about 30,000
clinical and molecular features from the MMRF CoMMPass IA9 dataset and GNS’ powerful causal AI platform, Reverse Engineering & Forward Simulation (REFSTM).

REFSTM uses an approximation of the causal relationships among the data in the variables (such as demographics, treatment, laboratories and somatic variants) and built nine independent predictive model ensembles on understanding predictive biomarkers for patient response to receiving different treatments (cetuximab, bevacizumab, or panitumumab) and to identify patient subpopulation-specific prognostic factors of OS, high risk and progression-free survival (PFS). This model is used to determine which patients will respond to stem cell transplants and to estimate the effectiveness of interventions based on the specific patient characteristics.

GNS also identifies gender-specific predictors, precisely creatinine level, intra-abdominal metastasis status, and the interaction of albumin and neutrophil levels. Urine protein levels were shown to predict better efficacy from cetuximab treatment in patients with mCRC and left-sided tumours.

Use case development template

Here is the guidance to help you develop business and a high-value use case for Personalised Medicine using Artificial Intelligence and Machine Learning.

This Personalised Medicine use case framework guidance describes Esdha’s current research on the topic and should be viewed only as recommendations, unless specific regulatory or statutory requirements are cited.

Challenges & Opportunities

Impact: Cancer Discovery in 2017 found that 93% of 1,000 patients did not benefit from precision medicine—and four studies presented in the American Society of Clinical Oncology(2018) found similar results. In particular, precision medicine failed to shrink tumours in 95% and 92% of cases.

Target group: only a few are suitable for—or can afford the cost. In a study, it was estimated that only 8 to 15% of patients end up being eligible for precision medicines, and of those patients, about half benefit from the targeted treatment.

Evidence base: overall knowledge creation with the clear evidence base for incorporating genomic data by a diagnostic group or a subgroup characterised by its response to a treatment is a challenge and requires new infrastructures, processes and costs. Furthermore, there are wide-ranging methodological issues, including conducting clinical trials, and measuring outcomes and effectiveness.

Data: there is a need for extensive data collection to gather insights from different groups of patients that understand the benefits of various interventions, which can create concerns about privacy and data quality.

Integrated health services: Building an integrated and accessible system across a healthcare system is complex. Personalised medicine needs an integrated approach, including access to healthcare, genomics and genetics, tissue biomarkers, imaging and radiology, environmental exposures, wireless monitoring, behaviours and personality and epigenetic modifications.

Strategy: we need to integrate AI strategy with the overall business strategy to systematically integrate into mainstream healthcare whilst ensuring the ethical, equality and economic implications are fully recognised and addressed.

Acceptance: get broad acceptance from different healthcare stakeholders, such as physicians, healthcare executives, insurance companies, and, ultimately, patients. Ensure that patients and the public are confident in using these technologies and that we can mitigate any potential concerns, particularly in data security and confidentiality.

Training: Lack of appropriate education and training with patient involvement and empowerment.

Cost: may incur high charges to the patient than traditional forms of effective treatment.

Regulatory approval: obtaining approval for routine use from various regulatory agencies.

Complexity: the system accurately maps demographics, treatment, laboratory and semantic variables for different patient groups to get the proper insight. 

Potential risks & mitigation

Data: The generation and curation of large clinico-genomic datasets can have missing, inaccurate, or incomplete source data. Imputation-based methods that replace missing data, however, imputation assumes data are missing at random, which they often are not. Expanding structured real-world data collection can isolate more social and biological factors that drive health behaviours, reducing the assumptions made in the models and improving their accuracy and generalizability. American Society of Clinical Oncology’s mCODE project collects structured patient data.

Privacy & quality: Ensure data privacy, quality, and access is carried out. A standardised approach to data collection can help to address this risk.

Transparency & replicability of results: Ensure the uniform workflow encourages algorithmic transparency and replicability of results.

Bias, overfitting and validity: build a rigorous criterion to evaluate for biases (such as statistical misrepresentation to the general population), overfitting, and validity.

Interdisciplinary team: We need an interdisciplinary team consisting of computer scientists, biologists, physicists, patients and clinicians to align goals, requirements and clinical trial outcomes.

Socio-environmental factors: The environmental factors and processes may impact model performance and clinical efficacy. 



More Posts

Modernising Applications

Overcoming Obstacles: Why Application Modernisation Matters

In today’s fast-paced digital landscape, legacy systems can often hinder #business growth and impede essential processes. Recognising this tipping point, leaders are turning to application modernisation to overcome these obstacles and propel their organisations forward.

Send Us A Message