As our healthcare standard continues to improve, the average life expectancy of a person with disease or multiple diseases also increases. Current clinical guidelines for drug prescriptions mostly operate on the assumption that patients have single, isolated diseases. This approach fails to account for the cumulative impact of multiple treatment aspects, such as the potentially harmful interaction between multiple drugs or the potentially contrasting effect of a single drug on multiple co-existing conditions, which may lead to potentially severe or fatal adverse outcomes.
We propose the creation of a graphical knowledge base, mapping the complete relationships between drugs and diseases, within drugs and within diseases, including the variety and frequency of drugs prescribed for each disease. This allows the identification of all predictable harmful effects elicited by drugs in treatment of multiple diseases. This knowledge base will also identify drugs and diseases involved in treatment dilemmas where the overall net benefit or harm of a drug in the presence of multiple diseases is unclear.
This work will help clinicians inform treatment decisions while accounting for the presence of multiple diseases and their drug interactions, and will help researchers identify and prioritise further investigation of the use of drugs with unclear treatment outcomes in the presence of multiple diseases, especially where a drug that is prescribed frequently may benefit a patient’s first condition but harm the second. Such findings will help to address complex treatment dilemmas and account for the cumulative impact of multiple drug treatments on disease or multiple diseases.
We aim to create an interpretable, accessible and user-friendly directed knowledge graph framework using standardised clinical terminology mapping the relationship between drugs and diseases, within drugs and within diseases. Using linked primary and secondary care data (CPRD/HES) will allow establishment of the relationship between drug prescriptions and conditions for which the drugs are prescribed for. Many existing databases attempting similar aims are incomplete and non-standardised, not available to end-users and exist in mostly unstructured free text. Our framework will consist of a backend SQL database containing entities labelled by standardised medical terminology (HES ICD-10, CPRD and OPSC read codes, and BNF labels) with front end knowledge graph accessibility through a basic user query tool and an integrated clinician/researcher clinical decision support system tool.
Relative risk ratios will be used to determine the risk of occurrence of comorbidities, the risk of diagnoses in different groups or given a prescribing pattern. Propensity score matching will be employed to minimise confounding effects. Cox proportional hazard regression will be used to ascertain associations between treatment and outcomes in single or multi-drug treatment. All-cause and cause-specific mortality in prevalence disease and at the 1 year post-diagnosis for incident disease will be established.
Such a framework will aim to improve and inform clinical treatment decisions by accounting for multimorbidity through identifying all known harmful interactions that may not be clearly presented using current singular drug prescription clinical guidelines. Our framework will also aim to identify drugs that may be used to treat a patient with commonly associated comorbidities that present as a contrasting indication-contraindication pair to the drug. In such a case, the net benefit/harm in drug administration within such multimorbidity is likely to be unclear and such findings can result in the formation of guiding hypotheses for further investigation into treatment dilemmas.
• Indication and contraindication relationship between drugs and diseases
• Interaction relationship between drugs and drugs
The above outcomes will be integrated using the following external sources of information: British National Formulary and British National Formulary for Children.
• Comorbidity relationship between diseases and diseases
• Prescribing patterns by drug and condition
• Frequency of prescribing patterns per conditions
• All-cause mortality, cause-specific mortality, mortality rate by disease and prescribing patterns
• Prevalence and incidence of diagnoses
• Identification of indication-contraindication pairs to drugs
Alvina Lai - Chief Investigator - University College London ( UCL )
Yen Yi Tan - Corresponding Applicant - University College London ( UCL )
Spiros Denaxas - Collaborator - University College London ( UCL )
Stefanie Mueller - Collaborator - University College London ( UCL )
Vaclav Papez - Collaborator - University College London ( UCL )
HES Accident and Emergency;HES Admitted Patient Care;HES Outpatient;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation