Skin cancer is the commonest form of cancer in humans and is divided in 3 types: basal cell cancer, squamous cell cancer and melanoma. Exposure to sunlight, a weaker immune system, environmental toxins and old age are factors that make people more likely to develop a skin cancer. Certain medicines, when taken for long periods of time, may make it more likely for patients to develop skin cancer, for example through making the skin more susceptible to the effect of sunlight one receives.
The aim of this study is to explore links between commonly prescribed medicines and their association with skin cancer.
Data from 79,428 patients in England between 2010 and 2019 with the two commonest types of skin cancer (50,000 with basal cell cancer and 29,428 with squamous cell cancer) will be each matched to 5 patients without skin cancer (397,140 patients in total) through a general practice records database (CPRD). All data will be anonymised to maintain confidentiality. The data will be analysed using modern machine learning and artificial intelligence methods, looking for possible links between medications and the chances of developing skin cancer. If associations are found, data from groups of patients and their other health problems will be analysed to find possible reasons.
The results could assist doctors and patients in better understanding the risk of skin cancer with some medications and tailor drug prescribing to patients’ best interests.
Cutaneous basal and squamous cell carcinomas (cBCC and cSCC, also known as ‘keratinocyte cancers’) are the commonest forms of skin cancer and their incidence is rising. It is thought that hydrochlorothiazide, amongst other medications known to increase the risk of developing cBCC and cSCC, does so via its photosensitising properties [1,2]. Additionally, immunosuppressive agents particularly increase cSCC risk [3].
The aim of this hypothesis-generating study is to generate clusters of exposures to commonly prescribed medications which are associated with risk of cBCC or cSCC.
A randomised selection of 79,428 cases (50,000 with cBCC and 29,428 with cSCC) in England will be extracted from national primary care data (CPRD Aurum) between 2010-2019. Preliminary counts have yielded over 176,000 cBCC, 29,428 cSCC, and over 20.2 million controls (i.e. patients without a diagnosis of any skin cancer, as per appendix code list). For every case, 5 controls without skin cancer will be matched on age, sex, GP practice and years of registration, using incidence density sampling.
To calculate the stratified risk of keratinocyte cancer following exposure to a therapeutic class (or a combination thereof), two approaches will be used: exploratory methods and descriptive statistics. Medications associated with possible risk of keratinocyte cancer will be explored using association rule mining. Patient characteristics associated with risk of cBCC or cSCC will be identified by cluster analysis. Sensitivity analyses will be conducted for those taking medications for a minimum of 1,3 or 6 months to exclude those with other cancer diagnoses, and decide the start of the observation period before index date (i.e. cSCC or cBCC diagnosis), with the end being development of skin cancer, death, date lost to follow up or January 2020.
This information will be used to generate machine learning/deep learning algorithms to assist doctors in identifying patients at risk of keratinocyte cancers.
Associations (i.e. increased or decreased risk) of medication classes with keratinocyte cancers (i.e. cBCC or cSCC).
This study will use artificial intelligence algorithms to compare patients with a diagnosis of keratinocyte cancer (cBCC or cSCC) between 2010-2019 to age-, sex- and years of GP registration-matched controls without skin cancer, to identify previously received medications associated with risk of cBCC or cSCC.
cBCC: cutaneous basal cell carcinoma; cSCC: cutaneous squamous cell carcinoma.
Dimitrios Karponis - Chief Investigator - University of East Anglia
Dimitrios Karponis - Corresponding Applicant - University of East Anglia
Daniella Sousa Massri - Collaborator - University of East Anglia
Geoffrey Guile - Collaborator - University of East Anglia
Henry Howard-Tripp - Collaborator - University of East Anglia
James Holmes - Collaborator - University of East Anglia
Jithu Kozhimannil Jose - Collaborator - University of East Anglia
Kathryn Richardson - Collaborator - University of East Anglia
Khaylen Mistry - Collaborator - University of East Anglia
Lesley Rhodes - Collaborator - University of Manchester
Mohammad Reza Zafari - Collaborator - University of East Anglia
Nicholas Levell - Collaborator - University of East Anglia
Sonia Gran - Collaborator - University of Nottingham
Wenjia Wang - Collaborator - University of East Anglia
Zenas Yiu - Collaborator - University of Manchester
Zoe Venables - Collaborator - University of East Anglia
Jithu Kozhimannil Jose - Collaborator - University of East Anglia
Nileena Ouseph - Collaborator - University of East Anglia
Patient Level Index of Multiple Deprivation