Use this URL to cite or link to this record in EThOS:
Title: Improving the measurement and detection of serious adverse drug reactions in databases of stored electronic health records
Author: Wing, Kevin
ISNI:       0000 0004 5922 3276
Awarding Body: London School of Hygiene & Tropical Medicine
Current Institution: London School of Hygiene and Tropical Medicine (University of London)
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Background: Adverse drug reactions are responsible for a significant proportion of hospitalisations. This PhD aimed to develop and optimise methods for detecting serious adverse drug reactions in databases of electronic health records for use in pharmacoepidemiology and genetic epidemiology, with a focus on cholestatic liver injury. Methods: A systematic review was performed before developing a multiple database source (“multisource”) algorithm for identifying cholestatic liver injury. Multisource algorithm case status was used to guide the development of another algorithm using data from a standard UK Clinical Practice Research Datalink (CPRD) record only (the CPRD algorithm). Testing of the CPRD algorithm was performed within a cohort analysis of an established cause of the injury (flucloxacillin), before carrying out a casecontrol study investigating a number of putative associations (drug exposures carbamazepine, celecoxib, duloxetine, ramipril and risperidone). Results: The majority of reviewed studies lacked a reproducible case definition, and case assignment generally required information external to database records. Secondary care (HES) data provided little additional information than that found in primary care (CPRD), meaning that the CPRD algorithm had a very good ability to discriminate between multisource algorithm cases statuses (ROC area under the curve 0.95). The flucloxacillin 45-day risk estimate obtained from the cohort study using the highest specificity CPRD algorithm (6.15 per 100 000 users, 95% CI 4.61 – 8.04) was very similar to previous studies. Celecoxib and risperidone were associated with cholestatic liver injury (celecoxib multivariablemultivariable RR recent vs. current users low specificity CPRD algorithm 1.89, 95% CI 1.11 – 3.22, risperidone multivariablemultivariable RR high specificity CPRD algorithm 2.59, 95% CI 1.41 – 4.75). Conclusions: The CPRD algorithm detected similar flucloxacillin effects as (1) the multisource algorithm and (2) previous studies. Associations with risperidone and celecoxib were also detected. Algorithm characteristics that could facilitate (1) pharmacovigilance and (2) recruitment to genetic association studies include the ability to (a) detect cases without using information external to the EHR and (b) apply varying levels of specificity and sensitivity.
Supervisor: Douglas, Ian Sponsor: PROTECT
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral