Evaluating and addressing the impact of COVID-19 restrictions on electronic health records in estimating causal effects
Lead Research Organisation:
London School of Hygiene & Tropical Medicine
Department Name: Epidemiology and Population Health
Abstract
This proposal aims to improve our ability to use routine health data collected during the COVID-19 pandemic to answer important clinical questions about the effectiveness and safety of medications.
While electronic health records (EHRs) are increasingly used for epidemiological research to estimate the causal effects between drug exposures and clinical outcomes, disruptions arising from the COVID-19 pandemic have had an as yet unquantified effect on our ability to successfully conduct this research. Notably, the COVID-19 pandemic led to lockdowns in many countries including the UK, resulting in behavioural changes in seeking healthcare services and thus prescribing patterns, recordings of clinical observations and measurements, and disease diagnoses in the EHRs. Specifically, the change in diagnostic recording could imply delays in diagnoses or missing diagnoses that would normally have been recorded. It could lead to measurement errors in the identification of study populations and ascertainment of outcomes, compromising the validity of study findings.
This proposed work will therefore identify and quantify potential measurement errors using two case-studies representing diverse clinical contexts as illustrations: 1) investigating the risks and benefits of long-term routine therapy with oral anticoagulants; and 2) quantifying the known side effect of tendon rupture associated with short course fluoroquinolone antibiotics. We will use data from the UK Clinical Practice Research Datalink Aurum linked with Hospital Episode Statistics and Office for National Statistics. This world-renowned primary care database has comprehensive medical records for a sample of ~19.8% of the UK population that is broadly representative in terms of age, sex and ethnicity. By categorising three periods which are pre-, during and post-pandemic periods, we will identify possible measurement errors by describing absolute rates of disease diagnoses for the identification of both study populations and outcomes. We will compare treatment effects using pre-pandemic data only with that combining pre-pandemic, during and post-pandemic data in each case study. The findings of the case studies using primary care data will first be validated against randomised controlled trials or a systematic review with meta-analysis. We will quantify the measurement errors using a period-treatment interaction to evaluate treatment effects in stratified periods. We will develop and evaluate approaches that attempt to correct for measurement errors due to pandemic restrictions, by exploiting and extending robust methods. These include using a simulation-extrapolation method and applying quantitative bias analysis. We will then recommend an optimal methodological approach to handle pandemic-related measurement errors using EHRs based on the findings.
This proposed work is highly feasible as the data is routinely collected and readily available for analysis. It will inform how measurement errors will impact the estimation of causal effect in different settings. Our findings will lead to recommendations for researchers using EHRs to design future studies that include data/follow-up spanning the pandemic period. The methods developed will allow future causal epidemiological questions to be answered as robustly as possible, and will benefit policymakers, clinicians, patients, carers to inform healthcare decision-making.
While electronic health records (EHRs) are increasingly used for epidemiological research to estimate the causal effects between drug exposures and clinical outcomes, disruptions arising from the COVID-19 pandemic have had an as yet unquantified effect on our ability to successfully conduct this research. Notably, the COVID-19 pandemic led to lockdowns in many countries including the UK, resulting in behavioural changes in seeking healthcare services and thus prescribing patterns, recordings of clinical observations and measurements, and disease diagnoses in the EHRs. Specifically, the change in diagnostic recording could imply delays in diagnoses or missing diagnoses that would normally have been recorded. It could lead to measurement errors in the identification of study populations and ascertainment of outcomes, compromising the validity of study findings.
This proposed work will therefore identify and quantify potential measurement errors using two case-studies representing diverse clinical contexts as illustrations: 1) investigating the risks and benefits of long-term routine therapy with oral anticoagulants; and 2) quantifying the known side effect of tendon rupture associated with short course fluoroquinolone antibiotics. We will use data from the UK Clinical Practice Research Datalink Aurum linked with Hospital Episode Statistics and Office for National Statistics. This world-renowned primary care database has comprehensive medical records for a sample of ~19.8% of the UK population that is broadly representative in terms of age, sex and ethnicity. By categorising three periods which are pre-, during and post-pandemic periods, we will identify possible measurement errors by describing absolute rates of disease diagnoses for the identification of both study populations and outcomes. We will compare treatment effects using pre-pandemic data only with that combining pre-pandemic, during and post-pandemic data in each case study. The findings of the case studies using primary care data will first be validated against randomised controlled trials or a systematic review with meta-analysis. We will quantify the measurement errors using a period-treatment interaction to evaluate treatment effects in stratified periods. We will develop and evaluate approaches that attempt to correct for measurement errors due to pandemic restrictions, by exploiting and extending robust methods. These include using a simulation-extrapolation method and applying quantitative bias analysis. We will then recommend an optimal methodological approach to handle pandemic-related measurement errors using EHRs based on the findings.
This proposed work is highly feasible as the data is routinely collected and readily available for analysis. It will inform how measurement errors will impact the estimation of causal effect in different settings. Our findings will lead to recommendations for researchers using EHRs to design future studies that include data/follow-up spanning the pandemic period. The methods developed will allow future causal epidemiological questions to be answered as robustly as possible, and will benefit policymakers, clinicians, patients, carers to inform healthcare decision-making.
