Elizabeth Williamson

Purpose: The COVID‐19 pandemic disrupted healthcare services, potentially affecting causal effect estimation using real‐world data. As a case study, we replicated a cohort study using Clinical Practice Research Datalink Aurum to evaluate whether pandemic‐related disruptions influenced the measured association between fluoroquinolone use and tendon rupture. Methods: We included adults initiating fluoroquinolones/cephalosporins during 1 January 2006–31 March 2023. Four analyses compared treatment effect estimates using different periods for exposure and outcome ascertainment: pre‐pandemic only, pre‐pandemic exposure with follow‐up extending through the pandemic, all available data and post‐pandemic only. We estimated hazard ratios (HRs) for tendon rupture using Cox regression. Results: Among 891 968 fluoroquinolone users and 895 229 cephalosporin users, fluoroquinolone users had more comorbidities than cephalosporin users in the pre‐pandemic period but fewer in the post‐pandemic period. Pre‐pandemic analysis showed an increased risk of tendon rupture associated with fluoroquinolones (Propensity score weighted [PSW]‐HR: 1.94; 95% CI: 1.46–2.59), consistent with previous evidence. Estimates using all available data were similar for the overall study period (PSW‐HR: 1.91; 95% CI: 1.46–2.51) and the pre‐pandemic period (PSW‐HR: 1.96; 95% CI: 1.47–2.61) but were less precise during/post‐pandemic period (PSW‐HR: 1.57; 95% CI: 0.72–3.41) with no evidence of a period‐treatment interaction (PSW‐HR: 0.80; 95% CI: 0.35–1.84). Post‐pandemic analysis showed very few events among cephalosporin users but not fluoroquinolone users with a larger but imprecise estimate (PSW‐HR: 3.32; 95% CI: 0.62–17.72). Conclusion: In this case study, there was no clear evidence that pandemic‐related healthcare disruptions materially altered causal estimates, albeit with limited power to detect differences between before and during/post‐pandemic effects. Studies using real‐world data across the pandemic should routinely examine temporal changes in baseline characteristics and outcome incidence.

Background: Flexible machine-learning (ML) models to generate imputations within the Multiple Imputation (MI) framework has recently gained traction, particularly in non-randomised observational settings. For randomised controlled trials (RCTs), it is unclear whether ML approaches to MI, in combination with Rubin’s Rules, provide valid inference in terms bias, confidence interval coverage, mean squared error (MSE), Type I error and power of the treatment effect estimator. Methods: We conducted two simulation studies in RCT settings that have an incomplete continuous outcome but fully observed covariates and treatment assignment. We compared Complete Cases, standard MI (MI-norm), MI with predictive mean matching (MI-PMM) and ML-based approaches to MI, including classification and regression trees (MI-CART), Random Forests (MI-RF) and SuperLearner when outcomes are missing completely at random or missing at random conditional on treatment/covariate. The first simulation explored a cross-sectional outcome with non-linear covariate-outcome relationships in the presence/absence of covariate-treatment interactions. The second simulation explored skewed repeated measures, motivated by a trial with digital outcomes. Results: For the cross-sectional simulation without interaction, we found that Complete Cases yielded valid inference; MI-norm performed similarly, except when there is a non-linear covariate-outcome relationship and missingness depends on the covariate. ML approaches led to smaller MSE in specific non-linear settings, but provided unreliable inference for others. MI-PMM, which is the default setting in the package , led to unreliable inference under some settings. In the presence of complex treatment-covariate interactions, performing MI separately by arm, either with MI-norm, MI-RF or MI-CART, provided inference that had comparable or better properties compared to Complete Cases when the analysis model omits the interaction term. In the repeated measures setting, ML approaches to MI and MI-PMM led to bias and under- or over-coverage, particularly when missingness depended on treatment. Conclusion: Based on the simulation findings, Complete Cases and MI-norm are more appropriate than ML approaches to MI for making inference, especially for late phase RCTs where Type I error control is crucial. While ML approaches may provide gains in MSE for complex covariate-outcome relationships, results should be interpreted with the caveat that Rubin’s Rules are not guaranteed to be valid when used with ML imputation methods, and can lead to bias in the estimated effect and/or its standard error.

Background Living with children has been associated with greater risks of SARS-CoV-2 infection, COVID-19 hospitalisation, and COVID-19 death. We examined how these associations varied during 2021–22 and according to the COVID-19 vaccination status of adults. Methods We carried out a population-based cohort study, with the approval of NHS England. Primary care data and pseudonymously-linked hospital and death records from England, between 20 th December 2020 and 21 st February 2022, were used for adults (≥18 years) registered at a general practice on 20 th December 2020. Adjusted hazard ratios (HRs) for SARS-CoV-2 infection, COVID-19 hospitalisation, or COVID-19 death, by presence of children in the household were calculated. Results The cohort included 9,417,278 adults aged ≤65 years and 2,866,602 adults aged >65 years. Adults aged ≤65 years living with children of any age ( versus no children) had greater risks of SARS-CoV-2 infection and COVID-19 hospitalisation (but not COVID-19 death), both when schools were open and closed ( e.g. HR = 1.50, 95% CI:1.49–1.51, for SARS-CoV-2 infection in the ‘Omicron dominant’ period, when schools were open, in adults living with children aged 0–11 years only). These associations also existed for adults aged >65 years, and there was some evidence that older adults living with children also had greater risks of COVID-19 death. Vaccinated adults living with children had greater risks of SARS-CoV-2 infection, but lower risks of COVID-19 hospitalisation and death, than unvaccinated adults not living with children. Conclusions In an era of widespread adult vaccination, adults living with children remained at increased risk of SARS-CoV-2 infection and COVID-19 hospitalisation.

Background: Cancer and its treatment can adversely affect lung health, but long-term respiratory risks in adult cancer survivors remain unclear. We investigated risks of developing new chronic respiratory conditions, or exacerbating existing ones, in survivors of 20 common cancers, compared with individuals with no cancer history.Methods: We used English electronic primary care records, linked to cancer registry, hospital admissions, and death records data. Adults with incident cancer were matched 1:10 to cancer-free individuals on age, sex and general practice. Relative risks were estimated using Cox models, adjusted for confounders. Absolute risks were estimated using adjusted incident rate differences and standardised cumulative incidence curves for new-onset disease, and mean cumulative count for exacerbations. We investigated effect modification by patient-level factors using interaction terms.Results: 789,254 adults with incident cancer were included. Compared with cancer-free individuals, elevated risks of developing chronic obstructive pulmonary disease were observed in 11 cancers (adjusted Hazard ratio [aHR] 5.19 increased risk for lung cancer; other HRs ranging from 1.06-1.72); for 8 of these cancers, raised risks persisted beyond 5 years from diagnosis. Interstitial lung disease risk was raised in 18 cancers (aHRs>5 in CNS, oesophageal and lung cancers, and aHRs>2 in 7 further cancers); raised risks persisted beyond 5 years after diagnosis in 11 cancers. New-onset bronchiectasis risk was increased in oesophageal, lung and haematological cancers while new-onset asthma risk was raised in non-Hodgkin lymphoma only. Rates of exacerbations of pre-existing respiratory diseases were raised in survivors of oesophageal, liver, lung, pancreatic, central nervous system, and haematological cancers, compared with cancer-free individuals.Conclusion: Survivors of most types of cancer had substantial and sometimes prolonged raised risk of a range of respiratory conditions, highlighting the need for respiratory health awareness in survivorship care. Targeted management, smoking cessation, and vaccination may be warranted for those at highest risk.

Background: There is growing interest in widening the use of sodium-glucose co-transporter 2 inhibitors (SGLT2i) to all people with type 2 diabetes mellitus (T2DM). However, pivotal randomized controlled trials (RCTs) evaluated these drugs only in highly selected populations, often lacking generalizability to real-world populations. Understanding the effects of SGLT2i in populations where RCT evidence may be lacking is essential to help inform guideline development. To address this, we estimated the effect of empagliflozin in real-world users, many of whom would not have been eligible for the pivotal EMPA-REG RCT. Methods: We designed a trial emulation in UK primary care data, based on the EMPA-REG RCT, to assess the effect of empagliflozin in a more clinically relevant population. Adults with T2DM initiating empagliflozin (intervention) or dipeptidyl peptidase-4 inhibitors (active control) between January 1, 2014 and December 31, 2022 were included. Eligibility was extended to both RCT-eligible and RCT-ineligible individuals. The effect of empagliflozin on all-cause mortality was estimated using an adjusted Cox proportional hazards model, with stratified analyses by RCT eligibility. Findings: The majority of people prescribed empagliflozin would not have met the EMPA-REG RCT eligibility criteria (11,011/13,239, 83.2% RCT-ineligible). During follow-up, all-cause mortality occurred in 551 out of 13,239 (4.2%) in the empagliflozin group and 6,589 out of 49,264 (13.4%) in the active control group (adjusted HR 0.76, 95% CI 0.69 to 0.83). There was no evidence of differential treatment effect by RCT eligibility status (p-interaction=0.27). Interpretation: Patients prescribed empagliflozin in real-world settings differ substantially from those enrolled in the EMPA-REG RCT. Using electronic health records, we demonstrate that the mortality benefit observed in EMPA-REG extends to a broader, more diverse real-world population, including those excluded from the original RCT. These findings provide a novel source of real-world evidence supporting the wider use of empagliflozin in routine clinical practice.

On March 11th 2020, the World Health Organization characterised COVID-19 as a pandemic. Responses to containing the spread of the virus have relied heavily on policies involving restricting contact between people. Evolving policies regarding shielding and individual choices about restricting social contact will rely heavily on perceived risk of poor outcomes from COVID-19. In order to make informed decisions, both individual and collective, good predictive models are required. For outcomes related to an infectious disease, the performance of any risk prediction model will depend heavily on the underlying prevalence of infection in the population of interest. Incorporating measures of how this changes over time may result in important improvements in prediction model performance. This protocol reports details of a planned study to explore the extent to which incorporating time-varying measures of infection burden over time improves the quality of risk prediction models for COVID-19 death in a large population of adult patients in England. To achieve this aim, we will compare the performance of different modelling approaches to risk prediction, including static cohort approaches typically used in chronic disease settings and landmarking approaches incorporating time-varying measures of infection prevalence and policy change, using COVID-19 related deaths data linked to longitudinal primary care electronic health records data within the OpenSAFELY secure analytics platform.

SUMMARY: Micro-randomized trials are commonly conducted for optimizing mobile health interventions such as push notifications for behavior change. In analyzing such trials, causal excursion effects are often of primary interest, and their estimation typically involves inverse probability weighting (IPW). However, in a micro-randomized trial, additional treatments can often occur during the time window over which an outcome is defined, and this can greatly inflate the variance of the causal effect estimator because IPW would involve a product of numerous weights. To reduce variance and improve estimation efficiency, we propose two new estimators using a modified version of IPW, which we call “per-decision IPW.” The second estimator further improves efficiency using the projection idea from the semiparametric efficiency theory. These estimators are applicable when the outcome is binary and can be expressed as the maximum of a series of sub-outcomes defined over sub-intervals of time. We establish the estimators’ consistency and asymptotic normality. Through simulation studies and real data applications, we demonstrate substantial efficiency improvement of the proposed estimator over existing estimators. The new estimators can be used to improve the precision of primary and secondary analyses for micro-randomized trials with binary outcomes.

Background: The UK delivered its first “booster” COVID-19 vaccine doses in September 2021, initially to individuals at high risk of severe disease, then to all adults. The BNT162b2 Pfizer-BioNTech vaccine was used initially, then also Moderna mRNA-1273. Methods: With the approval of the National Health Service England, we used routine clinical data to estimate the effectiveness of boosting with BNT162b2 or mRNA-1273 compared with no boosting in eligible adults who had received two primary course vaccine doses. We matched each booster recipient with an unboosted control on factors relating to booster priority status and prior COVID-19 immunization. We adjusted for additional factors in Cox models, estimating hazard ratios up to 182 days (6 months) following booster dose. We estimated hazard ratios overall and within the following periods: 1–14, 15–42, 43–69, 70–97, 98–126, 127–152, and 155–182 days. Outcomes included a positive SARS-CoV-2 test, COVID-19 hospitalization, COVID-19 death, non-COVID-19 death, and fracture. Results: We matched 8,198,643 booster recipients with unboosted controls. Adjusted hazard ratios over 6-month follow-up were: positive SARS-CoV-2 test 0.75 (0.74, 0.75); COVID-19 hospitalization 0.30 (0.29, 0.31); COVID-19 death 0.11 (0.10, 0.14); non-COVID-19 death 0.22 (0.21, 0.23); and fracture 0.77 (0.75, 0.78). Estimated effectiveness of booster vaccines against severe COVID-19-related outcomes peaked during the first 3 months following the booster dose. By 6 months, the cumulative incidence of positive SARS-CoV-2 test was higher in boosted than unboosted individuals. Conclusions: We estimate that COVID-19 booster vaccination, compared with no booster vaccination, provided substantial protection against COVID-19 hospitalization and COVID-19 death but only limited protection against positive SARS-CoV-2 test. Lower rates of fracture in boosted than unboosted individuals may suggest unmeasured confounding. Observational studies should report estimated vaccine effectiveness against nontarget and negative control outcomes.

Cardiovascular disease (CVD) is a leading cause of death globally. Angiotensin-converting enzyme inhibitors (ACEi) and angiotensin receptor blockers (ARB), compared in the ONTARGET trial, each prevent CVD. However, trial results may not be generalisable and their effectiveness in underrepresented groups is unclear. Using trial emulation methods within routine-care data to validate findings, we explored generalisability of ONTARGET results. For people prescribed an ACEi/ARB in the UK Clinical Practice Research Datalink GOLD from 1/1/2001-31/7/2019, we applied trial criteria and propensity-score methods to create an ONTARGET trial-eligible cohort. Comparing ARB to ACEi, we estimated hazard ratios for the primary composite trial outcome (cardiovascular death, myocardial infarction, stroke, or hospitalisation for heart failure), and secondary outcomes. As the pre-specified criteria were met confirming trial emulation, we then explored treatment heterogeneity among three trial-underrepresented subgroups: females, those aged ≥75 years and those with chronic kidney disease (CKD). In the trial-eligible population (n=137,155), results for the primary outcome demonstrated similar effects of ARB and ACEi, (HR 0.97 [95% CI: 0.93, 1.01]), meeting the pre-specified validation criteria. When extending this outcome to trial-underrepresented groups, similar treatment effects were observed by sex, age and CKD. This suggests that ONTARGET trial findings are generalisable to trial-underrepresented subgroups.

We aimed to explore whether the ONTARGET trial results, which led to an end of recommendations of dual angiotensin-converting enzyme inhibitor (ACEi) and angiotensin receptor blocker (ARB) use, extended to patients with chronic kidney disease (CKD) who were underrepresented in the trial. We selected people prescribed an ACEi and/or an ARB in the UK Clinical Practice Research Datalink Aurum during 1/1/2001-31/7/2019. We specified an operational definition of dual users and applied ONTARGET eligibility criteria. We used propensity-score—weighted Cox-proportional hazards models to compare dual therapy to ACEi for the primary composite trial outcome (cardiovascular death, myocardial infarction, stroke, or hospitalisation for heart failure), as well as a primary composite renal outcome (≥50% reduction in GFR or end-stage kidney disease), and other secondary outcomes, including hyperkalaemia. Conditional on successfully benchmarking results against the ONTARGET trial, we explored treatment effect heterogeneity by CKD at baseline. In the propensity-score—weighted trial-eligible analysis cohort (n=412 406), for dual therapy vs ACEi we observed hazard ratio (HR) 0.98 (95% CI: 0.93, 1.03), for the primary composite outcome, consistent with the trial results (ONTARGET HR 0.99, 95% CI: 0.92, 1.07). Dual therapy use was associated with an increased risk of the primary renal composite outcome, HR 1.25 (95% CI: 1.15, 1.36) vs ONTARGET HR 1.24 (1.01, 1.51) and hyperkalaemia, HR 1.15 (95% CI: 1.09, 1.22) in the trial eligible cohort, consistent with ONTARGET. The presence of CKD at baseline had minimal impact on results. Translational statement: We extended ONTARGET trial findings of the comparative effectiveness of dual ARB and ACEi therapy use compared to ACEi alone for a composite cardiovascular outcome to UK patients at high-risk of cardiovascular disease, including those with CKD. As in ONTARGET, we found an increased risk of a composite renal outcome (≥50% reduction in GFR or end-stage kidney disease) and an increased risk of hyperkalaemia among dual users compared to ACEi alone. Consistent results were observed among patients with CKD at baseline. This is evidence against the hypothesis that dual blockade provides cardiorenal benefits among high-risk cardiovascular patients with CKD.