Effectiveness of mRNA COVID-19 Vaccines as First Booster Doses in England: An Observational Study in OpenSAFELY-TPP.

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.

OpenSAFELY: A platform for analysing electronic health records designed for reproducible research.

Electronic health records (EHRs) and other administrative health data are increasingly used in research to generate evidence on the effectiveness, safety, and utilisation of medical products and services, and to inform public health guidance and policy. Reproducibility is a fundamental step for research credibility and promotes trust in evidence generated from EHRs. At present, ensuring research using EHRs is reproducible can be challenging for researchers. Research software platforms can provide technical solutions to enhance the reproducibility of research conducted using EHRs. In response to the COVID-19 pandemic, we developed the secure, transparent, analytic open-source software platform OpenSAFELY designed with reproducible research in mind. OpenSAFELY mitigates common barriers to reproducible research by: standardising key workflows around data preparation; removing barriers to code-sharing in secure analysis environments; enforcing public sharing of programming code and codelists; ensuring the same computational environment is used everywhere; integrating new and existing tools that encourage and enable the use of reproducible working practices; and providing an audit trail for all code that is run against the real data to increase transparency. This paper describes OpenSAFELY's reproducibility-by-design approach in detail.

On variance estimation of the inverse probability-of-treatment weighting estimator: A tutorial for different types of propensity score weights.

Propensity score methods, such as inverse probability-of-treatment weighting (IPTW), have been increasingly used for covariate balancing in both observational studies and randomized trials, allowing the control of both systematic and chance imbalances. Approaches using IPTW are based on two steps: (i) estimation of the individual propensity scores (PS), and (ii) estimation of the treatment effect by applying PS weights. Thus, a variance estimator that accounts for both steps is crucial for correct inference. Using a variance estimator which ignores the first step leads to overestimated variance when the estimand is the average treatment effect (ATE), and to under or overestimated estimates when targeting the average treatment effect on the treated (ATT). In this article, we emphasize the importance of using an IPTW variance estimator that correctly considers the uncertainty in PS estimation. We present a comprehensive tutorial to obtain unbiased variance estimates, by proposing and applying a unifying formula for different types of PS weights (ATE, ATT, matching and overlap weights). This can be derived either via the linearization approach or M-estimation. Extensive R code is provided along with the corresponding large-sample theory. We perform simulation studies to illustrate the behavior of the estimators under different treatment and outcome prevalences and demonstrate appropriate behavior of the analytical variance estimator. We also use a reproducible analysis of observational lung cancer data as an illustrative example, estimating the effect of receiving a PET-CT scan on the receipt of surgery.

Sustained Hypothetical Interventions on Midlife Alcohol Consumption in Relation to All-Cause and Cancer Mortality: The Australian Longitudinal Study on Women's Health.

No randomized controlled trial has evaluated the effect of long-term alcohol interventions on mortality. Results reported in existing observational studies may be subject to selection bias and time-varying confounding. Using data from the Australian Longitudinal Study on Women's Health 1946-1951 birth cohort, collected regularly from 1996-2016, we estimated all-cause and cancer mortality had women been assigned various alcohol interventions (in categories ranging from 0 to >30 g/day ethanol, or reduced to ≤20 g/day if higher) at baseline, and had they maintained these levels of consumption. The cumulative risks for all-cause and cancer mortality were 5.6% (10,118 women followed for 20 years) and 2.9% (18 years), respectively. For all-cause and cancer mortality, baseline ethanol up to 30 g/day showed lower risk and >30 g/day showed higher risk relative to abstention. Had women sustainedly followed the interventions, a similar relationship was observed for all-cause mortality. However, the negative association observed for intakes ≤30 g/day and positive association for intakes >30 g/day was not evident for cancer mortality. Our findings suggest that all-cause mortality could have been lower than observed if this cohort of women had consumed some alcohol (no more than 30 g/day) rather than no consumption, but cancer mortality might not.

Prevalent new user designs: A literature review of current implementation practice.

PURPOSE: Prevalent new user (PNU) designs extend the active comparator new user design by allowing for the inclusion of initiators of the study drug who were previously on a comparator treatment. We performed a literature review summarising current practice. METHODS: PubMed was searched for studies applying the PNU design since its proposal in 2017. The review focused on three components. First, we extracted information on the overall study design, including the database used. We summarised information on implementation of the PNU design, including key decisions relating to exposure set definition and estimation of time-conditional propensity scores. Finally, we reviewed the analysis strategy of the matched cohort. RESULTS: Nineteen studies met the criteria for inclusion. Most studies (73%) implemented the PNU design in electronic health record or registry databases, with the remaining using insurance claims databases. Of 15 studies including a class of prevalent users, 40% deviated from the original exposure set definition proposals in favour of a more complex definition. Four studies did not include prevalent new users but used other aspects of the PNU framework. Several studies lacked details on exposure set definition (n = 2), time-conditional propensity score model (n = 2) or integration of complex analytical techniques, such as the high-dimensional propensity score algorithm (n = 3). CONCLUSION: PNU designs have been applied in a range of therapeutic and disease areas. However, to encourage more widespread use of this design and help shape best practice, there is a need for improved accessibility, specifically through the provision of analytical code alongside guidance to support implementation and transparent reporting.

Changes in COVID-19-related mortality across key demographic and clinical subgroups in England from 2020 to 2022: a retrospective cohort study using the OpenSAFELY platform.

BACKGROUND: COVID-19 has been shown to differently affect various demographic and clinical population subgroups. We aimed to describe trends in absolute and relative COVID-19-related mortality risks across clinical and demographic population subgroups during successive SARS-CoV-2 pandemic waves. METHODS: We did a retrospective cohort study in England using the OpenSAFELY platform with the approval of National Health Service England, covering the first five SARS-CoV-2 pandemic waves (wave one [wild-type] from March 23 to May 30, 2020; wave two [alpha (B.1.1.7)] from Sept 7, 2020, to April 24, 2021; wave three [delta (B.1.617.2)] from May 28 to Dec 14, 2021; wave four [omicron (B.1.1.529)] from Dec 15, 2021, to April 29, 2022; and wave five [omicron] from June 24 to Aug 3, 2022). In each wave, we included people aged 18-110 years who were registered with a general practice on the first day of the wave and who had at least 3 months of continuous general practice registration up to this date. We estimated crude and sex-standardised and age-standardised wave-specific COVID-19-related death rates and relative risks of COVID-19-related death in population subgroups. FINDINGS: 18 895 870 adults were included in wave one, 19 014 720 in wave two, 18 932 050 in wave three, 19 097 970 in wave four, and 19 226 475 in wave five. Crude COVID-19-related death rates per 1000 person-years decreased from 4·48 deaths (95% CI 4·41-4·55) in wave one to 2·69 (2·66-2·72) in wave two, 0·64 (0·63-0·66) in wave three, 1·01 (0·99-1·03) in wave four, and 0·67 (0·64-0·71) in wave five. In wave one, the standardised COVID-19-related death rates were highest in people aged 80 years or older, people with chronic kidney disease stage 5 or 4, people receiving dialysis, people with dementia or learning disability, and people who had received a kidney transplant (ranging from 19·85 deaths per 1000 person-years to 44·41 deaths per 1000 person-years, compared with from 0·05 deaths per 1000 person-years to 15·93 deaths per 1000 person-years in other subgroups). In wave two compared with wave one, in a largely unvaccinated population, the decrease in COVID-19-related mortality was evenly distributed across population subgroups. In wave three compared with wave one, larger decreases in COVID-19-related death rates were seen in groups prioritised for primary SARS-CoV-2 vaccination, including people aged 80 years or older and people with neurological disease, learning disability, or severe mental illness (90-91% decrease). Conversely, smaller decreases in COVID-19-related death rates were observed in younger age groups, people who had received organ transplants, and people with chronic kidney disease, haematological malignancies, or immunosuppressive conditions (0-25% decrease). In wave four compared with wave one, the decrease in COVID-19-related death rates was smaller in groups with lower vaccination coverage (including younger age groups) and conditions associated with impaired vaccine response, including people who had received organ transplants and people with immunosuppressive conditions (26-61% decrease). INTERPRETATION: There was a substantial decrease in absolute COVID-19-related death rates over time in the overall population, but demographic and clinical relative risk profiles persisted and worsened for people with lower vaccination coverage or impaired immune response. Our findings provide an evidence base to inform UK public health policy for protecting these vulnerable population subgroups. FUNDING: UK Research and Innovation, Wellcome Trust, UK Medical Research Council, National Institute for Health and Care Research, and Health Data Research UK.

Comparative effectiveness of BNT162b2 versus mRNA-1273 covid-19 vaccine boosting in England: matched cohort study in OpenSAFELY-TPP.

OBJECTIVE: To compare the effectiveness of the BNT162b2 mRNA (Pfizer-BioNTech) and mRNA-1273 (Moderna) covid-19 vaccines during the booster programme in England. DESIGN: Matched cohort study, emulating a comparative effectiveness trial. SETTING: Linked primary care, hospital, and covid-19 surveillance records available within the OpenSAFELY-TPP research platform, covering a period when the SARS-CoV-2 delta and omicron variants were dominant. PARTICIPANTS: 3 237 918 adults who received a booster dose of either vaccine between 29 October 2021 and 25 February 2022 as part of the national booster programme in England and who received a primary course of BNT162b2 or ChAdOx1. INTERVENTION: Vaccination with either BNT162b2 or mRNA-1273 as a booster vaccine dose. MAIN OUTCOME MEASURES: Recorded SARS-CoV-2 positive test, covid-19 related hospital admission, covid-19 related death, and non-covid-19 related death at 20 weeks after receipt of the booster dose. RESULTS: 1 618 959 people were matched in each vaccine group, contributing a total 64 546 391 person weeks of follow-up. The 20 week risks per 1000 for a positive SARS-CoV-2 test were 164.2 (95% confidence interval 163.3 to 165.1) for BNT162b2 and 159.9 (159.0 to 160.8) for mRNA-1273; the hazard ratio comparing mRNA-1273 with BNT162b2 was 0.95 (95% confidence interval 0.95 to 0.96). The 20 week risks per 1000 for hospital admission with covid-19 were 0.75 (0.71 to 0.79) for BNT162b2 and 0.65 (0.61 to 0.69) for mRNA-1273; the hazard ratio was 0.89 (0.82 to 0.95). Covid-19 related deaths were rare: the 20 week risks per 1000 were 0.028 (0.021 to 0.037) for BNT162b2 and 0.024 (0.018 to 0.033) for mRNA-1273; hazard ratio 0.83 (0.58 to 1.19). Comparative effectiveness was generally similar within subgroups defined by the primary course vaccine brand, age, previous SARS-CoV-2 infection, and clinical vulnerability. Relative benefit was similar when vaccines were compared separately in the delta and omicron variant eras. CONCLUSIONS: This matched observational study of adults estimated a modest benefit of booster vaccination with mRNA-1273 compared with BNT162b2 in preventing positive SARS-CoV-2 tests and hospital admission with covid-19 20 weeks after vaccination, during a period of delta followed by omicron variant dominance.

The DetectDeviatingCells algorithm was a useful addition to the toolkit for cellwise error detection in observational data.

OBJECTIVES: We evaluated the error detection performance of the DetectDeviatingCells (DDC) algorithm which flags data anomalies at observation (casewise) and variable (cellwise) level in continuous variables. We compared its performance to other approaches in a simulated dataset. STUDY DESIGN AND SETTING: We simulated height and weight data for hypothetical individuals aged 2-20 years. We changed a proportion of height values according to predetermined error patterns. We applied the DDC algorithm and other error-detection approaches (descriptive statistics, plots, fixed-threshold rules, classic, and robust Mahalanobis distance) and we compared error detection performance with sensitivity, specificity, likelihood ratios, predictive values, and receiver operating characteristic (ROC) curves. RESULTS: At our chosen thresholds error detection specificity was excellent across all scenarios for all methods and sensitivity was higher for multivariable and robust methods. The DDC algorithm performance was similar to other robust multivariable methods. Analysis of ROC curves suggested that all methods had comparable performance for gross errors (e.g., wrong measurement unit), but the DDC algorithm outperformed the others for more complex error patterns (e.g., transcription errors that are still plausible, although extreme). CONCLUSIONS: The DDC algorithm has the potential to improve error detection processes for observational data.

Factors associated with COVID-19 vaccine uptake in people with kidney disease: an OpenSAFELY cohort study.

OBJECTIVE: To characterise factors associated with COVID-19 vaccine uptake among people with kidney disease in England. DESIGN: Retrospective cohort study using the OpenSAFELY-TPP platform, performed with the approval of NHS England. SETTING: Individual-level routine clinical data from 24 million people across GPs in England using TPP software. Primary care data were linked directly with COVID-19 vaccine records up to 31 August 2022 and with renal replacement therapy (RRT) status via the UK Renal Registry (UKRR). PARTICIPANTS: A cohort of adults with stage 3-5 chronic kidney disease (CKD) or receiving RRT at the start of the COVID-19 vaccine roll-out was identified based on evidence of reduced estimated glomerular filtration rate (eGFR) or inclusion in the UKRR. MAIN OUTCOME MEASURES: Dose-specific vaccine coverage over time was determined from 1 December 2020 to 31 August 2022. Individual-level factors associated with receipt of a 3-dose or 4-dose vaccine series were explored via Cox proportional hazards models. RESULTS: 992 205 people with stage 3-5 CKD or receiving RRT were included. Cumulative vaccine coverage as of 31 August 2022 was 97.5%, 97.0% and 93.9% for doses 1, 2 and 3, respectively, and 81.9% for dose 4 among individuals with one or more indications for eligibility. Delayed 3-dose vaccine uptake was associated with younger age, minority ethnicity, social deprivation and severe mental illness-associations that were consistent across CKD severity subgroups, dialysis patients and kidney transplant recipients. Similar associations were observed for 4-dose uptake. CONCLUSION: Although high primary vaccine and booster dose coverage has been achieved among people with kidney disease in England, key disparities in vaccine uptake remain across clinical and demographic groups and 4-dose coverage is suboptimal. Targeted interventions are needed to identify barriers to vaccine uptake among under-vaccinated subgroups identified in the present study.

Recording of 'COVID-19 vaccine declined': a cohort study on 57.9 million National Health Service patients' records in situ using OpenSAFELY, England, 8 December 2020 to 25 May 2021.

BackgroundPriority patients in England were offered COVID-19 vaccination by mid-April 2021. Codes in clinical record systems can denote the vaccine being declined.AimWe describe records of COVID-19 vaccines being declined, according to clinical and demographic factors.MethodsWith the approval of NHS England, we conducted a retrospective cohort study between 8 December 2020 and 25 May 2021 with primary care records for 57.9 million patients using OpenSAFELY, a secure health analytics platform. COVID-19 vaccination priority patients were those aged ≥ 50 years or ≥ 16 years clinically extremely vulnerable (CEV) or 'at risk'. We describe the proportion recorded as declining vaccination for each group and stratified by clinical and demographic subgroups, subsequent vaccination and distribution of clinical code usage across general practices.ResultsOf 24.5 million priority patients, 663,033 (2.7%) had a decline recorded, while 2,155,076 (8.8%) had neither a vaccine nor decline recorded. Those recorded as declining, who were subsequently vaccinated (n = 125,587; 18.9%) were overrepresented in the South Asian population (32.3% vs 22.8% for other ethnicities aged ≥ 65 years). The proportion of declining unvaccinated patients was highest in CEV (3.3%), varied strongly with ethnicity (black 15.3%, South Asian 5.6%, white 1.5% for ≥ 80 years) and correlated positively with increasing deprivation.ConclusionsClinical codes indicative of COVID-19 vaccinations being declined are commonly used in England, but substantially more common among black and South Asian people, and in more deprived areas. Qualitative research is needed to determine typical reasons for recorded declines, including to what extent they reflect patients actively declining.

Comparative effectiveness of ChAdOx1 versus BNT162b2 covid-19 vaccines in health and social care workers in England: cohort study using OpenSAFELY.

OBJECTIVE: To compare the effectiveness of the BNT162b2 mRNA (Pfizer-BioNTech) and the ChAdOx1 (Oxford-AstraZeneca) covid-19 vaccines against infection and covid-19 disease in health and social care workers. DESIGN: Cohort study, emulating a comparative effectiveness trial, on behalf of NHS England. SETTING: Linked primary care, hospital, and covid-19 surveillance records available within the OpenSAFELY-TPP research platform, covering a period when the SARS-CoV-2 Alpha variant was dominant. PARTICIPANTS: 317 341 health and social care workers vaccinated between 4 January and 28 February 2021, registered with a general practice using the TPP SystmOne clinical information system in England, and not clinically extremely vulnerable. INTERVENTIONS: Vaccination with either BNT162b2 or ChAdOx1 administered as part of the national covid-19 vaccine roll-out. MAIN OUTCOME MEASURES: Recorded SARS-CoV-2 positive test, or covid-19 related attendance at an accident and emergency (A&E) department or hospital admission occurring within 20 weeks of receipt of the first vaccine dose. RESULTS: Over the duration of 118 771 person-years of follow-up there were 6962 positive SARS-CoV-2 tests, 282 covid-19 related A&E attendances, and 166 covid-19 related hospital admissions. The cumulative incidence of each outcome was similar for both vaccines during the first 20 weeks after vaccination. The cumulative incidence of recorded SARS-CoV-2 infection 20 weeks after first-dose vaccination with BNT162b2 was 21.7 per 1000 people (95% confidence interval 20.9 to 22.4) and with ChAdOx1 was 23.7 (21.8 to 25.6), representing a difference of 2.04 per 1000 people (0.04 to 4.04). The difference in the cumulative incidence per 1000 people of covid-19 related A&E attendance at 20 weeks was 0.06 per 1000 people (95% CI -0.31 to 0.43). For covid-19 related hospital admission, this difference was 0.11 per 1000 people (-0.22 to 0.44). CONCLUSIONS: In this cohort of healthcare workers where we would not anticipate vaccine type to be related to health status, we found no substantial differences in the incidence of SARS-CoV-2 infection or covid-19 disease up to 20 weeks after vaccination. Incidence dropped sharply at 3-4 weeks after vaccination, and there were few covid-19 related hospital attendance and admission events after this period. This is in line with expected onset of vaccine induced immunity and suggests strong protection against Alpha variant covid-19 disease for both vaccines in this relatively young and healthy population of healthcare workers.

Waning effectiveness of BNT162b2 and ChAdOx1 covid-19 vaccines over six months since second dose: OpenSAFELY cohort study using linked electronic health records.

OBJECTIVE: To estimate waning of covid-19 vaccine effectiveness over six months after second dose. DESIGN: Cohort study, approved by NHS England. SETTING: Linked primary care, hospital, and covid-19 records within the OpenSAFELY-TPP database. PARTICIPANTS: Adults without previous SARS-CoV-2 infection were eligible, excluding care home residents and healthcare professionals. EXPOSURES: People who had received two doses of BNT162b2 or ChAdOx1 (administered during the national vaccine rollout) were compared with unvaccinated people during six consecutive comparison periods, each of four weeks. MAIN OUTCOME MEASURES: Adjusted hazard ratios for covid-19 related hospital admission, covid-19 related death, positive SARS-CoV-2 test, and non-covid-19 related death comparing vaccinated with unvaccinated people. Waning vaccine effectiveness was quantified as ratios of adjusted hazard ratios per four week period, separately for subgroups aged ≥65 years, 18-64 years and clinically vulnerable, 40-64 years, and 18-39 years. RESULTS: 1 951 866 and 3 219 349 eligible adults received two doses of BNT162b2 and ChAdOx1, respectively, and 2 422 980 remained unvaccinated. Waning of vaccine effectiveness was estimated to be similar across outcomes and vaccine brands. In the ≥65 years subgroup, ratios of adjusted hazard ratios for covid-19 related hospital admission, covid-19 related death, and positive SARS-CoV-2 test ranged from 1.19 (95% confidence interval 1.14 to 1.24)to 1.34 (1.09 to 1.64) per four weeks. Despite waning vaccine effectiveness, rates of covid-19 related hospital admission and death were substantially lower among vaccinated than unvaccinated adults up to 26 weeks after the second dose, with estimated vaccine effectiveness ≥80% for BNT162b2, and ≥75% for ChAdOx1. By weeks 23-26, rates of positive SARS-CoV-2 test in vaccinated people were similar to or higher than in unvaccinated people (adjusted hazard ratios up to 1.72 (1.11 to 2.68) for BNT162b2 and 1.86 (1.79 to 1.93) for ChAdOx1). CONCLUSIONS: The rate at which estimated vaccine effectiveness waned was consistent for covid-19 related hospital admission, covid-19 related death, and positive SARS-CoV-2 test and was similar across subgroups defined by age and clinical vulnerability. If sustained to outcomes of infection with the omicron variant and to booster vaccination, these findings will facilitate scheduling of booster vaccination.

Safety of COVID-19 vaccination and acute neurological events: A self-controlled case series in England using the OpenSAFELY platform.

INTRODUCTION: We investigated the potential association of COVID-19 vaccination with three acute neurological events: Guillain-Barré syndrome (GBS), transverse myelitis and Bell's palsy. METHODS: With the approval of NHS England we analysed primary care data from >17 million patients in England linked to emergency care, hospital admission and mortality records in the OpenSAFELY platform. Separately for each vaccine brand, we used a self-controlled case series design to estimate the incidence rate ratio for each outcome in the period following vaccination (4-42 days for GBS, 4-28 days for transverse myelitis and Bell's palsy) compared to a within-person baseline, using conditional Poisson regression. RESULTS: Among 7,783,441 ChAdOx1 vaccinees, there was an increased rate of GBS (N = 517; incidence rate ratio 2·85; 95% CI2·33-3·47) and Bell's palsy (N = 5,350; 1·39; 1·27-1·53) following a first dose of ChAdOx1 vaccine, corresponding to 11.0 additional cases of GBS and 17.9 cases of Bell's palsy per 1 million vaccinees if causal. For GBS this applied to the first, but not the second, dose. There was no clear evidence of an association of ChAdOx1 vaccination with transverse myelitis (N = 199; 1·51; 0·96-2·37). Among 5,729,152 BNT162b2 vaccinees, there was no evidence of any association with GBS (N = 283; 1·09; 0·75-1·57), transverse myelitis (N = 109; 1·62; 0·86-3·03) or Bell's palsy (N = 3,609; 0·89; 0·76-1·03). Among 255,446 mRNA-1273 vaccine recipients there was no evidence of an association with Bell's palsy (N = 78; 0·88, 0·32-2·42). CONCLUSIONS: COVID-19 vaccines save lives, but it is important to understand rare adverse events. We observed a short-term increased rate of Guillain-Barré syndrome and Bell's palsy after first dose of ChAdOx1 vaccine. The absolute risk, assuming a causal effect attributable to vaccination, was low.

Planning a method for covariate adjustment in individually randomised trials: a practical guide.

BACKGROUND: It has long been advised to account for baseline covariates in the analysis of confirmatory randomised trials, with the main statistical justifications being that this increases power and, when a randomisation scheme balanced covariates, permits a valid estimate of experimental error. There are various methods available to account for covariates but it is not clear how to choose among them. METHODS: Taking the perspective of writing a statistical analysis plan, we consider how to choose between the three most promising broad approaches: direct adjustment, standardisation and inverse-probability-of-treatment weighting. RESULTS: The three approaches are similar in being asymptotically efficient, in losing efficiency with mis-specified covariate functions and in handling designed balance. If a marginal estimand is targeted (for example, a risk difference or survival difference), then direct adjustment should be avoided because it involves fitting non-standard models that are subject to convergence issues. Convergence is most likely with IPTW. Robust standard errors used by IPTW are anti-conservative at small sample sizes. All approaches can use similar methods to handle missing covariate data. With missing outcome data, each method has its own way to estimate a treatment effect in the all-randomised population. We illustrate some issues in a reanalysis of GetTested, a randomised trial designed to assess the effectiveness of an electonic sexually transmitted infection testing and results service. CONCLUSIONS: No single approach is always best: the choice will depend on the trial context. We encourage trialists to consider all three methods more routinely.

Comparison of methods for predicting COVID-19-related death in the general population using the OpenSAFELY platform.

BACKGROUND: Obtaining accurate estimates of the risk of COVID-19-related death in the general population is challenging in the context of changing levels of circulating infection. METHODS: We propose a modelling approach to predict 28-day COVID-19-related death which explicitly accounts for COVID-19 infection prevalence using a series of sub-studies from new landmark times incorporating time-updating proxy measures of COVID-19 infection prevalence. This was compared with an approach ignoring infection prevalence. The target population was adults registered at a general practice in England in March 2020. The outcome was 28-day COVID-19-related death. Predictors included demographic characteristics and comorbidities. Three proxies of local infection prevalence were used: model-based estimates, rate of COVID-19-related attendances in emergency care, and rate of suspected COVID-19 cases in primary care. We used data within the TPP SystmOne electronic health record system linked to Office for National Statistics mortality data, using the OpenSAFELY platform, working on behalf of NHS England. Prediction models were developed in case-cohort samples with a 100-day follow-up. Validation was undertaken in 28-day cohorts from the target population. We considered predictive performance (discrimination and calibration) in geographical and temporal subsets of data not used in developing the risk prediction models. Simple models were contrasted to models including a full range of predictors. RESULTS: Prediction models were developed on 11,972,947 individuals, of whom 7999 experienced COVID-19-related death. All models discriminated well between individuals who did and did not experience the outcome, including simple models adjusting only for basic demographics and number of comorbidities: C-statistics 0.92-0.94. However, absolute risk estimates were substantially miscalibrated when infection prevalence was not explicitly modelled. CONCLUSIONS: Our proposed models allow absolute risk estimation in the context of changing infection prevalence but predictive performance is sensitive to the proxy for infection prevalence. Simple models can provide excellent discrimination and may simplify implementation of risk prediction tools.

Mortality among Care Home Residents in England during the first and second waves of the COVID-19 pandemic: an observational study of 4.3 million adults over the age of 65.

BACKGROUND: Residents in care homes have been severely impacted by COVID-19. We describe trends in the mortality risk among residents of care homes compared to private homes. METHODS: On behalf of NHS England we used OpenSAFELY-TPP to calculate monthly age-standardised risks of death due to all causes and COVID-19 among adults aged >=65 years between 1/2/2019 and 31/03/2021. Care home residents were identified using linkage to Care and Quality Commission data. FINDINGS: We included 4,340,648 people aged 65 years or older on the 1st of February 2019, 2.2% of whom were classified as residing in a care or nursing home. Age-standardised mortality risks were approximately 10 times higher among care home residents compared to those in private housing in February 2019: comparative mortality figure (CMF) = 10.59 (95%CI = 9.51, 11.81) among women, and 10.87 (9.93, 11.90) among men. By April 2020 these relative differences had increased to more than 17 times with CMFs of 17.57 (16.43, 18.79) among women and 18.17 (17.22, 19.17) among men. CMFs did not increase during the second wave, despite a rise in the absolute age-standardised COVID-19 mortality risks. INTERPRETATION: COVID-19 has had a disproportionate impact on the mortality of care home residents in England compared to older residents of private homes, but only in the first wave. This may be explained by a degree of acquired immunity, improved protective measures or changes in the underlying frailty of the populations. The care home population should be prioritised for measures aimed at controlling COVID-19. FUNDING: Medical Research Council MR/V015737/1.

Overall and cause-specific hospitalisation and death after COVID-19 hospitalisation in England: A cohort study using linked primary care, secondary care, and death registration data in the OpenSAFELY platform.

BACKGROUND: There is concern about medium to long-term adverse outcomes following acute Coronavirus Disease 2019 (COVID-19), but little relevant evidence exists. We aimed to investigate whether risks of hospital admission and death, overall and by specific cause, are raised following discharge from a COVID-19 hospitalisation. METHODS AND FINDINGS: With the approval of NHS-England, we conducted a cohort study, using linked primary care and hospital data in OpenSAFELY to compare risks of hospital admission and death, overall and by specific cause, between people discharged from COVID-19 hospitalisation (February to December 2020) and surviving at least 1 week, and (i) demographically matched controls from the 2019 general population; and (ii) people discharged from influenza hospitalisation in 2017 to 2019. We used Cox regression adjusted for age, sex, ethnicity, obesity, smoking status, deprivation, and comorbidities considered potential risk factors for severe COVID-19 outcomes. We included 24,673 postdischarge COVID-19 patients, 123,362 general population controls, and 16,058 influenza controls, followed for ≤315 days. COVID-19 patients had median age of 66 years, 13,733 (56%) were male, and 19,061 (77%) were of white ethnicity. Overall risk of hospitalisation or death (30,968 events) was higher in the COVID-19 group than general population controls (fully adjusted hazard ratio [aHR] 2.22, 2.14 to 2.30, p < 0.001) but slightly lower than the influenza group (aHR 0.95, 0.91 to 0.98, p = 0.004). All-cause mortality (7,439 events) was highest in the COVID-19 group (aHR 4.82, 4.48 to 5.19 versus general population controls [p < 0.001] and 1.74, 1.61 to 1.88 versus influenza controls [p < 0.001]). Risks for cause-specific outcomes were higher in COVID-19 survivors than in general population controls and largely similar or lower in COVID-19 compared with influenza patients. However, COVID-19 patients were more likely than influenza patients to be readmitted or die due to their initial infection or other lower respiratory tract infection (aHR 1.37, 1.22 to 1.54, p < 0.001) and to experience mental health or cognitive-related admission or death (aHR 1.37, 1.02 to 1.84, p = 0.039); in particular, COVID-19 survivors with preexisting dementia had higher risk of dementia hospitalisation or death (age- and sex-adjusted HR 2.47, 1.37 to 4.44, p = 0.002). Limitations of our study were that reasons for hospitalisation or death may have been misclassified in some cases due to inconsistent use of codes, and we did not have data to distinguish COVID-19 variants. CONCLUSIONS: In this study, we observed that people discharged from a COVID-19 hospital admission had markedly higher risks for rehospitalisation and death than the general population, suggesting a substantial extra burden on healthcare. Most risks were similar to those observed after influenza hospitalisations, but COVID-19 patients had higher risks of all-cause mortality, readmission or death due to the initial infection, and dementia death, highlighting the importance of postdischarge monitoring.

Transparency of high-dimensional propensity score analyses: Guidance for diagnostics and reporting.

PURPOSE: The high-dimensional propensity score (HDPS) is a semi-automated procedure for confounder identification, prioritisation and adjustment in large healthcare databases that requires investigators to specify data dimensions, prioritisation strategy and tuning parameters. In practice, reporting of these decisions is inconsistent and this can undermine the transparency, and reproducibility of results obtained. We illustrate reporting tools, graphical displays and sensitivity analyses to increase transparency and facilitate evaluation of the robustness of analyses involving HDPS. METHODS: Using a study from the UK Clinical Practice Research Datalink that implemented HDPS we demonstrate the application of the proposed recommendations. RESULTS: We identify seven considerations surrounding the implementation of HDPS, such as the identification of data dimensions, method for code prioritisation and number of variables selected. Graphical diagnostic tools include assessing the balance of key confounders before and after adjusting for empirically selected HDPS covariates and the identification of potentially influential covariates. Sensitivity analyses include varying the number of covariates selected and assessing the impact of covariates behaving empirically as instrumental variables. In our example, results were robust to both the number of covariates selected and the inclusion of potentially influential covariates. Furthermore, our HDPS models achieved good balance in key confounders. CONCLUSIONS: The data-adaptive approach of HDPS and the resulting benefits have led to its popularity as a method for confounder adjustment in pharmacoepidemiological studies. Reporting of HDPS analyses in practice may be improved by the considerations and tools proposed here to increase the transparency and reproducibility of study results.

Rates of serious clinical outcomes in survivors of hospitalisation with COVID-19 in England: a descriptive cohort study within the OpenSAFELY platform.

Background: Patients surviving hospitalisation for COVID-19 are thought to be at high risk of cardiometabolic and pulmonary complications, but quantification of that risk is limited. We aimed to describe the overall burden of these complications in people after discharge from hospital with COVID-19.   Methods: Working on behalf of NHS England, we used linked primary care records, death certificate and hospital data from the OpenSAFELY platform. We constructed three cohorts: patients discharged following hospitalisation with COVID-19, patients discharged following pre-pandemic hospitalisation with pneumonia, and a frequency-matched cohort from the general population in 2019. We studied seven outcomes: deep vein thrombosis (DVT), pulmonary embolism (PE), ischaemic stroke, myocardial infarction (MI), heart failure, AKI and new type 2 diabetes mellitus (T2DM) diagnosis. Absolute rates were measured in each cohort and Fine and Gray models were used to estimate age/sex adjusted subdistribution hazard ratios comparing outcome risk between discharged COVID-19 patients and the two comparator cohorts. Results: Amongst the population of 77,347 patients discharged following hospitalisation with COVID-19, rates for the majority of outcomes peaked in the first month post-discharge, then declined over the following four months. Patients in the COVID-19 population had markedly higher risk of all outcomes compared to matched controls from the 2019 general population. Across the whole study period, the risk of outcomes was more similar when comparing patients discharged with COVID-19 to those discharged with pneumonia in 2019, although COVID-19 patients had higher risk of T2DM (15.2 versus 37.2 [rate per 1,000-person-years for COVID-19 versus pneumonia, respectively]; SHR, 1.46 [95% CI: 1.31 - 1.63]).  Conclusions: Risk of cardiometabolic and pulmonary adverse outcomes is markedly raised following discharge from hospitalisation with COVID-19 compared to the general population. However, excess risks were similar to those seen following discharge post-pneumonia. Overall, this suggests a large additional burden on healthcare resources.