Outcomes in acute/general medical inpatients in the Hospital Episode Statistics

Oxford University Hospitals NHS Foundation Trust (OUH) requires HES and linked mortality data for the purpose of the Antibiotic Reduction & Konservation (ARK) project.

The objective for processing is to identify whether there is any evidence from existing electronic health record data to support the key prescriber concern that early antibiotic review leading to reduced antimicrobial usage will cause greater rates of treatment failure/mortality.

Antibiotic overuse increases the risks of bacteria developing resistance and therefore antimicrobial resistance is one of the greatest threats facing healthcare today. However, reducing antibiotic use is not straightforward. In primary care, trials have shown that it is possible to do this by either not prescribing antibiotics or using a delayed (“wait and see”) prescription. However, this approach isn’t useful in secondary care, where patients present acutely unwell and it is generally unclear at the point of presentation whether or not they have an underlying bacterial illness.

ARK is an NIHR-funded Programme Grant, awarded to the Oxford University Hospitals NHS Foundation Trust (OUH) through competitive funding in 2016, which is trying to address this challenge. ARK’s overarching aim is to reduce the incidence of serious infections caused by antibiotic-resistant bacteria in the future, through substantially and safely reducing antibiotic use in hospitals now. The programme has three specific research questions

(i) how can antibiotic prescription ‘review & revise’ strategies be implemented optimally to reduce antibiotic use safely within hospitals?

(ii) can a feasible inexpensive package of interventions that increase prescriber compliance with ‘review & revise’, and patient acceptability of shorter antibiotic therapy durations driven by ‘review & revise’ be built?

(iii) are ‘review & revise’ strategies cost-effective across a range of scenarios reflecting plausible associations between antibiotic use now and future resistance leading to loss of antibiotic options?

The goal of reducing total antibiotic burden in acute/general medical inpatients by at least 15% will be addressed through 6 work-packages (WP). WP1-WP3 will provide underpinning data for design and piloting in WP4 of a ‘review & revise’ intervention package for inpatients/carers and healthcare professionals. WP5 will evaluate its effectiveness and safety, and WP6 will conduct health-economic evaluations.

This request for data is for observational analysis as part of WP2. As above, the objective is to identify whether there is any evidence from existing electronic health record data to support the key prescriber concern that early antibiotic review leading to reduced antimicrobial usage will cause greater rates of treatment failure/mortality. National data is requested to give OUH the greatest power to identify these associations. Further, analyses will effectively compare adjusted (or standardised) event rates calculated per Trust with measures of antibiotic consumption, also requiring national data. 8 years data is requested to enable OUH to estimate how any associations might vary over time, as antibiotic use also varies within individual Trusts. Mortality is an absolutely key outcome to assess absolute risks associated with greater or lesser antibiotic use, since it is objective and most relevant to patients. OUH are only able to assess treatment failure indirectly through re-admissions, and ultimately mortality.

Oxford University Hospitals NHS Foundation Trust is the administering authority for the programme of research. The University of Oxford is a collaborator in the programme. For the purpose of WP2, individuals from the University of Oxford are involved but these individuals either have employment contracts or honorary contracts with the OUH and are acting as agents of OUH when controlling or processing the data under this Agreement. OUH is the sole data controller in respect of the data under this Agreement and its use for WP2. Data will be stored and analysed within NHS firewalls, hence the Trust is the sole controller and processor. No other individual acting on behalf of another organisation is involved in determining the purpose for or manner of use of the data or will process the data under this Agreement.

Case-mix adjusted outcomes in patients admitted to acute/general medicine will be compared with Trust-level antibiotic usage data from the English Surveillance Programme for Antimicrobial Utilisation and Resistance (ESPAUR), in an observational ecological (Trust-level) analysis. Hospital-level data will be used as a proxy for consumption in acute/general medicine as speciality-level data is not yet available in ESPAUR. Four outcomes will be considered:

• Mortality by 14 and 30 days after admission (in and out of hospital) (binary indicator)

• Admission to intensive care unit or high-dependency unit within this admission (identified from number of days of high-dependency/augmented care within each spell)

• Length of stay of the admission spell, both to actual discharge date and date medically ready for discharge

• Re-admission (non-elective) within 30 days of discharge (regardless of re-admission speciality)

Antibiotic use will be considered at the level of each Trust in terms of defined daily doses (DDDs: a World Health Organisation system for standardising antibiotic usage), overall and by drug class, per quarter, per year and over the 5 year study period. Broad spectrum will be defined as: co-amoxiclav; meropenem; second (eg cefuroxime), third (e.g. ceftriaxone ceftazidime) or fourth (e.g. cefepime) generation cephalosporins; quinolones; piperacillin/tazobactam.

The following potential confounders will be adjusted for age at admission (years); gender; ethnicity; index of multiple deprivation (IMD) score; Clinical Classifications Software (CCS) group of primary diagnosis code; Charlson co-morbidity score (calculated from secondary diagnosis codes associated with the first consultant episode within each spell, or the second consultant episode if the first is A&E); immunosuppression (calculated from the secondary diagnosis codes); intended management (admitted overnight, not admitted overnight, etc.); patient classification (actual management: admitted overnight, not admitted overnight etc.); admission day of the week, day of the year; calendar year; admission method; admission source; number of admissions (excluding as day case) in the previous year.

The null hypothesis is that there is no association between Trust-level antibiotic usage and outcomes in patients admitted to acute/general medicine.

The analysis will be conducted by medical statisticians only. Any individual using the data for such analysis will be either employed by Oxford University Hospitals NHS Foundation Trust (OUH) or by the University of Oxford, with an honorary contract with the Trust in place. The data will be stored and processed on an NHS server housed within OUH, within the NHS N3 firewall. A database manager/software engineer will process the data onto the NHS server.

No third parties will store, process or access record-level data.

All organisations party to this Agreement must comply with the Data Sharing Framework Contract requirements, including those regarding the use (and purposes of that use) by “Personnel” (as defined within the Data Sharing Framework Contract - i.e. employees, agents and contractors of the Data Recipient who may have access to that data).

The results of these analyses will be published in a peer-reviewed medical journal - target submission date is March 2019, but current experience is that manuscripts are taking a long time to get through the review and acceptance process with a large number of queries from reviewers that would likely need access to the original datasets to address adequately. For this reason, whilst OUH anticipate that all work on the data should be complete by 31 December 2019, OUH are requesting access to the data through to 31 December 2020 to ensure that OUH can complete the publication process before destroying the data.

No record-level data will be an output and any small numbers will be suppressed in line with the HES analysis guide. These analyses are comparing Trust-level outcome data with Trust-level antibiotic usage (see protocol for further details). Therefore, even though OUH will make every attempt to adjust for case-mix and other factors that could influence outcomes and antibiotic usage, such adjustments may be imperfect, and any residual association does not necessarily imply causation. Interpretation of results from these analyses will explicitly highlight this. Standardly, such biases can occur in either direction, making it impossible to work out whether effects observed in observational studies are optimistic or conservative; hence, the need to rely on randomised controlled trials for unbiased inference regarding intervention effects. However, a priori, in this specific context of adults admitted to acute general/medical specialities, it is highly likely that any residual bias is primarily in one direction, namely that “less sick” individuals (at lower risk of the various clinical outcomes) have lower antibiotic exposure. Given this, not observing harm in these observational analyses is necessary to conclude that no harm would be associated with an intervention to reduce antibiotic use in this group of patients. If one observes evidence for harm from this observational analysis, after adjustment for as many confounders as possible, this would seriously undermine the rationale for the proposed trial within the larger Programme Grant, necessitating high-level review of the larger project.

The Trust will also disseminate findings through the patient and public engagement activities ongoing within ARK and the Oxford Biomedical Research Centre within which key team members also work.

Within ARK is a Community Advisory Board (CAG) made up of 4 patient representations, one of whom also sites as the community member on the Programme Steering Committee. These individuals have met three times so far through the course of the project to review outputs from other parts of the programme. When the analyses from this workstream are complete, the researchers at the OUH plan to meet with them to discuss how they feel the findings should be best disseminated to patients and the public. At minimum, a blog will be written about each paper published from the research group within which ARK is situation (see: https://modmedmicro.nsms.ox.ac.uk/mmm-news-and-updates/). These are always reviewed by PPI representatives before being posted to try to make them as accessible as possible and ARK CAB members would do that for this study. Depending on whether PPI representatives think papers are interesting to the general public, there may be a press release or an opinion piece may be produced and published. Any blog or press release would be published on the main ARK website as well as the Modernising Medical Microbiology (MMM) Group’s website. ARK and MMM also have Twitter feeds which would also be used for dissemination.

On a wider scale, all the PPI for the different projects within the MMM Group, including ARK, are managed under a PPI strategy which is published at: http://modmedmicro.nsms.ox.ac.uk/mmm-ppie-strategy/. The PPI Steering Committee meets face to face approximately every 3-4 months. It currently contains 8 active community members. The Committee discusses all projects and the various activities being planned for each and feedback is given. Information about the 2018 activities are detailed on https://modmedmicro.nsms.ox.ac.uk/ppi/. This includes a mixture of schools work, museum talks, and a Royal Society Summer Science exhibition lasting 6 days. A 6-monthly newsletter is provided to ~150 people who have signed up via the website or through these activities and analyses like ARK are sometimes summarised in these newsletters. Again, the PPI representatives advise on what they think are the most interesting aspects of the work and check the language is appropriate for the target audience before it is published.

The intended benefits from this analysis were to inform the development of a behavioural intervention to reduce antibiotic usage in acute/general medical inpatients. If the behaviour intervention reduces antibiotic use without changing patient outcomes, then it would be immediately ready for NHS deployment. It would be freely available to the NHS (under the terms of the contract with NIHR).

Providing that the results do not suggest that substantial harm could result from reducing antibiotic use (in which case the whole Programme Grant will be reviewed by the funders and the Programme Steering Committee), the results will be used, together with other published studies, to inform the development of a ‘review &revise’ behavioural intervention package for inpatients/carers and healthcare professionals aimed at reducing antibiotic usage. This behavioural intervention will then be tested in a large cluster-randomised stepped-wedge trial during years 3-5 of the Programme Grant.

The analysis completed to date has contributed, together with results from other published studies, to the development of a ‘review &revise’ behavioural intervention package for inpatients/carers and healthcare professionals aimed at safely reducing antibiotic usage in acute/general medical inpatients. It has provided data to support the safety of reduced antibiotic usage, by demonstrating that there is no association between mortality (and other outcomes) and antibiotic usage across Trusts with varying degrees of antibiotic usage. If the behaviour intervention (currently being tested in a large Trust (cluster)-randomised stepped wedge trial) reduces antibiotic use without changing patient outcomes, then it would be immediately ready for NHS deployment. It would be freely available to the NHS (under the terms of the contract with NIHR).

As above, results from the planned analyses generated to date have contributed to the development of the ARK intervention which is currently undergoing evaluation in a large cluster-randomised trial. Continuing access to the data is required to complete the write up of the results for publication in a peer-reviewed manuscript, and to appropriately address queries which will inevitably arise from reviewers.