Learning From Patients (2): Developing a National Resource for Hip Fracture Research

The National Hip Fracture Database (NHFD) is a clinically led, web-based audit of hip fracture care that grew out of collaboration between the British Orthopaedic Association and the British Geriatrics Society and is now managed by the Royal College of Physicians (RCP) and commissioned by Healthcare Quality Improvement Partnership (HQIP).

The NHFD currently provides statistics on hospital performance outcomes but does not capture all variables to account for patient case mix, for example any patient who has a pre-existing health condition such as diabetes or heart attack.

The University of Oxford requires access to pseudonymised record level HES data linked with NHFD data plus vital status indicators for the purpose of a project which aims to:

1. Develop an improved risk adjustment model for the NHFD using data linkage and prospective collection of additional variables.

2. Use the new risk adjustment model to compare hip fracture patient outcomes between the UK, US, and Australia to identify opportunities for improving care.

3. Demonstrate the feasibility of using the enriched NHFD dataset to answer specific health policy questions that could inform future research priorities and health service planning.

This is a standalone PhD study undertaken by a researcher who holds a UCB-Oxford Prize Fellowship in Biomedical Research and is employed by University of Oxford and supported by a researcher from Yale University. UCB is a global biopharmaceutical company. The study aims to improve the case mix adjustment model to collect co-morbidities and enrich the data held in the NHFD. No data will be shared with UCB.

There are 70,000 hip fractures every year in the United Kingdom at a total cost exceeding £2 billion. Mortality is high, with 8.5% of patients dying within 30 days of admission. The NHFD captures data from 95% of hip fractures treated in England, Wales, and Northern Ireland; and is the largest comparable registry in the world. The NHFD exists primarily to drive quality improvement and has been used infrequently for research.

The NHFD reports NHS hospital outcomes (e.g. mortality) that are adjusted for differences in patient case mix using statistical techniques, such as multivariable regression. This is important because some hospitals might treat patients that are - for example - older and sicker than other hospitals. If statistical techniques are not used to "adjust" outcomes, then a hospital with older patients might appear to be worse than one with a younger population, even though it is actually providing better care. However, these models have been developed based on limited fields available within the dataset, such as age and pre-morbid mobility. It is possible that additional variables could have greater predictive value for outcomes reported by the NHFD. The majority of hip fractures occur in older adults and so it is likely that co-morbid diseases affect patient outcomes. Although the NHFD includes American Society of Anaesthesiologists (ASA) score (which is a measure of overall "sickness" on a scale from 1-5), other validated co-morbidity measures (such as the Charlson and Elixhauser indices, which convert each patient's diagnoses into a single numeric score) cannot be calculated because the NHFD does not record individual comorbidity diagnoses. Inadequate risk adjustment could result in failure to identify poorly performing hospitals and/or unfair attribution of poor outcomes to hospitals with unfavourable case mixes.

One of the three components of this project is to investigate the effect of the Best Practice Tariff (BPT) on hip fracture outcomes. The BPT is an initiative that was launched in April 2010 and provides additional payments to hospitals that meet defined quality criteria when treating patients with hip fractures, such as providing assessment by a senior orthogeriatrician and an operation within 36 hours. The National Hip Fracture Database (NHFD) was already in existence when the BPT was introduced and was chosen as the vehicle for determining which cases should attract the additional payment. One consequence was that this financial incentive dramatically increased the proportion of cases captured by the NHFD once the BPT was introduced. This creates an obvious problem for using the NHFD to evaluate the BPT as any change in outcomes might simply reflect differences in the cases being reported by individual hospitals. A solution is to use an external dataset (such as HES APC) that was capturing cases before and after the BPT was introduced and that is independent of the NHFD. The University of Oxford intend to use this dataset to undertake an interrupted time series “natural experiment” to look for any changes in outcome that came about after the BPT was introduced.

The aim of processing these data is to improve the NHFD risk adjustment model and develop this comprehensive national registry into a valuable resource for hip fracture research.

This study will mainly focus on a cohort of patients originally identified from the National Hip Fracture Database (NHFD), all of whom are aged >60 and have been treated for a hip fracture at a hospital in England. The University of Oxford has previously received an extract of HES data linked to this cohort under a separate Data Sharing Agreement (ref DARS-NIC-61090-T9Y0G) for the purpose of an unrelated study. The University of Oxford will take a copy of some of that data to reuse it for this purpose. The detailed process to do this is as follows:

1. From the dataset that the University of Oxford is authorised to hold and process under the Data Sharing Agreement DARS-NIC-61090-T9Y0G, the encrypted HESIDs will be extracted and supplied to NHS Digital along with the date of the hip fracture for each individual patient.

2. NHS Digital will decrypt those IDs to identify the individuals and extract linked ONS mortality data (date of death) which NHS Digital will use to calculate vital status at 30, 60, 90 and 365 days post-date of hip fracture.

3. NHS Digital will apply Type 2 patient objections to this list of patients. The data of any individual who has registered a Type 2 objection since the data was supplied under DARS-NIC-61090-T9Y0G will be removed.

4. NHS Digital will remove all identifiers and the date of death; re-encrypt the HESIDs using the same encryption key as for DARS-NIC-61090-T9Y0G, and disseminate the data, as pseudo-anonymised data, to the University of Oxford.

5. The University of Oxford would extract a copy of the linked HES and NHFD data (but not the linked ONS mortality data) supplied under DARS-NIC-61090-T9Y0G for all individuals whose data is included in the output file from NHS Digital. This data would be linked with the mortality indicators supplied and the resulting dataset will be stored separately from the dataset supplied under DARS-NIC-61090-T9Y0G with access controls ensuring the two will not be linked or accessed together.

6. In addition, NHS Digital will produce a new unlinked pseudonymised extract of HES data for which Type 2 patient objections will not be applied. This will be an extract of patients with a record of hip fracture (ICD-10 S72.0*, S72.1*, S72.2*) fracture between 1st January 2000 and the latest possible date with age >=60 at the time of diagnosis. For each patient, vital status at 30, 60, 90 and 365 days post-admission will be calculated. This will be supplied to the University of Oxford. This will be used as external validation of the Hip Fracture Best Practice Tariff work, as this initiative might itself also influence reporting of cases to the NHFD. The University of Oxford will not attempt to link this data with any other dataset.

The University of Oxford will then hold two separate datasets: one NHFD-HES with vital status indicators under this Data Sharing Agreement and a second as NHFD-HES-ONS with date of death under the Agreement DARS-NIC-61090-T9Y0G.

The respective datasets will only be used for the projects authorised under their respective Data Sharing Agreements. The two datasets will be stored on separate hardware-encrypted hard drives that can only be accessed by physically entering a passcode onto the device itself. This will mean that the datasets will be physically separated (i.e. on separate devices) and accessed using different passcodes. As well as being separated in space, the datasets will also be separated in time as they will not be accessed simultaneously.

The researcher from the University of Oxford has previously undertaken analyses directly from these hard drives so that data does not have to be stored on a single computer or network. If it does become necessary to transfer a dataset to a University of Oxford computer (e.g. because of the size or analysis complexity), this will only be done temporarily while the dataset is being worked on. It will be returned to the appropriate hard drive at the end of the day to mitigate the risk of "cross contamination" between the two datasets.

The linked mortality data provided under Data Sharing Agreement DARS-NIC-61090-T9Y0G will not be copied, linked or used in any way for the purpose of this project under this Data Sharing Agreement.

The datasets will be stored on encrypted hard drives in a locked safe in the Botnar Secure Computing Room, which is an access-controlled facility within the Botnar Research Centre. The Botnar is one of three facilities run by the Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences (NDORMS) at the University of Oxford. It is based on the Nuffield Orthopaedic Centre site, which is operated by Oxford University Hospitals NHS Foundation Trust. In addition to access controls, there is CCTV in the building and 24/7 security. Data will only be analysed within the Botnar Secure Computing room on computers that are not connected to any external network. Security within the room is regularly audited by the Information Governance Manager and the computers are kept up-to-date with anti-virus software.

The Kadoorie Centre for Critical Care Research is a second site operated by NDORMS and will serve as a secondary processing location for analysing derived (i.e. less sensitive) data. This data will only include fields that are required for the analyses being undertaken at any one time. Sensitive variables (e.g. individual diagnoses) will be recoded (e.g. to Charlson co-morbidity score). Data at the Kadoorie Centre will be stored on an encrypted hard drive in a locked filing cabinet in a secure office. This room requires two levels of swipe card access and a physical key to open the door itself. Data will be analysed on a computer that is not connected to any external network and analyses will be undertaken using the encrypted hard drive rather than by transferring data to the computer itself. This will mean that no sensitive data will be stored on the computer hard drive overnight.

Data will only be transported occasionally between the Botnar Secure Computing Room and the Kadoorie Centre. When such transfers do occur, it will always be a direct journey and the data will always be stored on an AES-256 encrypted hard drive.

Data from the NHFD and NHS Digital will only be stored and accessed at the University of Oxford. It will not be linked to any datasets other than the NHFD. There will be no requirement nor attempt to re-identify individuals from the data by the research team.

The members of the research team are all directly employed or working under appropriate supervision on behalf of University of Oxford. All researchers who will have access to the data are subject to the same policies, procedures and sanctions as substantive employees. Supporting the research team will be one individual on secondment from the University of Yale in the USA. This individual will work under an honorary contract with the University of Oxford and will only access the data on site at the University of Oxford under the same conditions as substantive employees of the University of Oxford described above.

All organisations party to this agreement will 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 data will only be used for the purposes described in this Agreement. The data will not be made available to any third parties other than those specified except in the form of aggregated outputs with small numbers suppressed in line with the HES Analysis Guide.

For clarification, comparisons of hip fracture patient outcomes between the UK, US and Australia will be undertaken using the new risk adjustment model produced as a result of processing the data covered by this Agreement. The data will not be shared with third parties in the UK, US or Australia. As above, only data in the form of aggregated outputs with small numbers suppressed in line with the HES Analysis Guide will be available to any third parties.

This project will principally be reported in a doctoral thesis by the researcher (DPhil in Musculoskeletal Sciences) and this is intended for submission to the University of Oxford in late 2018. The findings from objective 1 (improving the NHFD risk adjustment model) will be communicated directly to the Healthcare Quality Improvement Partnership (HQIP) which commissions the Royal College of Physicians to manage the National Hip Fracture Database (NHFD). This is anticipated to occur in the latter half of 2018 and in time for a revised case mix adjustment algorithm to be incorporated into the NHFD 2018 annual report. If additional variables are required for collection in future iterations of the NHFD, this could happen as early as 2019.

The case mix adjustment data will be presented in a research methods paper that is likely to appear in an open access journal. This will provide a reference point for other research papers using the NHFD and needing to justify their statistical approach to risk adjustment. This will be submitted for publication in advance of the doctoral thesis submission in 2018.

The other papers (e.g. international benchmarking and health service research projects) will be submitted for publication in peer-reviewed journals that will depend on the study findings. The target journals are usually the Bone & Joint Journal (for findings that are applicable to orthopaedic surgeons) and the British Medical Journal (in the case of findings that are of interest to a broad clinical audience). These are intended for submission in early 2019. The University anticipate presenting all available results for each project at the British Orthopaedic Association Annual Congress 2019.

All outputs will be aggregated with small numbers suppressed in line with the HES Analysis Guide.

The two changes likely to occur as a result of this project are:

(1) Revision of the existing risk adjustment model within the NHFD and;

(2) Refinement of the variables collected by the NHFD for case mix adjustment.

These changes will bring a number of benefits for patients and commissioning organisations. In particular, it will improve the ability of the NHFD to fulfil a key audit role, which is to identify hospitals with worse-than-expected and better-than-expected hip fracture outcomes. Suboptimal risk adjustment may lead to hospitals with worse case mix being inappropriately highlighted as providing poor care and - conversely - failure to identify some hospitals that provide sub-standard care but that benefit from treating "fitter" patients. This will help drive quality of care for older adults with hip fractures. It will also open up the NHFD as a tool for hip fracture researchers studying factors associated with hip fracture outcomes. The project will achieve this final outcome in a number of ways, i.e. by (1) showcasing the value of the NHFD to the clinical research community, (2) describing the degree to which the variables in the existing NHFD can predict mortality, i.e. so increasing researcher confidence in their ability to risk adjust hip fracture outcomes.

The risk adjustment model in the existing NHFD could be improved in time for the next annual report in 2018. If optimal risk adjustment requires the collection of new variables, this could take place from 2019.

Yielded Benefits is not a requirement for new applications.