OpenSAFELY NHS Service Restoration Observatory
Introduction
During the COVID-19 pandemic, routine healthcare services faced significant levels of disruption. The NHS in England responded to the emerging pandemic by stopping non-urgent work in hospitals, and suggesting that, where possible, patients should have non-urgent primary care appointments remotely. On July 21st, one of the many recommendations by NHS England was to restore services to near-normal levels where clinically appropriate before winter, whilst remaining vigilant for a second wave.
OpenSAFELY is a new secure analytics platform for electronic patient records built by our group on behalf of NHS England to deliver urgent academic and operational research during the pandemic: analyses can currently run across all patients’ full raw pseudonymised primary care records at 40% of English general practices, with patient-level linkage to various sources of secondary care data; all code and analysis is shared openly for inspection and re-use. One aim of OpenSAFELY is to assess “COVID Aftershocks” where we monitor data to measure and mitigate the indirect health impacts of COVID-19. In order to produce the best possible insights across a range of diverse topics using this huge volume of activity and data, we are working with NHS England to create a programme of work that we have titled the OpenSAFELY NHS Service Restoration Observatory.
Method
General practice clinical activity was described by conducting a retrospective cohort study using raw data from English NHS general practices. Methods are described in detail elsewhere and briefly below.
Data Source: Primary care records managed by TPP were analysed through OpenSAFELY, a data analytics platform created by our team on behalf of NHS England to address urgent COVID-19 research questions (opensafely.org). The dataset contains information on 23.8 million people registered with GP surgeries using TPP SystmOne software as at 30 September 2020.
Study Population: We included all patients registered with any practice using TPP EHR software (those with a registration beginning on or before 30 September 2020, which was still live until at least this date). All coded events between January 2019 and the end of September 2020 for this cohort were included.
Data Processing: We employed a data-driven approach in order to capture the most common coded events in primary care. We ranked codes according to the number of total occurrences in January-September 2020, excluding those with fewer than 1000. Data was grouped at the practice level. We used each patient’s latest practice (as at 30 September 2020) as their assigned practice for all activity throughout the study period. The total population in each practice was calculated as the total registered patients at 30 September 2020 and the same value used for every month.
Clinical Codes: Clinical codes included are described in each chapter.
Study Measures: We calculated the monthly incidence of each code (or group of codes) per 1000 registered patients at each practice. We calculated the median and deciles across all practices each month for each code, after excluding practices which never used that code in the study period. We present the data in time trend charts. All charts were collaboratively reviewed by clinicians and researchers; selected charts are shown in the Results section to illustrate key patterns, and all charts will be shown in the Appendix.
In the case of all grouped (“parent”) codes, we present a table of the top (up to) five “child” codes to illustrate examples of individual codes captured, and we plot time trends of the top most common code. Fewer than five child codes being shown indicates that fewer than five codes exist in the group, or fewer than five reached the 1,000 minimum 2020 activity threshold.
Classification of service restoration: With each chart we display the median value and interdecile range (IDR) for February, April and September 2020. April was identified as the first full month after full lockdown whilst September was the latest full month at the time of the initiating this analysis and before the “second wave” of infections would have been expected to influence health services once again. To aid interpretation, we provide an approximate classification based on changes to the median compared to baseline (the same month the previous year).
Service change classification
For April and September:
no change: activity remained within 15% of the baseline level; increase: an increase of >15% from baseline; small drop: a reduction of between 15% and 60% from baseline; large drop: a reduction of >60% from baseline.
Overall classification:
no change: no change in both April and September; increase: an increase in either April or September; sustained drop: a small or large drop in April which has not returned to within 15% of baseline by September 2020; recovery: a small or large drop in April, which returned to within 15% of baseline (“no change”) by September 2020.
Software and Reproducibility: Data management was performed using Python and the OpenSAFELY software, with data extracted via SQL Server Management Studio and analysis carried out using Python. All of the code used for data management and analyses is openly shared online for review and re-use (https://github.com/opensafely/restoration-observatory-intro-notebook).
Patient and Public Involvement: This analysis relies on the use of large volumes of patient data. Ensuring patient, professional and public trust is therefore of critical importance. Maintaining trust requires being transparent about the way OpenSAFELY works, and ensuring patient voices are represented in the design of research, analysis of the findings, and considering the implications. For transparency purposes we have developed a public website which provides a detailed description of the platform in language suitable for a lay audience; we will be co-developing an explainer video; and we have presented at a number of online public engagement events to key communities. To ensure the patient voice is represented, we are working closely with appropriate medical research charities.
We will publish analysis of each separate clinical area as a report chapter below. If you would like to get involved please do get in touch.