OpenSAFELY: The impact of COVID-19 on azathioprine, leflunomide and methotrexate monitoring, and factors associated with change in monitoring rate

Aims The COVID-19 pandemic created unprecedented pressure on healthcare services. This study investigates whether disease-modifying antirheumatic drug (DMARD) safety monitoring was affected during the COVID-19 pandemic. Methods A population-based cohort study was conducted using the OpenSAFELY platform to access electronic health record data from 24.2 million patients registered at general practices using TPP's SystmOne software. Patients were included for further analysis if prescribed azathioprine, leflunomide or methotrexate between November 2019 and July 2022. Outcomes were assessed as monthly trends and variation between various sociodemographic and clinical groups for adherence with standard safety monitoring recommendations. Results An acute increase in the rate of missed monitoring occurred across the study population (+12.4 percentage points) when lockdown measures were implemented in March 2020. This increase was more pronounced for some patient groups (70–79 year-olds: +13.7 percentage points; females: +12.8 percentage points), regions (North West: +17.0 percentage points), medications (leflunomide: +20.7 percentage points) and monitoring tests (blood pressure: +24.5 percentage points). Missed monitoring rates decreased substantially for all groups by July 2022. Consistent differences were observed in overall missed monitoring rates between several groups throughout the study. Conclusion DMARD monitoring rates temporarily deteriorated during the COVID-19 pandemic. Deterioration coincided with the onset of lockdown measures, with monitoring rates recovering rapidly as lockdown measures were eased. Differences observed in monitoring rates between medications, tests, regions and patient groups highlight opportunities to tackle potential inequalities in the provision or uptake of monitoring services. Further research should evaluate the causes of the differences identified between groups.


| INTRODUCTION
The COVID-19 pandemic markedly disrupted delivery of primary care services globally, 1 and within the National Health Service (NHS), with a notable impact on chronic disease management. 2 Disease-modifying antirheumatic drugs (DMARDs) are typically prescribed for chronic autoimmune conditions such as rheumatoid arthritis, Crohn's disease and severe psoriasis.Many DMARDs have narrow therapeutic indexes, and can cause potentially fatal adverse events such as blood dyscrasias and liver toxicity. 3,4To mitigate these risks, DMARDs are typically initiated by a specialist clinician in secondary care.Once patients have been stabilized on treatment, they can be transferred to primary care under a shared care arrangement, which specifies longterm safety monitoring recommendations (Table 1) for general practitioners (GPs) to oversee.Under the arrangement GPs have guidance on when to suspend a prescription or seek specialist advice, e.g. if they observe unsafe monitoring results, nonadherence to monitoring or emergence of adverse effects.
The first wave of the COVID-19 pandemic in March 2020 prompted a national lockdown in the UK.GP practices made several changes with the aim of reducing viral transmission, ranging from reducing face-to-face appointments, to temporary practice closure where outbreaks occurred amongst practice staff. 5Moreover, many patients worldwide chose to avoid healthcare services for fear of contracting COVID-19, [6][7][8] and those who were recommended to isolate through shielding faced additional barriers to healthcare access. 9In April 2020, the British Medical Association and Royal College of GPs issued guidance suggesting clinicians consider DMARD monitoring a high priority task. 10However, The NHS Specialist Pharmacy Service and British Society of Rheumatology suggested that monitoring intervals for certain medications, including DMARDs, could be considered for extension. 11,12Several laboratory tests in GP practices experienced substantial activity reductions during lockdown, 13 including tests related to methotrexate monitoring. 14To date, no research has described the extent to which adherence with wider DMARD monitoring recommendations was affected during the COVID-19 pandemic.
OpenSAFELY is a secure analytics platform for electronic patient records built with the approval of NHS England to deliver urgent academic 15 and operational NHS service research 13,16 on the impacts of the pandemic.Analyses run across patients' full raw pseudonymized primary care records at English GP practices with patient-level linkage to sources of secondary care data.

What is already known about this subject
• Disease-modifying antirheumatic drugs (DMARDs) carry risks of serious adverse effects, so national guidelines recommend that patients taking them adhere to regular safety monitoring.
• The COVID-19 pandemic disrupted general monitoring services in England, with certain groups disproportionately affected in their health outcomes.

What does this study adds
• DMARD monitoring rates deteriorated during the COVID-19 pandemic but recovered quickly after the easing of lockdown measures.
• Substantial variations in DMARD safety monitoring rates were found amongst different demographic, clinical and regional subgroups in England-suggesting that opportunities exist to address inequalities in safety monitoring.
We set out to assess the impact of COVID-19 on the monitoring of DMARDs using OpenSAFELY, and evaluate whether any such effect was associated with other variables, such as patient demographics or comorbidities.

| Study design
We conducted a retrospective cohort study across 24 million patients by using EHR data from all GP practices in England supplied by software provider TPP.From this data we sought to identify patients prescribed DMARDs between November 2019 and July 2022.

| Study population
We included all patients who were: alive; aged 18-120 years; and registered with a TPP practice.Index of multiple deprivation (IMD) 17 and rurality classification 18 were derived for each patient based on their address.We excluded patients missing age, sex, IMD and rurality classification, since their omission may indicate generally poor data quality.
We defined patients on each DMARD as those recorded with at least 2 prescriptions (one issued within 3 months before the search date, and the other between 3 and 6 months before the search date).
Three-month search windows for prescriptions were chosen because in England GPs issue 93% of repeat prescriptions for a duration of 3 months or less. 19athioprine, leflunomide and methotrexate were chosen as a representative sample of DMARDs.These 3 medications are consistently classified as shared care across England, which avoids introducing potential confounding in the analysis of monitoring rates from inconsistent shared care statuses.We created codelists for each DMARD using NHS dictionary of medicines and devices (dm+d).

| Study measures and statistics
Monitoring tests relevant to the prescribed medication were selected (Table 1).These included full blood counts (FBCs), liver function tests (LFTs), urea and electrolytes (U&E), and blood pressure (BP).We excluded weight (for leflunomide) because it is not consistently listed as a monitoring requirement in all sources; namely the British National Formulary and Summary of Product Characteristics.Moreover, BP and weight checks would usually be carried out at the same time, so BP can be considered broadly representative of physical health checks.
We selected patient groups which were relevant to health inequality priorities in England, 20 or deemed susceptible to variation in monitoring.We grouped age into 7 categories (18-29, 30-39, 40-49, 50-59, 60-69, 70-79, ≥80 years), sex into male or female, ethnicity into 5 categories (Black, Mixed, South Asian, White, Other), region into 9 categories (East, East Midlands, London, North East, North West, South East, South West, West Midlands, Yorkshire and The Humber), rural-urban into 8 categories (1: most urban, to 8: most rural), 18 and IMD into 5 quintiles.Dementia, learning disability, serious mental illness, care home status and housebound status were grouped as binary characteristics.
Codelists were specified to represent monitoring tests and patient groups using SNOMED CT.Each monitoring test typically represents a set of tests conducted together, therefore a single representative test was chosen (Table S1).For example, where FBCs are recommended we looked for the presence of a Red blood cell count code.
Patients were defined as having missed monitoring if any of the relevant tests required for their given drug were not recorded in T A B L E 1 Examples of licensed indications typically seen in primary care, and long-term monitoring requirements for 3 commonly prescribed disease-modifying antirheumatic drugs (DMARDs).the last 3 months.Indicators (see Table S2) expressed the proportion of patients deemed to have missed monitoring (numerator), of all relevant patients prescribed the DMARD(s) being assessed (denominator).
Higher percentages represent poorer monitoring performance.Each indicator was specified in analytic code within the OpenSAFELY framework, and was calculated from patient counts rounded to the nearest 5 to ensure anonymity.

| Trend in DMARD monitoring rates
The numerator and denominator were generated for each indicator at monthly intervals between November 2019 and July 2022, then percentages calculated for each time-period.Note that each month in our results reflects a 3-month rolling average corresponding to data for the named month and the 2 months preceding it.For example, May 2020 represents patients not having tests from 1 March 2020 to 31 May 2020.Time-periods of interest are defined in Table 2.
Total counts of the numerator and denominator were calculated for each indicator across the full study period.In this cumulative data, repeated events were counted for each period in which the event occurs, e.g. if a patient missed monitoring in 2 separate periods this was represented as 2 separate events.To report the extent of monitoring being repeatedly missed for the same patients, we also calculated the ratio of total missed monitoring events to unique patients with missed monitoring events across the full study period.
We conducted t-tests to investigate whether a significant change in monitoring rate occurred between February-May 2020.

| Variability of change across GP practices
The indicator percentages for overall DMARD missed monitoring rate were summarized as deciles across all GP practices and presented as a decile chart.

| Factors associated with change in monitoring rate
Cochran's Q heterogeneity tests were conducted within each demographic/clinical subgroup to check for associations between patient factors and changes in monitoring rate.

| Software and reproducibility
Data management and analysis was performed using the OpenSA-FELY software libraries and Python, both implemented using Python 3.8.Inferential statistics were performed using R.This analysis was delivered using federated analysis, which involves carrying out patient-level analysis in multiple secure datasets, then later combining them: codelists and code for data management and analysis were specified once using the OpenSAFELY tools; then securely transmitted to the OpenSAFELY-TPP platform for execution against local patient data within TPP's secure environment.Summary results were reviewed for disclosiveness and released for final outputs.All code for the OpenSAFELY platform for data management, analysis and secure code execution is shared for review and re-use under open licences at github.com/OpenSAFELY, or for this study at github.com/opensafely/Shared-Care-Monitoring. importance of large datasets for research; we have presented at several public engagement events.We also work closely with appropriate medical research charities.

| RESULTS
Demographic, regional and clinical characteristics of patients being regularly prescribed azathioprine, leflunomide or methotrexate are reported in Table 3 according to the final month of the study period (July 2022).Of these 94 611 patients, 60.9% were aged ≥60 years, 60.9% were female and 91.1% were White.

| Trend in DMARD monitoring rates
Throughout the study, patients missed monitoring at an average rate of 31.1% (977 354 of 3 146 849 prescribing events).The ratio of missed monitoring events to unique patients was 8.3.Between the baseline and lockdown periods, there was a significant rise in missed monitoring rates across the population, increasing from 28.4 to 40.8% (+12.4 percentage points, P < .001).However, between the lockdown and recovery periods, rates decreased to 28.1% (À12.7 percentage points).For a detailed breakdown of results by DMARD and monitoring test, see Table S3.For breakdowns by demographic and clinical subgroups (including statistical analysis of change over time), see Table S4.

| Medication type
The percentage of patients identified as missing monitoring was greatest for leflunomide (65.0%; 125 850 of 193 665 patients, ratio 13.5 events per unique patient) and lowest for methotrexate (24.6%; 517 770 of 2 106 050 patients, ratio 7.0 events per unique patient).
All DMARDs exhibited an increase in missed monitoring rates immediately following the onset of lockdown (April-June 2020; Figure 1).These rates showed considerable recovery through July-August 2020.The increase in missed monitoring rates between the baseline and lockdown period was similar for methotrexate (+11.7 percentage points) and azathioprine (+12.3 percentage points), whereas a noticeably greater change was seen for leflunomide (+20.All monitoring tests exhibited an increase in missed monitoring immediately following the onset of lockdown (April-June 2020; Figure 2).The increase in missed monitoring rates between the baseline and lockdown period was similar for FBC (+12.61 percentage points), LFT (+12.75 percentage points) and U&E (+12.66 percentage points), whereas a greater change was seen for BP (+24.46 percentage points).All rates showed considerable recovery through July-August 2020.By the recovery period most tests had returned to similar rates as at the baseline period: FBC (23

| Variability of change across GP practices
At the baseline period (February 2020), the median missed monitoring rate across GP practices was 25.0%.There was a large interdecile range, from 8.3% in the 1st decile to 63.3% in the 9th decile (Figure 3).
At the lockdown period (May 2020), the median missed monitoring rate was 40.6% (+15.6 percentage points from baseline).The interdecile range widened slightly with the 9th decile increasing by +12.6 percentage points from baseline, compared to +7.6 percentage points for the 1st decile.
At the recovery period (July 2022), the interdecile range had narrowed compared to baseline.

| Factors associated with change in monitoring rate
Mean monitoring rates at baseline, lockdown and recovery periods are presented for each group of patient characteristics in Table S4.
All groups showed significant increase in missed monitoring rates between the baseline and lockdown period.This increase differed for:

| Strengths and limitations
The scale and comprehensiveness of data in the OpenSAFELY platform is greater than any other method for accessing GP data.Historically, attempts to audit monitoring compliance in primary care would have typically relied on manual audit within a practice.By contrast, OpenSAFELY allows efficient analyses using data, which is broadly representative of the English population, providing a national overview of monitoring compliance. 22A second strength is the study's  Secondly, in some instances a clinician or the patient may have decided to stop the medication since the prescription issue date (especially during the pandemic months), which may invalidate the requirement for monitoring to be completed in the epected period.
Both of these limitations may lead to some patients falsely appearing overdue for monitoring.Conversely, some tests may have been counted that were carried out for reasons unrelated to DMARD monitoring.We were also unable to capture prescriptions issued by secondary care, these data have historically been challenging to obtain. 23Recently, NHS Digital have made prescription data available for some hospitals-we will aim to incorporate these in future research. 24Lastly, we report data for 3 DMARDs, therefore findings may not be generalisable for other high-risk drugs.

| Comparison with existing literature
This study is the first to comprehensively evaluate DMARD monitoring at a national scale during the pandemic.Our findings are consistent with previous OpenSAFELY research, which identified deterioration in monitoring across a range of other medications and monitoring tests following the March 2020 lockdown. 13,14For example, Fisher et al. 14 25 Although, it should be noted that many patient-related factors have been reported as having an inconsistent impact, suggesting that determinants of adherence are multifaceted and context-specific. 26,27

| Policy implications and interpretation
This research highlights that DMARD monitoring is often not adhered to.However, wide variance in monitoring rates between practices indicates that some practices have been able to implement successful strategies.This presents opportunities for sharing best practice, which NHS commissioners may be able to facilitate at scale.
The deterioration that this study observed in DMARD monitoring during lockdown, and that other studies have observed in routine services, 13 offers insight for health policymakers regarding the effects of the pandemic and its management strategies.A combination of factors probably contributed to this deterioration including: diversion of primary care resources towards managing COVID-19, guidance from national bodies to enable reduced monitoring frequency, supply problems with blood sample bottles and patients choosing to avoid healthcare for fear of exposure to COVID-19.This understanding will help inform assessment of the risks and benefits of lockdown measures in future pandemics, and how the impact on groups that were more severely affected could be mitigated, through strategies that preserve patient engagement and accessibility.
The disproportionately large deterioration observed in BP monitoring may suggest that clinicians deemed it less critical than blood tests, reflecting appropriate prioritization of monitoring.Regional differences in deterioration may suggest that regions prioritized DMARD monitoring differently during the pandemic.Factors underpinning regional differences may include the availability of local workforce and proximity of facilities to support the pandemic response; both have been suggested as relevant to service resilience during the pandemic. 28Differences in resilience may also be related to known variability in the robustness of local shared care processes and documentation, 29 and regional differences which have been observed in the time between first rheumatoid arthritis review to DMARD initiation. 30e overall differences in monitoring rates observed between various groups highlight ongoing inequalities.
Primary care records managed by TPP are available in OpenSAFELY, a data analytics platform created by our team on behalf of NHS England to address urgent COVID-19 research questions (https:// opensafely.org).OpenSAFELY provides a secure software interface allowing analysis of pseudonymized primary care patient records from England in near real-time within the EHR vendor's secure data centre, avoiding the need for large volumes of potentially disclosive pseudonymized patient data to be transferred off-site.This, in addition to other technical and organizational controls, minimizes risk of re-identification.Similarly pseudonymized datasets from other data providers are securely provided to the EHR vendor and linked to the primary care data.The dataset analysed within OpenSAFELY is based on people currently registered with GP practices using TPP Syst-mOne software.It includes pseudonymized data such as coded diagnoses, medications and physiological parameters.No free text data are included.Further details can be found under information governance and ethics.
age-group categories (heterogeneity test: Cochran's Q = 37.526, P < .001;Figure4), ranging from +8.3 percentage points in the 18-29 years age-group to +13.7 percentage points in the 70-79 years age-group; sex categories, (Q = 4.5, P = .034;Figure5), ranging from +11.9 percentage points in males to +12.8 percentage points in females; and regions (Q = 78.869,P < .001;Figure6), ranging from +9.2 percentage points in the North East to +17.0 percentage points in the North West.Throughout the study, substantial overall differences were apparent in missed monitoring rates within the groups of age, sex and region (Figures4-6) and also ethnicity, learning disabilities, serious mental illness, deprivation and rural-urban score (FiguresS1-S5).No T A B L E 3 Cohort description for patients who were included in at least 1 disease-modifying antirheumatic drug (DMARD) denominator at the end of the study period (July 2022).
subgroups were rounded to the nearest 5, so percentages were calculated as a proportion of the subgroup total.F I G U R E 1 Proportion of patients overdue monitoring between November 2019 and July 2022, broken down by shared care medication.The baseline period before lockdown is shown as an orange dashed vertical line.The monitoring window, measured as 3 months from the onset of the March 2020 COVID-19 lockdown, is shown as a green dashed vertical line.reproducibility and transparency.The complete set of code for the platform and all data curation/analysis from raw data to final output, is shared publicly on GitHub for peer review and efficient reuse.We note some limitations.Firstly, our data only include results from tests carried out in primary care, or carried out in secondary care where results are returned to GPs as structured data.Test results performed outside GP practices are omitted from our data unless they are communicated to the GP practice and coded as structured data by GP practice staff.In most circumstances, where GP practices prescribe the medications, they will usually also perform the monitoring.

F I G U R E 2
Proportions of patients overdue disease-modifying antirheumatic drug (DMARD) monitoring between November 2019 and July 2022, broken down by monitoring test type.The baseline period before lockdown is shown as an orange dashed vertical line.The monitoring window, measured as 3 months from the onset of the March 2020 COVID-19 lockdown, is shown as a green dashed vertical line.F I G U R E 3 Practice level decile charts for proportions of patients overdue disease-modifying antirheumatic drug (DMARD) monitoring between November 2019 and July 2022.

F
I G U R E 4 Proportions of patients overdue disease-modifying antirheumatic drug (DMARD) monitoring between November 2019 and July 2022, broken down by age-band.The baseline period before lockdown is shown as an orange dashed vertical line.The monitoring window, measured as 3 months from the onset of the March 2020 COVID-19 lockdown, is shown as a green dashed vertical line.F I G U R E 5 Proportions of patients overdue disease-modifying antirheumatic drug (DMARD) monitoring between November 2019 and July 2022, broken down by sex.The baseline period before lockdown is shown as an orange dashed vertical line.The monitoring window, measured as 3 months from the onset of the March 2020 COVID-19 lockdown, is shown as a green dashed vertical line.

F I G U R E 6
Proportions of patients overdue disease-modifying antirheumatic drug (DMARD) monitoring between November 2019 and July 2022, broken down by region.The baseline period before lockdown is shown as an orange dashed vertical line.The monitoring window, measured as 3 months from the onset of the March 2020 COVID-19 lockdown, is shown as a green dashed vertical line.

4. 5 |
Future researchTo inform policy interventions, further research is needed to explore the causes of poorer performance amongst specific practices and patient groups.This could be done through qualitative studies surveying/interviewing those with poorer monitoring rates to identify the individual traits, attitudes, behaviours and circumstances underpinning their poorer monitoring adherence.More broadly, the pandemic offers an unprecedented opportunity of a natural experiment, since changes to clinical practice (such as the frequency of safety monitoring) have occurred that would have been unethical in normal circumstances.This allows researchers to test whether certain principles of clinical practice could be optimized.Recent research has suggested that the clinical impact of missing DMARD monitoring is minimal, with 1 study suggesting that reducing the frequency of methotrexate monitoring from 3 to 6 monthly did not increase abnormal test results nor cause harm to patients.35However, such analysis is easily confounded by patient-specific factors related to clinical risk, which influence the likelihood of monitoring occurring.For example, patients at lower risk of negative outcomes may be less motivated to attend monitoring, or less proactively followed up by clinicians.By contrast, patients at higher risk may be more likely to have accessibility issues inhibiting monitoring, or more reluctant to attend monitoring due to fear of contracting COVID-19 with their clinical vulnerability.Consequently, patients who miss monitoring may have an intrinsically different likelihood of negative outcomes.Studies could attempt to evaluate the clinical impact of reduced monitoring frequency within patient groups stratified by clinical risk factors, and break results down for various medications/tests.This may reveal whether certain medications or patient groups benefit from monitoring to different extents, informing more tailored costeffective strategies for who should receive which monitoring tests, and how often.

4. 6 |
Conclusion A transient deterioration in DMARD safety monitoring occurred across all groups during the COVID-19 pandemic, acutely around the time of lockdown measures with performance mostly recovering rapidly as lockdown measures were eased.Long-term differences in safety monitoring rates exist between several sociodemographic and clinical subgroups, and the pandemic impacted monitoring in these populations to varying extents.Health commissioners should identify causes of poorer monitoring in specific groups and consider implementing supportive measures.Monitoring rates varied substantially between practices, suggesting opportunities exist to improve service consistency by sharing good practice and adopting standardized shared care processes.These analyses demonstrate the capability of the OpenSAFELY platform as an effective tool to provide actionable insights on health service provision and inequalities.AUTHOR CONTRIBUTIONS Conceptualization: Andrew D. Brown, Ruth E. Costello and Brian MacKenna.Data curation: Andrew D. Brown, Louis Fisher, Lisa E.M. Hopcroft, Christine Cunningham, Seb C.J. Bacon, Amir Mehrkar, George Hickman, Chris Bates, Jonathan Cockburn, John Parry, Frank Hester and Sam Harper.Formal analysis: Andrew D. Brown, Louis Fisher, Milan Wiedemann, William J. Hulme and Helen J. Curtis.Funding acquisition: Ben Goldacre.Investigation: Andrew D. Brown, Louis Fisher, Milan Wiedemann, William J. Hulme and Helen J. Curtis.
reported 19.6% of patients on methotrexate missed LFT on average between January to March 2020, whilst in the current study we reported that 22.1% of patients on methotrexate missed LFT, FBC or U&E across the same period.This small difference is to be expected as our Fisher et al. study used indicators specific to single ethnicity, cognitive function and socioeconomic status) can influence medication-taking adherence.