Abstract
Objective Online patient-reported outcome measures (PROMs) enable remote collection of perceptions of health status, function, and well-being. We aimed to explore patterns of PROM completion in patients with early inflammatory arthritis (EIA) recruited to the National Early Inflammatory Arthritis Audit (NEIAA).
Methods NEIAA is an observational cohort study design; we included adults from this cohort with a new diagnosis of EIA from May 2018 to March 2020. The primary outcome was PROM completion at baseline, 3 months, and 12 months. Mixed effects logistic regression and spatial regression models were used to identify associations between demographics (age, gender, ethnicity, deprivation, smoking, and comorbidity), clinical commissioning groups, and PROM completion.
Results Eleven thousand nine hundred eighty-six patients with EIA were included, of whom 5331 (44.5%) completed at least 1 PROM. Patients from ethnic minority backgrounds were less likely to return a PROM (adjusted odds ratio [aOR] 0.57, 95% CI 0.48-0.66). Greater deprivation (aOR 0.73, 95% CI 0.64-0.83), male gender (aOR 0.86, 95% CI 0.78-0.94), higher comorbidity burden (aOR 0.95, 95% CI 0.91-0.99), and current smoker status (aOR 0.73, 95% CI 0.64-0.82) also reduced odds of PROM completion. Spatial analysis identified 2 regions with high (North of England) and low (Southeast of England) PROM completion.
Conclusion We define key patient characteristics (including ethnicity) that influence PROM engagement using a national clinical audit. We observed an association between locality and PROM completion, with varying response rates across regions of England. Completion rates could benefit from targeted education for these groups.
Patient-reported outcome measures (PROMs) are standardized, validated questionnaires, completed by patients to identify perceptions of their health status, function, and well-being.1 These measures were initially designed for use in research and clinical trials but have been adopted across a wide range of healthcare settings. Increasingly, PROMs are being used in national clinical audits, as means of monitoring the performance of healthcare providers and informing individualized patient care.
The National Early Inflammatory Arthritis Audit (NEIAA), commissioned by the Healthcare Quality Improvement Partnership (HQIP) and run by the British Society of Rheumatology is one such audit. NEIAA collects information on all patients with suspected inflammatory arthritis over the age of 16 seen for the first time in rheumatology departments in England and Wales. As part of the outcome measurements, NEIAA collects 4 key PROMs, using a combination of clinician-transcribed data and patient-entered online data. These comprise the Musculoskeletal Health Questionnaire (MSK-HQ),2 Health Assessment Questionnaire (HAQ), Patient Health Questionnaire-4 Anxiety and Depression Scale (PHQ4-ADS), and the Work Productivity and Activity Index (WPAI). These PROMs provide information about disease impact, functional impairment, as well as mental health impact and effects on work.
The completion of a PROM assessment requires patient cooperation. As such, PROM completion is susceptible to a higher rate of missing data compared with information collected by clinicians or researchers in clinical trials and audits. Fundamental to the utility and reliability of PROMs is that they are representative of the population they are sampled from.3 Our group has previously shown that socioeconomic position (or deprivation), defined as the social standing of an individual within a population, predicts clinical outcomes.4
Previous studies have demonstrated that social deprivation, female gender, and increasing age are predictors of not returning PROMs in the context of orthopedic procedures.3,5 However, there is little in the way of evidence looking at PROM engagement in inflammatory arthritis, which may differ from that seen in orthopedics because of its chronicity and lack of operative interventions. PROMs are also increasingly integrated with clinical decision making in rheumatology to enhance disease control.6 Ensuring equal access to PROMs across patients with varying demographics and geographic locations is currently of particular relevance. The coronavirus disease 2019 (COVID-19) pandemic has strengthened the desire to collect PROMs as a supplement to remote consultation; in parallel, the pandemic has renewed focus on tackling health inequalities.
This study explores the predictors of PROM completion in NEIAA. The specific objectives are to (1) assess demographic associations, including social deprivation, with PROM completion in patients with EIA; and (2) explore regional patterns in PROM completion in patients with EIA, examining for differences in completion rates.
METHODS
The HQIP commissioned the British Society of Rheumatology to deliver NEIAA with the aim to improve healthcare quality and outcomes of adult patients with a new diagnosis of EIA in England and Wales.
Full details of the data collection can be found in the project’s annual report,7 with summarized details below.
NEIAA is a prospective cohort study design. It has data on adults (aged > 16 years) with suspected EIA referred to secondary care rheumatology services in England and Wales. Patients with a confirmed EIA diagnosis are eligible for further follow-up. This report did not include data from Wales owing to limitations of data availability.
Patient and public involvement. A patient panel has been central to the design and delivery of NEIAA since its outset. The PROM collection tool was developed in conjunction with the patient panel, the members of whom expressed a desire for us to explore reasons behind PROM noncompletion. This informed the need for this analysis.
Patient population. We included adults with a confirmed new diagnosis of EIA recruited to NEIAA in England. Recruitment commenced May 7, 2018. The data cut-off date for this study was March 31, 2020.
Predictor variables. Rheumatology clinicians collect the following data for all patients in NEIAA (we only included baseline information in this report):
• Demographics, including age, gender (as defined in the clinical record at time of diagnosis), smoking status (current smoker, ex-smoker, never smoker), ethnicity, work status, and deprivation.
• Service quality measures, including time to referral and treatment, and referral route.
• Clinical measures, including comorbidities, symptom duration, inflammatory markers (C-reactive protein [CRP]/erythrocyte sedimentation rate [ESR]), disease severity, and treatments.
Ethnicity was coded by the treating clinician according to the 5 main ethnic groups in accordance with the UK Office for National Statistics categorization system8 into White, Black (Black British/Caribbean/African), Asian (Asian/Asian British), mixed (White and Black Caribbean/White, Black African/White, Asian, or any other mixed or multiple ethnic backgrounds), and other (Arab or any other ethnic group). To draw comparisons between groups, ethnicity was converted into a binary variable of Black, Asian, and Minority Ethnic (BAME) or White. There have been different opinions about the use of the term BAME in grouping ethnic minority patients. We acknowledge that this term is not ideal as its emphasis on the Black and Asian ethnic groups underestimates other ethnic groups. However, we decided to group ethnic minority into binary groups for statistical reasons only, as ethnicity was used as a covariate and not the primary outcome in this study.
Work status was categorized as those working > 20 hours/week or not. Comorbidity burden was scored using the Rheumatic Disease Comorbidity Index (RDCI). The RDCI is a validated tool that gives a weighted cumulative comorbidity burden score based on history of chronic lung disease, cardiovascular disease, hypertension, diabetes mellitus, cancer, peptic ulcer disease, and depression. RDCI is scored from 0 to 9, where higher scores indicate higher comorbidities burden.9 Disease severity was assessed using tender joint count, swollen joint count, CRP, or ESR, and a visual analog scale to calculate the disease activity score in 28 joints (DAS28).
Deprivation was collected using the English Index of Multiple Deprivation (IMD), which uses 7 domains to give an area level composite score of socioeconomic position. The following 7 domains were combined using specific weights to produce the overall IMD: (1) income (22.5%); (2) employment (22.5%); (3) education, skills, and training (13.5%); (4) health and disability (13.5%); (5) crime (9.3%); (6) barriers to housing and services (9.3%); and (7) living environment deprivation (9.3%). The score of each area is used to generate an IMD rank.10 The IMD quintile was obtained for each patient by linking to their postal code. We used an IMD to measure deprivation as NEIAA does not collect information on individual income.
Patients’ general practitioner (GP) postal code was linked to Clinical Commissioning Groups (CCGs) to enable geospatial mapping. At the time of data collection, healthcare delivery in England was organized by CCG, who funded healthcare services within their boundaries. These funding arrangements make CCG a rational area level measure for assessing differences between parts of England. In addition, CCGs have well defined geographic boundaries, making them suitable for geospatial analyses.11
Outcome. The primary outcome was completing 1 of 4 PROMs collected in NEIAA. The PROMs collected at baseline, 3 months, and 12 months follow-up were the following:
1. MSK-HQ, a 15-item questionnaire that evaluates how musculoskeletal symptoms affect day-to-day life. Scores range from 0 to 56, with higher scores indicating better musculoskeletal health.12
2. HAQ-II, a 10-item assessment of disability, summed to provide a total score ranging from 0 to 3. A higher score indicates worse function.13
3. PHQ4-ADS, a 4-item measure to assess low mood or anxiety. A combined score is calculated from 2 questionnaires. Each questionnaire has 2 items, with a score ranging from 0 to 6. The lower combined score indicates a lower likelihood of having low mood or anxiety.14
4. WPAI, a 6-item assessment of work status and impact. WPAI overall impairment is expressed as percentages (from low to high impact) and incorporates absenteeism (percentage of work hours missed in relation to total hours worked) and presenteeism (extent by which patients’ work productivity was affected by their health as a percentage).15
Patients can return PROMs information through 1 of 3 mechanisms based on their preference: online data entry by the patient into the patient audit website; paper form completion by the patient, which is subsequently uploaded to the audit website; or online entry by the healthcare provider into the staff audit website. Patients who agreed to online PROM completion were sent a link to the online portal to record their PROMs. If no data are completed, a follow-up reminder email is sent 2 weeks later.
Since few patients completed a PROM at all 3 timepoints, we defined PROM completion in our primary analysis as the following: any patient who completed ≥ 1 PROM was categorized as having “completed PROMs.” Patients who completed no PROMs were categorized as having “not completed PROMS.” Questionnaires that were started but not finished were considered “not complete.” We also calculated the proportion of available PROMs that were completed for each patient.
Statistical analysis. Characteristics of patients by PROM completion were tabulated and tested for statistically significant imbalance using chi-square, Mann-Whitney U, or t tests, as appropriate.
A mixed effects logistic regression model was constructed to examine associations between patient factors, geographic location, and PROM completion at any timepoint. Due to limited data, PROM completion was used as a binary outcome (at least 1 PROM completed vs never completed). The primary predictor variable was IMD, which was converted to a binary variable, with the most deprived quintile compared to the rest. Baseline data, including age (by decade), gender, ethnicity, smoking status, and comorbidity burden, were included as covariates. There were few missing baseline data, and a complete case approach was performed for the main analysis.
A random effect was included in the model to account for clustering for patients within centers, consistent with methodology we have used previously.4
To explore area level differences in PROM completion, mean PROM completion by CCG was mapped across England using the grmap Stata package (StataCorp). Shapefiles detailing coordinates of English CCG boundaries from 2018 were downloaded from the Office of National Statistics website and merged to the NEIAA dataset via participants’ GP postal code.16
In line with the Office of National Statistics small numbers reporting guidance, CCGs with ≤ 5 patients recruited to NEIAA were excluded from mapping to ensure patient confidentiality was maintained.4,17
A hot spot spatial analysis, similar to what was used in NEIAA previously,4 was performed to identify local correlations in PROM completion, to assess if geographic area associated with PROM completion. This was performed using the Getis-Ord statistic.18 The statistic identifies if a CCG and its neighbors form a spatial cluster or outlier compared to the overall sample by calculating the ratio of local PROM completion to total PROM completion for each CCG. This is used to generate a z statistic for each area, with values ± 1.96 and ± 2.58 corresponding to 5% and 1% significance levels, respectively.19 Hot spots were areas with high PROM completion rates, whereas cold spots were areas with low completion rates.
Secondary analyses. To account for missing covariate data, the mixed effects regression model was rerun following multiple imputation. Demographic information was mostly complete; however, imputation was used for missing data in deprivation (IMD), comorbidity (RDCI), smoking, and ethnicity (Supplementary Material S1, available with the online version of this article). A multiple imputation model using chained equations (20 imputed datasets) was used for adjusted analyses. Linear and logistic regression were performed to impute the missing variables. The following variables were used for the imputation: age, gender, and treating center and region.
To explore PROM completion patterns in different ethnicities, the regression model was rerun with ethnicity as a categorical variable rather than as a binary one.
The regression model was rerun limited to patients who had agreed to online PROM completion (as opposed to paper-based completion and then clinician transcription to the online portal). We had intended to look at response rate according to method of completion based on feedback from our patient expert panel; however, most questionnaires were returned electronically. The other groups were too small to look at individually.
Ethics and consent. NEIAA successfully obtained a Secretary of State for Health Section 2.51 approval, allowing the collection of patient identifiable data without consent. This paper is a Healthcare Quality Improvement Partnership–approved output of the project (Confidentiality Advisory Group [CAG] reference: 18/CAG/0063 [resubmission of 18/CAG/0017]).
RESULTS
Patient characteristics. From its inception until March 31, 2020, a total of 34,856 patients with suspected EIA were recruited to NEIAA. Of these, 11,986 were diagnosed with EIA across England, recruited from 205 CCGs. Baseline characteristics are presented in the Table. The mean age at recruitment was 56.8 (SD 16.2) years. Thirty-eight percent of the cohort were male and 13.1% were BAME. Rheumatoid arthritis was the predominant EIA diagnosis. The cohort was evenly distributed across the IMD quintiles.
Baseline demographics of the NEIAA cohort, by PROM completion.
PROM completion. At least 1 PROM was completed by 5331 (44.5%) patients. Of these, 3846 (72.1%) had completed half of all the available PROMs, whereas 412 (7.7%) had completed all available PROMs. In total, 6587 patients had at least 12 months of follow-up and thus were eligible to complete PROMs at all 3 timepoints. Of these patients, PROMs were completed at just 1 timepoint by 2313 patients, at 2 timepoints by 458 patients, and at all 3 timepoints by 195 patients (Supplementary Material S1, available with the online version of this article).
There were differences in completion of individual PROMs, with a greater percentage of completion seen with MSK-HQ, followed by PHQ4-ADS and HAQ, which followed the sequences that the PROMs are presented in the online platform. Only 2556 patients (21.3%) completed WPAI. Of the 5700 patients in paid work > 20 hours a week, approximately two-thirds (3487 patients) failed to complete a WPAI PROM.
Patients who did not complete a PROM had a mean age of 56.2 (SD 16.4) years compared to 57.5 (SD 15.9) years for those who did complete a PROM, which was a significant difference (P < 0.001). PROM noncompleters were more likely to be BAME (16% vs 9%, P < 0.001), current smokers (21% vs 17%, P < 0.001), and be in the most deprived IMD quintile (22% vs 18%, P < 0.001) than PROM completers. There was no difference in PROM completion between genders, patients with comorbidity, or patients in paid work (Table).
Association between patient factors and PROM completion. IMD predicted PROM completion in the logistic regression model (Figure 1). The odds of completing at least 1 PROM were reduced for people in the most deprived IMD quintile (adjusted OR [aOR] 0.73, 95% CI 0.64-0.83). Male gender (aOR 0.86, 95% CI 0.78-0.94), BAME (aOR 0.57, 95% CI 0.48-0.66), comorbidity burden (aOR 0.95, 95% CI 0.91-0.99), and current smoker status (aOR 0.73, 95% CI 0.64-0.82) also reduced the odds of PROM completion. The greatest effect was seen with BAME. Advancing age (by decade) was associated with an increase in the odds of PROM completion (aOR 1.04, 95% CI 1.01-1.07).
Mixed effects adjusted multivariable logistic regression model showing patient level associations with PROM completion. Ethnicity had the strongest association. BAME: Black, Asian, and Minority Ethnic; IMD: Index of Multiple Deprivation; PROM: patient-reported outcome measure; RDCI: Rheumatic Diseases Comorbidity Index.
Area level patterns of PROM completion. Of the 205 CCGs across England, 195 were included in the analysis (10 CCGs had ≤ 5 patients recruited to NEIAA, so they were excluded). The proportion of PROM completion varied across these CCGs with a range of 0-95%. Figure 2 depicts a map of mean PROM completion by CCG. Spatial analysis identified 2 autocorrelated regions, with a hot spot of PROM completion in the North, and a cold-spot in the Southeast (Figure 3). The z statistics for each CCG are mapped in Figure 3 (also tabulated in the Supplementary Material, available with the online version of this article). The PROM spatial analysis map (Figure 3) is striking and contrary to geographic patterns for IMD, where greater deprivation is seen in the North of England and less deprivation in seen in the Southeast (see Supplementary Material).
Map showing mean PROM completion by CCG. Darker colors indicate higher PROM completion; lighter colors indicate lower PROM completion. White indicates no data available. CCG: Clinical Commissioning Group; PROM: patient-reported outcome measure.
Hot spot spatial analysis to identify local correlations in PROM completion by CCGs. Two autocorrelated geographic regions with a hot spot of PROM completion in the North, and a cold spot in the Southeast. Blue represents a cluster of CCG with lower PROM completion, with orange and red representing a cluster of higher PROM completion. CCG: Clinical Commissioning Group; PROM: patient-reported outcome measure.
Secondary analysis. The associations with PROM completion were consistent when rerun with imputed data. They were preserved when the analysis was limited to patients who agreed to online PROM completion, with the point estimates for IMD strengthened but with a widened 95% CI. See the Supplementary Material (available with the online version of this article) for further details. To further investigate the effect of ethnicity on PROM completion, we reran our regression model with ethnicity as a categorical variable. This identified that Black British/African/Caribbean (aOR 0.59, 95% CI 0.43-0.82) and Asian/Asian British (aOR 0.52, 95% CI 0.42-0.64) both associated with lower PROM completion compared to White ethnicity.
DISCUSSION
This is the first study, to our knowledge, to examine the association between demographics and PROM engagement in patients with EIA and in the context of a national clinical audit. Patient factors, including age, gender, ethnicity, deprivation, comorbidity, and smoking status predict PROM completion, with the greatest effect seen with ethnicity. The spatial analysis identified differences in PROM engagement across England, with higher completion in the North and an area of lower completion in the Southeast. This suggests a locally acting organizational influence on PROM completion that may be an additional driving force alongside sociodemographic factors.
The finding that noncompletion of PROMs was associated with Black, Asian, and minority ethnic backgrounds is salient. We have previously shown that people of non-White backgrounds have poorer clinical outcomes.20-22 If we are identifying fewer PROMs on ethnic minorities, we could be minimizing the results for patients from minority ethnic backgrounds and overestimating clinical outcomes for the population as a whole. The relationship between ethnicity and PROM return is likely to be complex and influenced by a number of social, cultural, and healthcare organizational factors.23 Previous work has linked ethnic disparities between patient and clinician to lower levels of patient-centered care24 and mutual understanding.25 An important consideration is that PROM forms in NEIAA are provided only in English owing to funding constraints, potentially excluding those patients for whom English is not their spoken or written language.5 It is also important to note that BAME is an umbrella term encompassing a diverse subgroup of patients and care must be taken when drawing conclusions about this group as one. We explored this with the secondary analysis that identified Black and Asian ethnicities as predictors of not returning PROMs.
The association between IMD and PROM completion is a key finding. As deprivation increased, the likelihood of PROM completion fell. Education is likely to be a driving factor,26 as a degree of literacy is required to complete PROMs even for patients who speak and read English as their first language. Education forms 1 of 7 components of the IMD score and receives greater weighting than 3 of the other domains. This relationship may more broadly reflect previous findings that patients from more deprived areas hold their doctor responsible for monitoring and managing their health, and thus feel less empowered or motivated to complete PROMs.27 The development of more collaborative clinical relationships, where patients take a greater degree of ownership for their condition, may encourage more complete PROM returns.
Our results suggest that there are underrepresented groups with the potential to bias in favor of those likely to experience better outcomes. It was shown previously that patients with better outcomes may be overrepresented in cohorts, which may lead to selection bias in registry data.28
Spatial analysis identified areas across England with high and low levels of PROM completion. The high hot spot area was found in the North, whereas the low cold spot was in the Southeast. The area of high PROM engagement could reflect shared learning between clinical departments, propagating best practice across local networks. The regional clustering of PROM completion is in stark contrast to regional clusters of deprivation, with greater deprivation in the North and less in the South (see Supplementary Material, available with the online version of this article). This suggests that locally acting factors may have substantial influence on PROM engagement. Areas of high PROM engagement may also reflect local efforts to engage patients and promote the value of PROMs data for informing clinical management. It implies that local/regional education and attitudes to PROMs could overcome demographic barriers to patient engagement and improve local uptake.
The sequence of delivering the PROM questionnaire likely affected our results, as the first PROM delivered had the highest completion rates. The length of the questionnaire did not appear to be a factor, with the MSK-HQ being the longest and most completed item. This could be explored in more detail in a follow-up study, where the order of questionnaires is randomized.
The major strength of this study is that it was conducted on a large dataset, with extensive patient level information. NEIAA recruited all patients referred to secondary care across England and Wales. We recognize that this study has several limitations. Limited knowledge of English literacy or education level can affect PROMs completion. Unfortunately, we only had area-level education data, rather than patient-level.
The NEIAA database is reliant on the quality of data uploaded. Data entry relies upon clinician engagement, which can lead to variations in the details recorded and coding and may increase susceptibility to missing data. To improve PROM completion, NEIAA relied on communication with patients by email. A proportion of patients do not engage with email correspondence, and this may have introduced bias. We have attempted to account for this in our secondary analyses. Last, NEIAA participants residing in Wales were excluded from our analyses because IMD rankings between England and Wales are not directly comparable. This affects generalizability to the Welsh population.
We need to acknowledge that we are presenting a population-level analysis, with some imprecision in the data. For example, we have examined patient behaviors (returning PROMs) using geospatial mapping attributable to their GP practice, which could be some distance from their residence and may be particularly relevant to rural areas in England.
We used a surrogate marker (IMD) to measure deprivation, as NEIAA does not collect information on individual income.
As a result of limited data, we defined PROM completion by completing at least 1 PROM at any timepoint; this may have led to attrition bias29 in our study. Loss of follow-up is inevitable in most cohort studies30; in our cohort, the number of individuals who were eligible to complete a PROM at all 3 timepoints was small.
Another limitation is that we did not account for variations in ethnicity distributions (clustering) in our spatial analyses. Last, the study period was mostly (bar 1 month) before any significant effects of the COVID-19 pandemic were felt in the United Kingdom. Given the effects of the pandemic on outpatient care delivery, it is likely that PROM completion behaviors will have changed during this period.
This study defines key patient characteristics that influence PROM engagement in a national clinical audit and identifies a novel association between locality and PROM completion. PROMs offer the potential to widely monitor the outcomes of patients in a way that is important to them and improve services accordingly. Integral to the utility of PROMs in NEIAA is that they are representative of the population with EIA, and it is imperative that nonresponse to PROMs among groups of patients does not bias outcome findings. Our results raise the possibility that there are underrepresented groups with the potential to bias in favor of those likely to experience better outcomes. Further understanding of the barriers and enablers (eg, patients’ engagement, overall eudcation level of the patient) to completing PROMs within these groups is required. This could in turn facilitate targeted patient education to specific groups to improve PROM completion rates.
Footnotes
NEIAA is funded by the Healthcare Quality Improvement Partnership (HQIP). MY’s salary is funded by grants from the British Society for Rheumatology and Versus Arthritis.
M. Yates and K. Bechman contributed equally to this work.
JG has received honoraria from AbbVie, Celgene, Chugai, Galapagos, Gilead, Janssen, Lilly, Pfizer, Roche, and UCB. MY and BC have received honoraria from AbbVie. The remaining authors declare no conflicts of interest relevant to this article.
- Accepted for publication April 27, 2023.
- Copyright © 2023 by the Journal of Rheumatology
REFERENCES
DATA AVAILABILITY
Data available upon application to HQIP.
