Discordance between patient and physician global assessments amongst South Africans with rheumatoid arthritis

Introduction

The management of rheumatoid arthritis (RA) has evolved with the adoption of the “Treat to Target” strategy, aiming for remission or low disease activity (1–3). Composite disease activity scores, such as the Simplified Disease Activity Index (SDAI), and Clinical Disease Activity Index (CDAI), are used to assess RA disease activity and to classify the disease into high, moderate or low disease activity states to inform treatment decisions which are shared between patient and physician (4). These scores combine the 28-joint tender and swollen joint counts (TJC and SJC), the patient global assessment (PGA) “How is your arthritis in the last week?” scoring between 0 (very good) and 10 (very poor), and physician global assessment (PhGA) scoring between 0 (very good) and 10 (very poor).

The PGA is an important widely used patient-reported outcome measure (PROM) and provides valuable insights into the patient's perspective on their disease activity. However, it is well recognized that PGA can be influenced by factors including socio-demographic characteristics, fatigue, sleep disturbances, mental health status, and pain (5–8).

Differences in the patient's perceptions of disease activity and their physicians’ assessment can be measured by comparing the PGA and the PhGA. The presence of patient-physician discordance in RA has been observed in clinical trials and clinical practice in up to half of patients, with patients generally reporting higher (worse) scores than physicians (9–11).

Understanding patient-physician discordance in RA is important. This discordance reflects a multidimensional misalignment in disease perception. While physicians tend to emphasize objective measures of inflammation, including the SJC and serum acute phase reactants, patients integrate the full experiential burden of their disease. Discordance may therefore arise from pain sensitization, structural joint damage, mood disorders, fatigue, or residual subclinical inflammation (9–16).

If unaddressed, this misalignment can lead to treatment decisions that are not based on shared decision-making. Indeed, discordance has been shown to negatively affect medical care with poor treatment adherence, higher levels of dissatisfaction, and an increased risk of joint damage, and is also associated with impaired health-related quality of life (HRQoL) and decreased work productivity (17–19). Interestingly, discordance has recently been shown to be associated with a higher grade of musculoskeletal ultrasound synovitis, suggesting that disease activity is underrated by one of the assessors (20).

Variation in PGA between countries is well described (7). There is a paucity of research specifically examining patient-physician discordance in South Africa (SA). Factors including language barriers, cultural beliefs, socioeconomic disparities, and healthcare access can influence communication and understanding between patients and physicians. Because SA has diverse cultures, we anticipate that studying patient-physician discordance may elucidate these specific challenges and guide the development of culturally sensitive and contextually appropriate interventions to reduce discordance, enhance the doctor-patient relationship, and improve disease management, patient satisfaction, and treatment adherence, leading to better long-term outcomes.

This study aimed to: (i) determine the frequency of patient-physician global assessment discordance in a South African RA cohort attending a tertiary academic hospital; (ii) identify clinical, demographic, and patient-reported outcome determinants of PGA and PhGA; and (iii) characterize the clinical features associated with patient-physician discordance.

Patients and methods

A cross-sectional study recruiting RA patients attending the arthritis outpatient clinics at a tertiary academic hospital. Consenting adults (≥18 years old) meeting the EULAR/ACR 2010 RA Classification Criteria (21) were invited to participate in the study.

Data collection

Demographic details including age, gender, self-expressed ethnicity, educational level, employment status, smoking status, and living situation were collated. A poor socio-economic setting (SES) was defined using a pooled index if any of the following were present: the highest level of education at primary school or less; more than 7 persons living in the household; clothes washing done by hand (ie no access to domestic washing machine) (22).

Disease-specific information including disease duration, medication, extra-articular manifestations, presence of rheumatoid factor and/or anti-cyclic citrullinated protein antibodies, joint surgery, and comorbidities was collated. Fibromyalgia Syndrome (FMS) was recorded as a comorbidity if it had been previously diagnosed by a treating clinician and was documented in the patient's clinical record. Active screening for FMS using formal diagnostic criteria was not performed in this study.

Clinical assessment with the 28-joint TJC, SJC, PGA, and PhGA. PGA and PhGA were recorded using a structured response card with five labelled anchor points at 0 (very good), 2.5 (good), 5 (fair), 7.5 (poor) and 10 (very poor). Intermediate values were occasionally recorded, and the scale therefore functioned as a near-continuous ordinal instrument. Disease activity was calculated using the CDAI, and classified into high, moderate, or low disease activity (HDA, MDA, or LDA) based on cut points of 22 and 10 (4).

The following PROMs were completed by patients, with the help of the study coordinator if required: the EuroQol (EQ-5D) to explore HRQoL, Health Assessment Questionnaire-Disability Index (HAQ-DI) to measure functional disability, Hospital Anxiety and Depression Score (HADS), FACIT -Fatigue Score and Brief Pain Inventory to assess anxiety and depression, fatigue and pain respectively (23–28).

PGA and the PhGA discordance

A discordance score was calculated by subtracting the PhGA score from the PGA score. Patients were categorized as Concordant (score between −2.5 and +2.5), Discordant (score ≥2.5) or Negatively Discordant (score < −2.5). A threshold of ≥2.5 was chosen as this represents the minimum detectable difference of one anchor-point step on the response instrument, consistent with published discordance thresholds of ≥2.0 to ≥3.0 on 0–10 global assessment scales (9–12).

Negatively discordant patients (n = 15, 2.7%) were excluded from the primary discordance analysis due to insufficient numbers for meaningful subgroup analysis, and because positive discordance represents the predominant and therapeutically relevant direction in the published literature (9, 10).

Statistical analysis

Pearson's correlation coefficients or Spearman's rank correlation coefficients were calculated for continuous variables and ordinal variables to identify factors correlating with PGA and PhGA. Correlation coefficients 0–0.19, 0.2–0.39, 0.40–0.59, 0.6–0.79 and 0.8–1 were considered as very weak, weak, moderate, strong, and very strong correlations respectively (29). Subsequently, the significant variables identified in univariate correlation were tested in multivariate analyses and further by linear stepwise regression to determine their contribution to PGA and PhGA. The proportion of variability of each outcome explained by each predictor was expressed by the partial R2. The proportions that could not be explained by the variables are presented as “unexplained”.

Variables achieving p < 0.05 in univariate analysis were entered as candidate predictors in multivariable regression models. The EQ5D Level Sum Score was not entered into the regression models; instead, the three functional EQ5D domains (mobility, self-care, and problems with usual activities) were entered individually as candidate predictors. The EQ5D pain and anxiety/depression subdomains were excluded as these constructs were more precisely assessed by the Brief Pain Inventory and HADS respectively. Regression analyses were conducted on 509 patients with complete data for all candidate predictors. Collinearity was assessed using Variance Inflation Factor (VIF); all values were <5, confirming no problematic multicollinearity. Standardised residual plots were examined to verify model assumptions.

Analysis of discordant and concordant groups was tested using Mann–Whitney U test and Pearson's Chi-Square test. Binomial logistic regression analysis was done with the inclusion of significant variables found in univariate analysis to assess the predictors of discordance.

Statistical analyses were conducted using SPSS version 25, and a p-value <0.05 was considered significant.

Ethical approval

This study was approved by the University of Cape Town Health Research Ethics Committee (HREC 629/2022). All participants provided written informed consent prior to enrolment

Results

Of 550 patients, the majority were female (84.9%), with a mean (SD) age and disease duration of 55.8 (12.9) and 11.5 (9.7) years, with 41.3% having longstanding RA (>10 years) (Table 1). A significant proportion (67.4%) were classified as having low SES; 32.7% of patients were smokers and the majority (75.1%) had at least one comorbidity. Most patients were in LDA or remission (47.8%), and the mean (SD) PGA was 4.9 (2.7) with a lower PhGA 3.3 (2.8). The majority of patients (58.3%) were prescribed ≥2 DMARDs concomitantly, 63.6% used low dose corticosteroids (CS) (≤5 mg/day), and only 13 (2.5%) were prescribed a biologic. The mean HAQ-DI score was 1.5 (0.7) suggesting significant physical disability, with global impairment revealed by the EQ-5D (Table 2). A third of patients had significant depression and anxiety, and the majority reported moderate to severe fatigue.

Table 1. Demographic and clinical details of 550 rheumatoid arthritis patients.

RA, rheumatoid arthritis; BMI, body mass index; RF, rheumatoid factor; ACPA, anti citrullinated peptide antibody; TJC, tender joint count; SJC, swollen joint count; PGA, patient global assessment; PhGA, physician global assessment; CDAI, clinical disease activity index; LDA, low disease activity; MDA, moderate disease activity; HDA, high disease activity; csDMARD, conventional synthetic disease modifying antirheumatic drug; bDMARD, biologic disease modifying antirheumatic drug.

Table 2. Patient-reported outcomes of 550 patients.

HAQ-DI, Health Assessment Questionnaire-Disability Index; EQ5D, European Quality of Life-5 Dimension; PSS, pain severity score; PIS, pain interference score, HADS, Hospital Anxiety and Depression scale.

Determinants of PGA and PhGA

The PGA correlated moderately with the pain severity score (PSS), TJC, SJC, and EQ5D Level Sum Score, and weakly with the pain interference score, HAQ-DI, fatigue, anxiety, and depression scores and with certain EQ5D domains (problems with usual activities, self-care, and mobility) and use of CS, and very weekly with SES (Table 3). There was no correlation with age, RA disease duration, or number of comorbidities. PhGA was strongly correlated with SJC and TJC and weakly with all PROMs.

Table 3. Factors influencing the PGA and PhGA—univariate analysis.

Data represent Pearson's correlation coefficient (r) or Spearman's rank correlation coefficient (ρ) for the Patient Global Assessment and the Physician Global Assessment. RA, rheumatoid arthritis; SES, socioeconomic status; TJC, tender joint count; SJC, swollen joint count; csDMARD, conventional synthetic disease modifying antirheumatic drug; OA, osteoarthritis; EQ5D LSS, European Quality of Life-5 Dimension; LSS, level sum score, HAQ-DI, Health Assessment Questionnaire-Disability Index; PSS, pain severity score; FACIT-F, Functional Assessment of Chronic Illness Therapy—Fatigue Scale; HADS, Hospital Anxiety and Depression scale.

Linear stepwise regression showed PGA was mainly determined by PSS (B = 0.449, R² = 0.257, p < 0.001), with additional contributions from TJC, SJC, problems with usual activities, prednisone use and fibromyalgia (Figure 1). These variables accounted for 38.3% of the variance in PGA, with 61.7% unexplained (p < 0.001). The major determinant of PhGA was SJC (B = 0.296, R² = 0.449, p < 0.001), which together with TJC, FACIT-fatigue, prednisone use and HAQ-DI explained 51.5% of the variance in PhGA, leaving 48.5% unexplained (p < 0.001). Regression analyses were conducted on 509 patients with complete data for all candidate predictors.

PGA and the PhGA discordance

Most patients (n = 399, 72.6%) and physicians had similar global assessments (Concordant group). Patient-physician discordance occurred in 151 (27.5%) of patients, with 136 patients (24.7%) rating their global assessments higher than their physician (Discordant group). Fifteen patients rated their global assessments lower than their physician's “Negatively Discordant”, and this group was excluded from this analysis.

Comparing the Discordant group with the Concordant group revealed significant differences in disease activity and some PROMs (Table 4). Patients in the Discordant group had lower TJC, lower SJC, and lower PhGA, with a higher proportion of patients in the LDA (58.1%) and MDA (32.4%) categories than in the HDA (9.6%) category (p = 0.001). Discordant patients also had a significantly worse EQ5D Level Sum Score, more problems with usual activities, and higher PSS, with a higher prevalence of fibromyalgia. There were no significant differences in age, gender, SES, duration of RA, comorbidities, or presence of osteoarthritis nor in HAQ-DI, anxiety, depression, or fatigue scores between the groups.

Table 4. Comparison of Concordant (PGA-PhGA: −2.5 to 2.5) and Discordant Group (PGA-PhGA ≥2.5).

p-value obtained from chi-square comparing the three groups.

CDAI incorporates both PGA and PhGA as components; group differences in CDAI-derived disease activity categories should therefore be interpreted with caution. TJC and SJC are reported as independent objective comparators.

RA, rheumatoid arthritis; CDAI, Clinical Disease Activity Index; OA, osteoarthritis; HAQ-DI, Health Assessment Questionnaire-Disability Index; EQ5D, European Quality of Life-5 Dimension; PSS, pain severity score; HADS, Hospital Anxiety and Depression scale; FACIT-F, Functional Assessment of Chronic Illness Therapy—Fatigue Scale.

Binomial logistic regression analysis showed that patients with lower SJC, high PSS, and problems with usual activities, are more likely to score their disease activity higher than their physician.

Discussion

Our study is the first to our knowledge to examine patient and physician global disease assessments, and the discrepancies between them in indigent RA patients living in SA. We show that pain severity had the strongest contribution to PGA, together with TJC, problems with usual activities, SJC, prednisone use and fibromyalgia. For PhGA, SJC was the most influential factor, with additional determinants including TJC, FACIT-fatigue, prednisone use and HAQ-DI. These findings align with studies conducted in various regions, indicating that patients perceptions of disease activity are influenced by the impact of RA, while physicians focus on objective signs of inflammation (12–16, 18, 30).

In our study, patient-physician discordance occurred in 27.5% of patients, with the majority of these rating their disease activity higher than their physicians. These results are somewhat lower than rates reported elsewhere, where up to half of patients score their arthritis differently to their physicians (10, 11, 31). Patients with discordance had poorer PROMs, particularly more problems with usual activities, higher PSS, a higher prevalence of fibromyalgia but notably lower TJC and SJC. The majority of discordant patients were in LDA or MDA states but did not feel they were doing well.

Although discordant patients were more frequently in LDA/remission by CDAI, this should be interpreted cautiously because CDAI inherently includes both global assessment scores. Because discordant patients contribute a higher PGA to their total CDAI score, this circularity mathematically biases them toward higher disease activity categories. The fact that these patients are still more frequently classified as being in remission or LDA is therefore striking, as it strongly emphasizes the absence of objective inflammation. Notably, objective measures such as TJC and SJC which are not influenced by global assessments were significantly lower in discordant patients (TJC 1.9 vs. 3.5, p < 0.001; SJC 1.3 vs. 3.5, p < 0.001). This confirms that discordance occurs predominantly in the context of objectively lower inflammation. Elsewhere, patients with longstanding disease have also been shown to have persistently high PGA scores despite low disease activity, joint counts, and inflammatory markers (17, 32). Pain and functional disability may have resulted in high PGA scores, even in the absence of joint inflammation (6, 33–36). Addressing both pain and activity limitations, particularly in patients with low joint counts, may reduce discordance and build a good doctor-patient relationship, leading to increased patient satisfaction with care.

Our findings align with the recent revision of the ACR/EULAR Boolean remission criteria, which raised the PGA threshold to ≤2. This revision notes that non-inflammatory factors such as pain and fatigue can raise PGA scores, hindering remission despite controlled inflammation (37). This directly supports our observation that discordant patients reported persistently elevated PGA driven by pain and functional limitations rather than active joint inflammation.

Mental health scores were not associated with discordance in our study although a third of patients had significant anxiety and/or depression, in contrast to findings by Barton et al. where higher depressive symptoms were associated with discordance (38).

There was significant unexplained variability in PGA in this study (61.7%). This is likely due to the subjective nature of PGA, influenced by individual perceptions and expectations together with other unmeasured factors, such as specific disease manifestations and social support. The cross-sectional design and inherent variability in PGA further contribute to this unexplained portion. Further research including qualitative interviews and longitudinal follow-up may allow a better understanding of this variability in PGA.

The majority of patients (89.1%) were receiving prednisone; most (63.6%) were on low doses of ≤5 mg/day. This is an area for improvement, given that most patients were in remission or LDA. In particular, discordant patients with low disease activity using CS require education regarding the toxicity of CS, and better pain management.

The main limitation of our study is the cross-sectional design, which restricts our ability to establish causal relationships and capture disease activity fluctuations over time. Additionally, as a cost–saving measure, acute phase reactants including CRP and ESR were not measured, and the CDAI was used to assess disease activity in keeping with practice in resource–constrained settings (39). The absence of CRP and ESR prevents us from distinguishing pain-driven discordance from discordance attributable to residual subclinical inflammation not captured by joint counts.

Additionally, musculoskeletal ultrasound synovitis scoring and radiographic erosion assessment were not performed; both may contribute to the unexplained variance in PGA and are recommended for future prospective studies, as demonstrated by Floris et al. (20).

Fibromyalgia was recorded as a pre-existing comorbidity rather than actively screened using formal diagnostic criteria; its true prevalence in this cohort may therefore be underestimated. Negative discordance (PhGA > PGA), observed in 15 patients (2.7%), represents a distinct phenomenon warranting dedicated investigation in future studies.

In conclusion, our findings in indigent RA patients with longstanding disease suggest that over a quarter of patients rate their disease activity higher than their clinicians, particularly when disease activity is low. Pain severity and functional limitations play a crucial role in patients' assessment of their disease; in this cross-sectional cohort, addressing these factors may reduce patient-physician discordance. Clinical review (with musculoskeletal ultrasound), pain control strategies, and assessment by occupational and physiotherapists together with exercise may reduce discordance. Additionally, the unexplained variability in PGA highlights the need for further research incorporating a wider range of variables including structural damage, health literacy and patient expectations.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The author(s) verify and take full responsibility for the use of generative AI in the preparation of this manuscript. Generative AI (Claude, Anthropic) was used to assist with proofreading and editing of manuscript text. All statistical analyses were conducted independently by the authors using SPSS version 25. All AI-assisted content was reviewed, verified, and approved by the authors prior to submission. The scientific content, interpretation of results, and conclusions are the sole responsibility of the authors.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Deshire Alpizar-Rodriguez, National Institute of Rehabilitation Luis Guillermo Ibarra Ibarra, Mexico

Reviewed by: Paloma De Abreu Trigueros, Sociedad Paraguaya de Reumatología, Paraguay

Rodrigo Garcia Salinas, Hospital Italiano de La Plata, Argentina