DIRECT AND INDIRECT DETERMINANTS OF THE PATIENT GLOBAL ASSESSMENT IN RHEUMATOID ARTHRITIS: DIFFERENCES BY LEVEL OF DISEASE ACTIVITY

Abstract

Objective

In rheumatoid arthritis (RA), the patient global assessment (PGA) has been strongly associated with pain severity, but less often with other measures, including disease activity measures. We tested if RA activity and psychological measures had direct associations with PGA, or indirect associations that were mediated by pain. We also tested if the correlates of PGA differed with the degree of RA activity.

Methods

We studied 260 patients with active RA on two visits in a prospective longitudinal study. We used path analysis to test direct and indirect associations of DAS28, morning stiffness, Health Assessment Questionnaire (HAQ), fatigue, physical role limitations, social functioning, depressive symptoms, and health distress with the PGA.

Results

Among the 509 visits, the median PGA was 50 (25^th, 75^th percentile 24, 66). Pain severity had the strongest association with PGA, but direct associations were also found for morning stiffness severity, health distress, fatigue, and DAS28. Morning stiffness severity, DAS28, health distress, and HAQ were also indirectly associated with PGA through pain. Among visits with DAS28 ≥ 5.4, pain, morning stiffness severity, and HAQ were the only determinants of PGA. Among visits with DAS28 < 4.2, health distress and age were additional determinants, and fatigue was marginally associated with PGA.

Conclusions

Although pain was the strongest determinant of PGA in RA, morning stiffness severity, health distress, fatigue, and DAS28 were also important. Determinants of PGA differed with RA activity, with health distress, age, and to a lesser degree, fatigue, contributing only in patients with less active RA.

The patient global assessment (PGA) is the most widely used patient-reported measure in rheumatoid arthritis (RA), being a component of disease activity scores, treatment response criteria, and definitions of remission. In marking the PGA, patients are implicitly asked to integrate many aspects of their arthritis, including pain, functional ability, stiffness, and joint swelling. Given its broad multi-attribute nature, patients may differ in which aspects of arthritis they consider when rating the PGA, and some patients may include factors unrelated to RA activity.

Many studies have shown that pain severity is the dominant predictor of the PGA, with functional limitations and fatigue having additional, although weaker, associations [1-10]. In contrast, joint counts and disease activity measures such as the Disease Activity Score (DAS) have been reported to have only weak, or no, associations with the PGA [4-10]. Its low association with these cardinal measures of RA activity has clouded the validity of the PGA. The broad nature of the PGA may also make it particularly susceptible to influence by the patient’s mood or situational factors present at the time of the assessment. Depression has been reported to affect the PGA, but the influence of other psychological factors has not been extensively studied [9-14]. Questions therefore remain about what aspects of RA are the most important determinants of the PGA, and the extent to which the PGA is influenced by factors other than RA activity.

Previous studies either reported only univariable associations between the PGA and other RA activity and psychological measures, or used multivariable regression modeling to find measures that were independently associated with the PGA. While this analytic approach can identify the most proximate determinants of the PGA, it does not consider indirect, or mediational, relationships among determinants of the PGA. For example, an apparent lack of association between the DAS and PGA in a multivariable analysis that includes both the DAS and patient-reported pain severity as predictors may be a consequence of stronger associations between pain and PGA than between the DAS and PGA. Such an analysis ignores the link between the DAS and pain, and the potential contribution of the DAS to the PGA through its influence on pain severity. Consequently, these analyses may have underestimated the full array of determinants of the PGA.

We hypothesized that many RA activity measures likely had indirect associations with the PGA that were mediated by pain severity, the measure most salient to patients. We tested this hypothesis using path analysis, an extension of multiple regression analysis that explicitly models associations among independent variables to test indirect as well as direct relationships with the dependent variable [15,16]. We also tested if patient-reported depression, health distress, social functioning, and role limitations were associated with the PGA. In addition, we examined if the correlates of PGA varied with the level of RA activity. We hypothesized that pain severity would predominate in states of high RA activity, while factors other than pain severity would surface when RA activity was lower.

METHODS

Participants and study design

We used data from a prospective longitudinal study of clinical changes in RA [17]. In this study, we enrolled adults with active RA who were having an escalation of their anti-rheumatic medication at an outpatient visit. We examined participants at two visits, before treatment escalation and either 1 month (for those treated with prednisone) or 4 months (for all others) after escalation. Joint counts, questionnaires, and laboratory testing were done on both visits. In this analysis, we included participants whether or not they completed both visits. The study was approved by the institutional review board. All participants provided written informed consent.

Measures

The PGA asked “Considering all the ways that your rheumatoid arthritis affects you, rate how you are doing,” and was measured on a visual analog scale with anchors of 0 (very well) and 100 (very poor). Pain severity and the severity of morning stiffness were also rated using visual analog scales (possible range 0 – 100, with anchors of none and severe). Participants were also asked the duration of morning stiffness. We used the Health Assessment Questionnaire (HAQ; possible range 0 – 3) Disability Index to measure physical functioning [18]. We measured fatigue, social functioning, and role limitations due to physical health problems using scales of the Short Form-36 (possible range 0 – 100, with higher scores indicating better health) [19]. We included social functioning and physical role limitations because they were thought possibly important to patients’ appraisals of how well they were doing.

To measure depressive symptoms, we used the Center for Epidemiological Studies – Depression (CESD) scale, a 20-item questionnaire that asks the frequency of depressive feelings. For analysis, we omitted 4 items that have been associated with RA activity rather than mood (possible range 0 – 48, with higher scores indicating more symptoms) [20]. We measured health distress using the Medical Outcomes Study Health Distress scale, a six-question scale that asks about worry, frustration, discouragement, and despair about health problems (0 – 100, with higher scores indicating less distress) (Supplemental table 1) [21]. We also calculated the three-variable DAS28, excluding the PGA from this measure to avoid overestimating its association with the PGA.

We excluded the visits of two patients who were missing data on more than one questionnaire measure. Complete data were available for 497 of the remaining 509 visits. We used mean imputation to estimate isolated missing values on 12 visits. No PGAs were missing.

Statistical analysis

We used path analysis to examine direct and indirect associations between the RA activity, functioning, and psychological measures and PGA. Path analysis is a modification of multivariable regression analysis that uses hierarchical models to test associations between a set of predictors and more than one dependent (or endogenous) variable simultaneously based on the covariance structure of the data [15,16].

Based on previous studies that reported pain severity as the dominant influence on PGA, we tested a model with pain and PGA as endogenous variables, pain as a mediator variable, and the nine other clinical measures as predictor (or exogenous) variables. We included age and sex as additional covariates. The model allowed for all measures to have both a direct effect on PGA and an indirect effect on PGA through pain (Supplemental Figure 1). To arrive at a final model, we first tested all direct effects on PGA and pain, and retained those that were significant (p < 0.05). We then sequentially added other paths based on model modification indices, and tested indirect associations through pain, retaining those that were significantly associated with pain, while also not worsening the overall model fit. We tested model fit using the root mean square error of approximation (RMSEA) and comparative fit index. Models are generally considered to have good fit when the RMSEA is close to 0 and the comparative fit index is greater than 0.90 [22,23].

Path coefficients (B) represent the magnitude and direction of associations between the measures, adjusted for other measures in the model. Path coefficients are the beta coefficients of the multivariable regression models that are standardized to the standard deviation of PGA. All path coefficients are therefore on the same scale and directly comparable.

To increase statistical power, we pooled data from both visits, and used cluster bootstrap sampling to estimate standard errors of the regression parameters to account for the clustering of observations within patients [24]. Including both the pre-treatment and post-treatment visits also increased the range of PGAs represented.

We repeated the analysis on subsets of visits stratified by tertile of DAS28 to determine if the major determinants of PGA varied with the level of RA activity. We used tertiles of the DAS28 rather than conventional cutpoints of low, moderate, and high disease activity because too few subjects had low disease activity by these criteria for a stable model (see Supplemental figure 2).

All statistical analyses were done using R version 3.2.0, and path analysis was performed using the R lavaan package (version 0.5-18) [25,26].

RESULTS

Study participants

The study included 260 patients (mean age 51.1 ± 13.6; 78% women; median duration of RA 6.4 years (25^th, 75^th percentile 2.2, 14.8); 80% seropositive; 64% erosive). Among all 509 visits (260 baseline visits and 249 follow-up visits), the median PGA was 50 (25^th, 75^th percentile 24, 66), with the full range of values (0 to 100) represented. The distributions of the RA activity and psychological measures that were tested as correlates of the PGA are shown in Table 1.

Table 1.

Values of clinical measures at either the baseline or follow-up visit (N = 509).

Measure	25th percentile	Median	75th percentile
Patient global assessment (0 – 100)	24	50	66
Pain severity (0 – 100)	28	51	73
Morning stiffness duration, min (0 – 480)	12	60	180
Morning stiffness severity (0 – 100)	22	49	73
Health Assessment Questionnaire (0 – 3)	0.625	1.25	1.75
DAS28 (1 – 9.4)	4.0	4.8	5.7
CESD (0 – 48)	5	12	19
Physical role limitations (0 – 100)*	0	25	50
Social functioning (0 – 100)*	37.5	62.5	87.5
Fatigue (0 – 100)*	25	45	60
Health distress (0 – 100)*	36.7	60	76.7

^*Higher values indicate better health.

DAS28 = Disease Activity Score 28; CESD = Center for Epidemiological Studies Depression scale.

Analysis of all visits

In the optimal-fitting model, pain severity had the strongest direct association with PGA (B = .42) (Figure 1). Greater severity of morning stiffness (B =.21) and more health distress (B = −.19) were also significantly directly associated with PGA, although their associations were only half as strong as that of pain. Higher levels of fatigue, and higher DAS28 were also directly associated with PGA, but these associations were weaker.

Figure 1.

Final path model including data from all visits (n = 509). Significant path associations are shown with corresponding path coefficients (standardized betas). Higher scores for fatigue and health distress indicate less fatigue and less distress; therefore negative coefficients for these measures indicate that more fatigue and more health distress are associated with higher patient global assessments. DAS28 = Disease Activity Score 28; HAQ = Health Assessment Questionnaire.

* p < .05 ^† p < .001 ^‡ p < .0001

Several measures were also associated with pain severity. There were direct associations between the severity of morning stiffness, DAS28, HAQ, and health distress and pain severity. Consequently, there were significant indirect associations between the morning stiffness severity, DAS28, HAQ, and health distress and the PGA that were mediated by pain severity. The indirect association of morning stiffness severity was twice as strong as its direct association with PGA (B = .44 versus B = .21), while the direct association of health distress with PGA was stronger than its association with pain.

Among the exogenous variables, severity of morning stiffness had the largest total (direct and indirect) effect on PGA (B = .39), followed by health distress (B = −.23), fatigue (B = −.14), DAS28 (B = .13), and HAQ (B = .11).

No significant direct or indirect associations were found for the CESD, physical role limitations, social functioning, age, or sex. The model fit was excellent, with RMSEA = .012 (p = .55) and comparative fit index = 1.0. The model R² was 0.66.

Subgroups of RA activity

To determine if the correlates of PGA differed with the degree of RA activity, we analyzed subgroups stratified by DAS28 (< 4.2; 4.2 to 5.39; ≥ 5.4) (Table 2 and Supplemental table 2).

Table 2.

Values of clinical measures in subgroups of Disease Activity Score-28.

Measure	25th percentile	Median	75th percentile
DAS28 ≥ 5.4 (N = 172 visits)
Patient global assessment (0 – 100)	48	61	77
Pain severity (0 – 100)	51	70	83
Morning stiffness duration, min (0 – 480)	36	120	285
Morning stiffness severity (0 – 100)	50	69	81
Health Assessment Questionnaire (0 – 3)	1.25	1.75	2.25
CESD (0 – 48)	8	14	21
Physical role limitations (0 – 100)*	0	0	25
Social functioning (0 – 100)*	37.5	50	75
Fatigue (0 – 100)*	15	35	50
Health distress (0 – 100)*	27	47	70
DAS28 4.2 – 5.39 (N = 184 visits)
Patient global assessment (0 – 100)	28	50	62
Pain severity (0 – 100)	30	50	67
Morning stiffness duration, min (0 – 480)	18	60	150
Morning stiffness severity (0 – 100)	28	48	70
Health Assessment Questionnaire (0 – 3)	0.75	1.125	1.625
CESD (0 – 48)	5	12	18
Physical role limitations (0 – 100)*	0	12.5	50
Social functioning (0 – 100)*	44	62	87
Fatigue (0 – 100)*	30	45	55
Health distress (0 – 100)*	40	60	77
DAS28 <4.2 (N = 153 visits)
Patient global assessment (0 – 100)	10	24	48
Pain severity (0 – 100)	10	28	50
Morning stiffness duration, min (0 – 480)	6	18	60
Morning stiffness severity (0 – 100)	9	25	49
Health Assessment Questionnaire (0 – 3)	0.125	0.625	1.125
CESD (0 – 48)	3	9	16
Physical role limitations (0 – 100)*	0	50	100
Social functioning (0 – 100)*	62.5	75	100
Fatigue (0 – 100)*	40	55	70
Health distress (0 – 100)*	46	72	88

^*Higher values indicate better health.

DAS28 = Disease Activity Score 28; CESD = Center for Epidemiological Studies Depression scale.

Among visits with DAS28 ≥ 5.4, only pain severity, stiffness severity, and HAQ had direct associations with PGA, with pain severity having the strongest association (Figure 2a). Stiffness severity and HAQ also were indirectly associated with PGA through pain. The associations of stiffness severity and HAQ with pain were much stronger than their direct associations with PGA, indicating that the dominant effects of these measures were on pain severity. The model R² was 0.47 (RMSEA = 0 (p = 1.0); comparative fit index = 1).

Figure 2.

Path models in subgroups of visits by Disease Activity Score 28. Significant path associations are shown with their corresponding path coefficients (standardized betas). Higher scores for fatigue and health distress indicate less fatigue and less distress; therefore negative coefficients for these measures indicate that more fatigue and more health distress are associated with higher patient global assessments. Dashed arrows in panel C represent non-significant associations. DAS28 = Disease Activity Score 28; CESD = Center for Epidemiological Studies Depression scale; HAQ = Health Assessment Questionnaire.

* p < .05 ^† p < .001 ^‡ p < .0001

Among visits with DAS28 between 4.2 and 5.39, pain severity, stiffness severity, and CESD had direct associations with the PGA, while stiffness severity and HAQ additionally had indirect associations through pain (Figure 2b). The indirect association of stiffness severity was twice as strong as its direct association with PGA. Pain severity had the strongest association with PGA among all measures in this subgroup, but stiffness severity and CESD also contributed substantially. The model R² was 0.55 (RMSEA = 0 (p = .61); comparative fit index =1).

Among visits with DAS28 < 4.2, pain severity again had the strongest association with PGA, but in addition, greater morning stiffness severity, more health distress, and older age were directly associated with higher PGA (Figure 2c). More fatigue (p = .08) and poorer social functioning (p = .08) had marginal direct associations with PGA in this subgroup. We retained these paths, along with the direct association of HAQ (p = .21) and the indirect associations with age (p = .50), health distress (p = .08), and fatigue (p = .41), because doing so greatly improved the model fit (RMSEA = 0 (p = .42); comparative fit index = 1). Considering both its direct and indirect associations, fatigue had a significant total association with PGA (p = .05). The model R² was 0.68. Morning stiffness severity was associated with PGA both directly and indirectly, although its indirect association was stronger.

DISCUSSION

Among our patients, who presented a broad range of RA activity, the PGA captured multiple aspects of RA, and therefore can truly be considered a global measure. In addition to pain severity, patients’ assessments were associated with the severity of morning stiffness, fatigue, health distress, and the DAS28. The importance of the HAQ, morning stiffness severity, DAS28, and health distress was revealed by the use of path analysis, which demonstrated the added associations of these measures with PGA that were mediated by pain severity. These indirect associations would not have been uncovered using conventional modeling approaches.

Similar to previous studies, pain severity was the strongest predictor of PGA, with an association twice as strong as any other measure in the overall group [1,3-7,9]. Pain was also the strongest predictor of PGA in each DAS28 subgroup. This consistency corroborates prior studies that reported strong associations between pain severity and the PGA in cohorts with low or high RA activity [4-7,9,10]. These findings confirm the centrality of pain to patients’ assessments of their RA. The severity of morning stiffness was the second strongest predictor in the overall model. It contributed directly to the PGA, and indirectly through an association with pain severity. Few studies have assessed morning stiffness severity in RA [27]. This measure appears to capture the importance of stiffness better than the duration of morning stiffness, which was not associated with the PGA in our analysis. Morning stiffness severity, pain severity, and PGA were all measured using visual analog scales, and use of the same measurement approach may have inflated the associations among these measures. Although these scales were separated in the questionnaire to reduce possible carry-over effects, we cannot exclude the possibility that some carry-over occurred. Additional research on morning stiffness severity that uses alternative scales will be important to dissect these associations further. Fatigue had weak but significant associations with the PGA in our cohort. In the only other study that included fatigue in a multivariable analysis, fatigue was second to pain severity in explaining variation in PGA [4].

The DAS28 was associated with PGA both directly and through an association with pain. The DAS28 association was weaker than that of any of the patient-reported measures, however. This finding is consistent with the results of other studies that examined the relative importance of patient-reported measures and joint counts to variation in PGA in multivariable analyses [4,5,7]. Joint counts had only minor associations with PGA in the studies of Khan et al and Studenic et al, with R² of less than 2% and 0.5%, respectively [4,5]. There was no significant association between joint count measures and PGA in the study of Furu, et al, after accounting for pain and functional class [7]. However, these studies examined patients with low or moderate RA activity, with mean DAS28 of 4.3, 3.9, and 3.2, respectively, and mean PGAs of 40, 38, and 35, respectively [4,5,7]. In contrast, our cohort had more active RA, with a median DAS28 of 4.8 and median PGA of 50. It is possible that associations with clinician-based measures such as the DAS28 and joint counts are more prominent at high levels of RA activity, because, in this setting, they provide additional discrimination among patients with similarly high self-reported RA activity.

A patient’s psychological state can influence their ratings of self-reported measures of RA activity [11,12, 28]. Most prior studies focused on depression and anxiety, and reported generally small to moderate associations of these measures with the PGA [4,9,10-12]. In our cohort, depression was not associated with the PGA in the overall cohort, but was significantly associated with the PGA among patients with moderately active RA. In contrast, health distress was associated with both pain and PGA in the overall cohort. Health distress encompasses concerns and worry about health problems specifically, and is distinct from depression and general anxiety [21]. Psychometric studies of the health distress scale have found it to be equally related to physical health and mental health [29-31]. In our cohort, health distress was correlated with depression (r = −.67), but also with pain (r = −.44), DAS28 (r = −.32), and HAQ (r = −.46). Its focus on health-related concerns may have enhanced its association with the PGA, as these concerns can be present in patients without mood disorders. Health distress may capture the emotional significance of illness better than measures of depression or generalized anxiety for most patients. As a health measure, health distress should be distinguished from coping strategies such as catastrophizing, which represent ways to deal with stressors. Further research should examine health distress as a measure of disease impact and assess its relationship with other patient-reported outcomes [32,33].

In the subgroup analysis, health distress was associated with PGA only among patients with less active RA, as were fatigue and age. These associations suggest that at lower levels of RA activity, the PGA is more likely to be affected by measures outside the core domains of RA. This finding is consistent with the hypothesis that as the symptom level decreases, patients give more consideration to secondary factors, so that health distress and fatigue contribute to variation in the PGA. The association with age may reflect decreased resilience or effects of comorbidities. In contrast, pain, morning stiffness severity, and HAQ were the only correlates of PGA in patients with very active RA. When the symptoms were severe, the core RA symptoms, particularly pain severity, dominated patients’ considerations when marking the PGA.

Despite several strengths, our study is limited in having few patients with low RA activity, inclusion of whom would have provided more opportunity to examine the contribution of non-core measures to PGA at this end of the activity spectrum. We also did not measure general anxiety, and did not assess pain and morning stiffness severity using scales other than visual analog scales. Using different question formats would have allowed us to distinguish these measures better. We chose to focus on constructs most closely related to the core feature of RA activity, rather than potentially even more indirect measures such as coping styles or sleep quality. In addition, models that tested mediators other than pain severity could have been examined but have less theoretical support. Lastly, all associations were correlational, and causation cannot be inferred.

Our findings indicate that the PGA captures multiple aspects of RA activity, but also have implications for how the PGA should be interpreted at high and low RA activity. Because PGA largely reflects pain severity at high levels of RA activity, these measures convey much the same information in these patients. Clinical trials in patients with active RA may therefore not be truly capturing separate aspects of RA activity when responses for both PGA and pain measures are counted, as for example, in the American College of Rheumatology response criteria. At low RA activity, the PGA does not reflect RA activity solely, as has been noted previously [10,13,14]. This finding suggests that use of the PGA as a component criterion of remission should be re-examined. Interpretation of studies that use the PGA as a predictor of other outcomes may be complicated because the PGA at high disease activity represents a different set of RA manifestations and concerns than the PGA at lower disease activity. Use of a more specific measure, such as pain, may be more interpretable. In clinical practice, these different influences may contribute to patient-clinician discordance, particularly among patients with less active RA.

SIGNIFICANCE AND INNOVATION.

• This is the first study to use path analysis in examining determinants of the patient global assessment in RA and to compare these determinants in very active and less active patients directly.

• This is the first study to investigate the association of health distress with the patient global assessment.