Article Text

Download PDFPDF

Psychometric properties of the Rheumatoid Arthritis Disease Activity Index (RADAI) in a cohort of consecutive Dutch patients with RA starting anti-tumour necrosis factor treatment
  1. M M Veehof1,
  2. P M ten Klooster1,
  3. E Taal1,
  4. P L C M van Riel2,
  5. M A F J van de Laar1,3
  1. 1
    Institute for Behavioral Research, University of Twente, Enschede, The Netherlands
  2. 2
    Department of Rheumatology, Radboud University Nijmegen Medical Centre, Nijmegen, The Netherlands
  3. 3
    Department of Rheumatology, Medisch Spectrum Twente, Enschede, The Netherlands
  1. Miss M M Veehof, Institute for Behavioral Research, Faculty of Behavioral Sciences, University of Twente, PO Box 217, 7500 AE Enschede, The Netherlands; M.M.Veehof{at}utwente.nl

Abstract

Objectives: To examine the psychometric properties of the self-administered Dutch Rheumatoid Arthritis Disease Activity Index (RADAI) and its short form (RADAI-SF) in patients with rheumatoid arthritis starting anti-tumour necrosis factor treatment.

Method: Internal consistency was assessed with Cronbach’s α. A confirmatory factor analysis (CFA) was carried out to test the single-factor structure. Construct validity was examined by correlating RADAI and RADAI-SF scores with Disease Activity Score in 28 joints (DAS28). Internal responsiveness was evaluated with the paired t test and the standardised response mean (SRM). External responsiveness was assessed with receiver operating characteristic analysis and the SRM, using the EULAR response criterion as external criterion. Change scores were correlated with changes in DAS28.

Results: At baseline and after 3 months’ treatment, respectively, 191 and 171 patients completed the RADAI. The internal consistency of the RADAI and the RADAI-SF was satisfactory. CFAs confirmed the single-factor structure of both RADAI versions, but the short form provided the best model fit. Moderate correlations were found with the DAS28. SRMs of the RADAI and the RADAI-SF were, respectively, 0.76 and 0.80. Both versions had moderate accuracy to distinguish responders from non-responders. Changes scores were moderately correlated with DAS28 change scores.

Conclusions: This study showed satisfactory psychometric properties of the Dutch version of the RADAI. Omission of the tender joint count (RADAI-SF) produced comparable results and is justified for research purposes. The tender joint count might be useful as additional clinical information in patient management.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Disease activity is an important concept in the evaluation of patients with rheumatoid arthritis (RA) in clinical care and research. Because there is no “gold standard” of disease activity in RA, multiple single variables (core-set variables) and index measures are used. An index measure of disease activity combines single variables representing several aspects of the disease. Index measures are considered to be more informative than single measures and have the advantage of avoidance of multiplicity and increased sensitivity to change.13

A widely used and accepted index measure of disease activity is the Disease Activity Score in 28 joints (DAS28).4 The DAS28, like most other indices of disease activity, primarily consists of physician-assessed and laboratory based variables. These variables are time consuming to assess, not always (directly) available and subject to interobserver variation. Furthermore, these variables do not take into account a patient’s perception of the burden of the disease, which has become increasingly important in the evaluation of treatment response and treatment management. The Rheumatoid Arthritis Disease Activity Score (RADAI) can be used as an alternative for, or complement to, these variables.5 6

The RADAI is a short and easy to complete self-administered index measure, combining a patient’s perception of past disease activity, current disease activity as measured by swollen and tender joints, pain, duration of morning stiffness and tender joint count into a single measure of disease activity. The RADAI has primarily been developed for use in clinical and epidemiological studies where clinical assessments are not available or too demanding.5 6 Nevertheless, the RADAI may also be useful in clinical practice.6 For research purposes, the tender joint count (item 5), which is time consuming and adds little or nothing to the measure, can be omitted from the RADAI.5

Previous studies on the psychometric properties of the RADAI have primarily focused on the five-item version of the RADAI and not on its short form (RADAI-SF). The RADAI has been shown to have adequate reliability, validity and responsiveness among Swiss patients with RA.57 Responsiveness was only investigated in patients showing worsening of disease activity over time.7 The results of this study cannot be generalised to patients showing improvement of disease activity. Demonstration of the responsiveness to detect improvement of disease activity is mandatory, especially in the present era where remission has become an option for patients with RA. The aim of this study was to assess the reliability, validity and responsiveness of the Dutch version of the RADAI and its short form (RADAI-SF) in a cohort of consecutive patients with RA starting with tumour necrosis factor (TNF) blocking treatment.

PATIENTS AND METHODS

Patients

Participants were from the continuing Dutch Rheumatoid Arthritis Anti-TNF Monitoring (DREAM) study, a multicentre study that started in April 2003 to monitor and evaluate prospectively the use of anti-TNF in patients with RA. Inclusion criteria for the DREAM study are a diagnosis of RA,8 active disease (DAS28 >3.2),4 previous treatment with at least two antirheumatic drugs including methotrexate at an optimal dose or intolerance for methotrexate, and no previous treatment with anti-TNF agents.

Measures

In the DREAM study, patients are seen every 3 months by trained research nurses who collect data on core disease activity variables, including 28 tender joint count (28-TJC, range 0–28), 28 swollen joint count (28-SJC, range 0–28), erythrocyte sedimentation rate (ESR), patient’s assessment of general health on a 100 mm visual analogue scale (VAS-GH), and Health Assessment Questionnaire Disability Index (range 0–3).9 10 For this part of the study we used data from a subset of centres that additionally administered the RADAI at study entry and after 3 months.

RADAI

The RADAI is a disease-specific questionnaire developed to measure self-assessed disease activity in patients with RA.5 The questionnaire has previously been translated into Dutch and was applied in several studies.1114 The RADAI contains five items on global disease activity during the past 6 months (item 1), current disease activity as measured by swollen and tender joints (item 2), current amount of arthritis pain (item 3), current duration of morning stiffness (item 4) and current number of tender joints in a joint list (item 5). The first three items are scored on an 11-point numerical rating scale, with verbal anchors from “no disease activity”/”no pain” (score 0) to “extreme disease activity”/”extreme pain” (score 10). The last two items are scored on a seven-point (item 4) and four-point (item 5) verbal rating scale. The scores on these two items range from 0 to 6 (item 4) and from 0 to 48 (item 5), and were transformed to a 0–10 scale, with higher scores indicating more disease activity. The total score of the RADAI was computed by summing the scores of the individual items and dividing this by five. The score of the short form (RADAI-SF) was computed by summing the scores of the first four items and dividing this by four, leaving out item 5 (tender joint count).

DAS28

From the 28-TJC, 28-SJC, ESR and the VAS-GH the DAS28 was computed. The DAS28 range is from 0 to approximately 10, where higher scores indicate more disease activity.4

Data analysis

Continuous data were presented as means with standard deviations (SDs) or medians with interquartile ranges (IQRs), depending on the distribution of the data (tested with the Kolmogorov–Smirnov test). Categorical data were presented as proportions. Analyses were performed using the statistical packages SPSS 12.0, LISREL 8.70, S-PLUS 6.1 and MedCalc 8.1.

Reliability

The internal consistency of the RADAI and the RADAI-SF was assessed with Cronbach’s α coefficient using the data obtained from the baseline assessment. According to Nunnally and Bernstein, a value of 0.80 is sufficient for research purposes and a value of 0.90 is recommended when individual decisions are made based on specific test scores.15

Construct validity

A confirmatory factor analysis (CFA) using the maximum likelihood estimation procedure was conducted with LISREL to test the single-factor structure of the RADAI and the RADAI-SF. Covariances between the (transformed) item scores at baseline were used as input. As recommended, multiple fit indices were used to evaluate the fit of the data to a single-factor model.1618 We used the following fit indices: χ2 statistic with degrees of freedom (df), non-normed fit index (NNFI) (comparable with Tucker–Lewis Index), comparative fit index (CFI), and root mean square error of approximation (RMSEA). A χ2/df ratio <2, combined with an NNFI value >0.95, a CFI value >0.90, and an RMSEA value <0.08 indicate a good model fit.1820

Correlation analysis was used to investigate further the construct validity. Pearson correlation coefficients were calculated between RADAI and RADAI-SF scores and DAS28 scores at baseline. Correlations ⩾0.90 were interpreted as very high, 0.70–0.89 as high, 0.50–0.69 as moderate, 0.26–0.49 as low and ⩽0.25 as little if any correlation.21

Responsiveness

In accordance with Husted et al, we distinguished between internal and external responsiveness.22 Internal responsiveness refers to the ability of a measure to change over a prespecified time frame, whereas external responsiveness describes the relationship between change in a measurement and change in a reference measure of disease activity.

Internal responsiveness was firstly assessed with the paired samples t test. Change between baseline and 3-month follow-up assessments was considered significant when p⩽0.05. Second, the standardised response mean (SRM) was calculated. The SRM is calculated as the mean change score divided by the standard deviation of the change score and is seen as an indicator of the ability to distinguish “signal” from “noise”.23 24 An SRM between 0.20 and 0.49 can be interpreted as a small effect, an SRM between 0.50 and 0.79 as a moderate effect and an SRM ⩾0.80 as a large effect.22 We applied a bootstrap procedure with S-PLUS to obtain the 95% confidence interval for the SRM.25

External responsiveness was assessed with receiver operating characteristic (ROC) curve analysis and the SRM. We used the EULAR response criterion as external criterion for clinical change.26 According to this criterion, patients were classified as (moderate or good) responders or non-responders, dependent on the individual change in DAS28 and the level of DAS28 reached. An ROC curve was created by plotting the true-positive proportion (sensitivity) versus the false-positive proportion (100−specificity) for the discrimination between responders and non-responders for multiple cut-off points. The area under the ROC curve was calculated to quantify the discriminative accuracy. The area under the curve ranges from 0.5 to 1.0, where an area of 0.5–0.7 indicates low accuracy, 0.7–0.9 moderate accuracy and >0.9 high accuracy.27 SRMs were calculated for responders and non-responders. To further investigate the external responsiveness, change scores of the RADAI and the RADAI-SF were correlated with change scores of the DAS28.

RESULTS

At baseline and after 3 months, respectively, 191 and 171 patients fully completed the RADAI. Mean (SD) age of the patients at baseline was 54.5 (13.3) years and median disease duration was 7.0 (IQR 3.0–17.0) years. The majority of the patients (71%) were female and had mild disability according to Steinbrocker’s functional classification (84% in class II). Table 1 shows the mean scores on disease activity measures at baseline and at 3 months.

Table 1 Mean scores (SD) on disease activity measures at baseline and after 3 months*

Reliability

The internal consistency of the RADAI and the RADAI-SF, measured with Cronbach’s α coefficients, was 0.84 and 0.82, respectively. Deletion of the items one by one did not change the coefficients substantially.

Construct validity

CFAs of the RADAI and the RADAI-SF showed that the RADAI-SF provided the best fit of the data to a single-factor structure (table 2). All fit indices of the RADAI-SF met the recommended criteria of acceptable model fit. For the RADAI, half of the fit indices (NNFI and CFI) satisfied the recommended criteria. The fit indices χ2/df and RMSEA were nearly acceptable. Post hoc modification analysis of the RADAI showed that the assumption of uncorrelated error terms did not hold. Correlated error terms were found between items 2 (current disease activity as measured by swollen and tender joints) and 5 (current number of tender joints in a joint list) (r = −0.08), and items 3 (current amount of arthritis pain) and 5 (r = 0.09). Inclusion of an error covariance between the pair of items with the highest correlated error terms (items 3 and 5) provided an acceptable model fit by all fit indices (without further suggestions for modification). Figure 1 shows the standardised factor loadings and error terms of the RADAI (including its refined version with an error covariance) and the RADAI-SF. The lowest factor loadings were found for items 4 (duration of morning stiffness) and 1 (past disease activity).

Figure 1 Standardised parameter estimates for the original and refined model of the Rheumatoid Arthritis Disease Activity Index (RADAI) and the model of the RADAI-short form (RADAI-SF).
Table 2 Fit indices for the one-factor structure of the RADAI and the RADAI-SF (n = 191)

RADAI and RADAI-SF scores correlated moderately (respectively 0.53 and 0.52) with DAS28 scores at baseline (n = 186).

Internal responsiveness

Scores on the RADAI, the RADAI-SF, and the DAS28 were significantly improved after 3 months of anti-TNF treatment (p<0.001) (table 1). The SRMs were respectively 0.80, 0.76, and 1.09, indicating moderate to large effects (table 3).

Table 3 Internal and external responsiveness indices (95% confidence intervals) for the RADAI and the RADAI-SF

External responsiveness

According to the EULAR response criterion, 71% of the patients were classified as responders (26% good responder; 45% moderate responder) after 3 months of anti-TNF treatment and 29% as non-responders. The responders showed significant improvement on both the RADAI (mean (SD) change score −2.17 (1.81)) and the RADAI-SF (mean (SD) change score −2.32 (2.05)). The non-responders showed no improvement on both versions (mean (SD) change score RADAI −0.29 (1.93); mean (SD) change score RADAI-SF −0.30 (2.17)).

Table 3 shows the external responsiveness indices for the RADAI and the RADAI-SF. The areas under the ROC curves show that the RADAI and the RADAI-SF had moderate accuracy to distinguish responders from non-responders (fig 2). The responders showed large improvements (SRM >0.80) of disease activity on both the RADAI and the RADAI-SF. The non-responders showed no improvements (SRM <0.20) of disease activity on both questionnaires. Change scores of the RADAI and RADAI-SF were moderately correlated with DAS28 change scores.

Figure 2 ROC curves for differences in RADAI and RADAI-SF scores between baseline and 3-month follow-up assessments using the EULAR response criteria as external criterion.

DISCUSSION

This study shows that the RADAI and its short form (RADAI-SF) have satisfactory reliability, validity and responsiveness among Dutch patients with RA starting with anti-TNF treatment. Omission of the tender joint count in the RADAI-SF did not harm the psychometric qualities of the RADAI. Results of the CFA even showed that the RADAI-SF provided the best fit of the data to a single-factor structure. The fit indices of the RADAI reflected a nearly acceptable model fit and met the recommended criteria after addition of an error covariance between items 3 (current amount of arthritis pain) and 5 (tender joint count). Correlated error terms may reflect either the omission of one or more relevant factors or the presence of overlapping item content.28 Since items 3 and 5 both asses current pain, the latter seemed most plausible. Leaving out the tender joint count, as earlier suggested,5 seems therefore justified and is recommended for research purposes. The tender joint count is the most time-consuming item to complete and omission of this item reduces the burden for the patient. In clinical practice, however, inclusion of the tender joint count might be useful as additional clinical information.

Of the single items of the RADAI, item 1 (global disease activity during the past 6 months) and item 4 (current duration of morning stiffness) had the lowest factor loadings. Apparently, these items contributed less to disease activity than the other items. The RADAI might be improved by modifications in the wording of these items. With regard to item 1, a shorter time frame, over which disease activity is measured, could be considered. In RA, self-reported health status is usually measured over a period of 1 week or 1 month. Measurement of global disease activity over a shorter period of time is more likely to be related to current disease activity than the measurement over a period of 6 months. With regard to item 4, replacement of stiffness duration by severity of stiffness, as earlier suggested,7 might be considered.

Results for the internal consistency and construct validity are in accordance with results reported in previous studies.5 6 The construct validity was demonstrated by correlating RADAI scores with DAS28 scores. A moderate correlation was found. Although the RADAI and the DAS28 are both intended to measure disease activity, higher correlations were not expected because of the different content of the measures. Where the RADAI consists of patient-assessed variables on signs and symptoms, the DAS28 primarily consists of physician-assessed and laboratory variables.

Only one study has previously reported on the responsiveness of the RADAI. Fransen et al have shown that the RADAI is a responsive measure to increases in disease activity.7 In this study we focused on the responsiveness of the RADAI to decreases in disease activity, which is an important feature to consider, especially if the aim is to assess treatment response. Since both studies differed from each other with regard to study sample, treatment, responsiveness indices and external criterion of change, absolute responsiveness values cannot easily be compared with each other and should be interpreted with caution.22 29 30 Two remarkable differences between the studies have to be mentioned, however. Fransen et al found comparable responsiveness of the RADAI and the DAS28 and a high correlation between the change scores of both measures. In our study we found less responsiveness of the RADAI than of the DAS28 and a moderate correlation between the change scores of both measures. Although these differences might be attributed to differences in methodology, the direction of change over which responsiveness is measured might also be of influence. An increase in disease activity, which is an unpleasant experience, is more likely to draw a patient’s attention than a decrease in disease activity.31 Therefore, the RADAI might be more responsive for worsening of disease activity than for improvement of disease activity.

In this study the responsiveness of the RADAI was evaluated over a period of 3 months. Because item 1 refers to global disease activity during the past 6 months, this item cannot expected to be very responsive. The responsiveness of the RADAI might have been underestimated. Especially if the aim is to measure treatment response, shortening of the time frame of 6 months seems necessary.

A limitation of this study concerns the generalisability of the results. For this study we used a cohort of consecutive patients with RA starting with anti-TNF treatment. At baseline, all patients had high disease activity, and major improvement was expected after 3 months of treatment. Therefore, the results should be generalised with caution to the whole population of patients with RA and other treatments.

In conclusion, this study supports the reliability, validity and responsiveness of the Dutch version of the RADAI. Omission of the tender joint count in the RADAI-SF produces comparable results and is justified and recommended for research purposes. The tender joint count might be useful as additional clinical information in patient management. Whether the RADAI can be improved by modification of the time frame of item 1 (past disease activity) and by replacement of stiffness duration by severity of stiffness (item 4) needs to be investigated. Moreover, to support the interpretation of RADAI scores in clinical practice and research, criteria for identifying low and high levels of disease activity and treatment response should be established in future research.

Acknowledgments

We thank T van Gaalen, W Kievit and P Welsing for their contribution to the organisation of the study and data management We thank the following rheumatologists and research nurses for their assistance in patient recruitment and data collection: J Alberts, C Allaart, A ter Avest, P Barrera Rico, T Berends, H Bernelot Moens, K Bevers, C Bijkerk, A van der Bijl, J de Boer, E Bos, B Botha, A Branten, F Breedveld, H van den Brink, G Bruyn, H Cats, M Creemers, J Deenen, C De Gendt, K Drossaers-Bakker, A van Ede, A Eijsbouts, S Erasmus, M Franssen, I Geerdink, M Geurts, E Griep, E de Groot, C Haagsma, J Harbers, A Hartkamp, J Haverman, H van Heereveld, A van de Helm-van Mil, I Henkes, S Herfkens, M Hoekstra, K van de Hoeven, F van den Hoogen, M Horbeek, P M Houtman, T Huizinga, H Hulsmans, T Jansen, M Janssen, M Jeurissen, A de Jong, Z de Jong, M Kleine Schaar, G Kloppenburg, H Knaapen, P Koelmans, M Kortekaas, B Kraft, A Krol, M Kruijssen, I Kuper, R Laan, J van de Laan, J van Laar, A Mooij, J Moolenburgh, N Olsthoorn, P van Oijen, M van Oosterhout, J Oostveen, P van’t Pad Bosch, K Rasing, K Ronday, D de Rooij, L Schalkwijk, A Spoorenberg, A Stenger, G Steup, W Swen, M Veerkamp, C Versteegden, H Visser, C Vogel, M Vonk, H Vonkeman, H van Wijk, N Wouters.

REFERENCES

Footnotes

  • Funding: This study was funded by an unrestricted educational grant by Schering-Plough and CVZ (the Dutch Health Care Insurance Board).

  • Competing interests: None.