Rheumatology Advance Access originally published online on October 25, 2005
Rheumatology 2006 45(3):308-313; doi:10.1093/rheumatology/kei150
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
A statistical analysis of the interrelationships between disease activity in different systems in systemic lupus erythematosus
Department of Statistical Science, University College London, 1 MRC Biostatistics Unit, University of Cambridge, 2 Centre for Rheumatology, Department of Medicine, The Middlesex Hospital, London and 3 Department of Rheumatology, Division of Immunity and Infection, The Medical School, University of Birmingham, Birmingham, UK.
Correspondence to: E. Allen, Mortimer Market Centre, off Capper Street, London WC1E 6AU, UK. E-mail: eallen{at}gum.ucl.ac.uk
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionAppendix 1Appendix 2References |
|---|
Objectives. To develop models for disease activity in patients with systemic lupus erythematosus (SLE) and to examine the hypothesis that possible subsets exist within the disease, notably renal disease and little else, mucocutaneous and musculoskeletal disease in isolation and more multisystem disease.
Methods. Four hundred and forty patients with SLE were followed for a period of 10 yr. Socio-demographic data were obtained at the first visit with disease activity being recorded at subsequent visits and damage scores at 6-monthly intervals. Prognostic factors for active disease in each of the mucocutaneous, musculoskeletal and renal systems were examined statistically. The results were then validated using data collected over 5 yr on a further 295 SLE patients from a different centre.
Results. Logistic regression analyses indicated that for all three systems studied a patient known to have an involvement in that system is more likely to present with active disease in that same system than a patient with no known prior involvement. Patients with a higher frequency of clinic visits with active disease in a system are more likely to represent with active disease than those with fewer visits. The results suggest that renal disease is most likely to occur on its own. Associations between activity in the mucocutaneous and musculoskeletal systems support the suggestion that patients with musculoskeletal and mucocutaneous disease alone represent a possible subset of SLE. None of the associations identified were modified by the medication a patient received.
Conclusions. Previous disease history and involvement of other systems determine a patient's chance of developing further episodes of active disease in SLE.
KEY WORDS: Systemic lupus erythematosus, Renal disease, Musculoskeletal disease, Mucocutaneous disease
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1Appendix 2References |
|---|
Systemic lupus erythematosus (SLE) is a multisystem autoimmune rheumatic disease that tends to remit and relapse despite therapy. Renal involvement is of considerable prognostic importance but is less common than musculoskeletal involvement. The use of serological markers, in most instances autoantibodies, to reflect disease activity or predict clinical flare has produced variable results. Antibodies to
double-stranded DNA are a good case in point. Some studies demonstrate good clinical correlation [1, 2] whereas some groups have little evidence of such a correlation [3, 4]. In general, serological markers of disease activity have proved disappointing in predicting non-renal flares [5]. Observation of patients with SLE suggests, to some physicians, that subsets exist within the disease. Common subgroups of SLE appear to be patients where the disease is focused on the kidneys, patients with mucocutaneous and musculoskeletal disease and little else, and patients with disease involving more systems. The research presented here represents the first development of models for multiorgan/system disease activity in SLE. The analysis undertaken examines the interrelationships between disease activity in the mucocutaneous, musculoskeletal and renal organs/systems to assess the validity of the clinical impression and to identify those patients who are most at risk of experiencing an increase in disease activity.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1Appendix 2References |
|---|
The primary database available for analysis comprises information on patients from SLE clinics at two hospitals in Birmingham, the Queen Elizabeth Hospital and the City Hospital. The data were collected prospectively on 440 patients over a period of 10 yr with ethical approval and signed informed consent from the patients. A second data set from the Centre for Rheumatology, Department of Medicine at the Middlesex Hospital was used to validate the final models. The second data set was collected on 295 patients over a period of 5 yr.
Data collected at the first recorded clinic visit incorporated socio-demographic data including age, sex, race and disease duration. Disease assessment using the British Isles Lupus Assessment Group (BILAG) index was then completed at each clinic visit with individual and total end-organ damage scores assessed using the Systemic Lupus International Collaborating Clinics/American College of Rheumatology (SLICC/ACR) damage index, completed at approximately 6-month intervals. Immunosuppressive drug therapy was recorded at every visit. This covered exposure and dose changes for steroids, azathioprine, cyclophosphamide, cyclosporin A, methotrexate and, rarely, intravenous immunoglobulin and rituximab. Patients were expected to attend 3-monthly for routine clinic visits but patients with renal involvement attended up to monthly as clinically indicated. The median time between visits was in fact 90 days [interquartile range (IQR) 43–105]. The median number of visits contributed by a patient was 11 (IQR 4–22).
The BILAG index [6] is a comprehensive activity index developed according to the principle of the ‘physicians’ intention to treat’. The index allocates separate alphabetical scores to each of eight organs and/or systems using the following ratings: A is the most active disease state requiring major immunosuppressive drugs; B, the patient is known to have active disease but does not require >20 mg prednisolone; C, the patient has relatively mild disease requiring only symptomatic therapy such as simple analgesics; D, there is no activity in this system now; E, no evidence of activity now or previously [6]. For the purposes of comparison, the BILAG index can be converted into a global score (A grade = 9 points, B = 3, C = 1, D and E = 0) [7].
Strong correlations have been shown between a BILAG global score and other measures designed as global score systems, notably the SLE Activity Measure (SLAM) and the SLE Disease Activity Index (SLEDAI) [8].
For the purpose of this analysis, disease activity in SLE will be regarded as a two-state process classified as active if the patient scores a BILAG A or B and inactive if the patient scores a BILAG C, D or E. (A BILAG score of C was categorized as inactive as a patient with a B or A score preceded by a C score is considered to have flared [9], and such disease activity does not respond to an increase in immunosuppressive therapy.)
The SLICC/ACR damage index was developed by the Systemic Lupus International Collaborating Clinics and adopted by the American College of Rheumatology as a valid measure of damage in SLE [10]. The index includes descriptors in 12 systems and damage is only considered if present for at least 6 months.
Statistical analysis
The primary medical focus of this work is the identification of those factors that affect a patient's chance of developing active disease. Consequently the objective of the analyses undertaken is the modelling of the transition probabilities between inactive and active disease, as defined by the BILAG index, in the specified individual systems at each clinic visit.
The analyses were based on logistic regression models with the dichotomous outcomes for each system being the grouped version of the BILAG index, i.e. active and inactive disease. As the objective of the analysis is to establish predictors of active disease, all models were based on a subset of visits where the patient was known to have had inactive disease at the previous visit. Thus a patient is included in the analysis each time they present with inactive disease at a visit and the models were therefore fitted using the generalized estimating equations approach of Liang and Zeger [11] to adjust for any heterogeneity not captured by conditioning on the previous visit. An example of the models used can be found in Appendix 1.
Explanatory variables for each patient included binary variables indicating whether or not a patient was known to have a history of disease in that system and a variable representing the number of times the patient had been observed with a BILAG score of A or B (separate explanatory variables were used for A and B scores) in the mucocutaneous, musculoskeletal and renal systems. These ‘counts’ were included in the models to ascertain whether or not the frequency of a patient presenting with active disease, as well as the simple knowledge that the system was active at least once in the past, affects their chance of re-presenting with active disease. A binary variable was included in all models indicating whether the patient is known to have had active disease in any of the other (not the mucocutaneous, musculoskeletal and renal systems), systems in the past. Binary variables were used to represent the presence or absence of damage in each of the systems.
Other explanatory variables that were considered included the time between clinic visits, time since the last visit, time since the patient's first recorded visit, time since disease onset, where appropriate the time since the last visit at which the patient presented with disease activity in the system under consideration and the time since the last visit at which the patient presented with disease activity in any system, and age, sex and age at disease onset.
In addition explanatory variables reflecting treatments a patient is receiving were included in all final models in order to establish whether any relationships found were modified by the addition of these variables. Variables were created that indicated what medication a patient was receiving at the previous clinic visit. For this purpose: four groups of treatments were considered, non-steroidal anti-inflammatory drugs (NSAIDs), steroids, immunosuppressants and other drugs.
In starting from a situation where the patient is known to have presented with inactive disease in all systems, the analysis undertaken considers three outcomes: active mucocutaneous disease, active renal disease and active musculoskeletal disease. These were chosen to represent the types of lupus involvement most commonly seen in our (rheumatology) clinics and cover a broad spectrum of lupus manifestations. These three outcomes incorporate a number of possible states: active disease in one of the systems only or concurrent active disease in more than one of the three systems. Preliminary analyses suggested, however, that there were no features of disease history that predisposed a patient to present with active disease in more than one of the three systems.
A number of different analyses were undertaken in order to arrive at the choice of variables included in the final models. Results from univariate analyses and full details of the variable selection process can be found in Allen [12]. The choice of variables to be included in the final models was initiated by a backwards selection with a significance level of 5%. Consideration was then given to the inclusion of interaction terms and finally previously excluded variables were checked to see whether they added to the final model. The model-building decisions are checked in the validation of the models using a second data set. Given the complexity of the results and the number of significant interactions, we have presented the results from each analysis in terms of odds ratios associated with combinations of factors. This leads to the necessary inclusion of some results that do not appear to be significant.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1Appendix 2References |
|---|
Four hundred and eleven of the Birmingham patients were women (93.4%), 277 (63.0%) Caucasian, 72 (16.4%) Afro-Caribbean and 67 (15.2%) Asian. Median age at the time of the last visit was 40.5 yr (range 16.3–81.6 yr), median follow-up was 3.8 yr (range 0.1–11 yr) and median disease duration was 12 yr (range 0–60 yr). Two hundred and thirty (56%) patients had active mucocutaneous disease, 262 (63.7%) had active musculoskeletal disease and 124 (30.1) had active renal disease. The number of visits at which these patients presented with active disease having had inactive disease in all systems at the previous visit is given in Table 1 along with the number of these patients who had damage at their first visit.
|
A second data set from the Centre for Rheumatology, Department of Medicine at the Middlesex Hospital was used to validate the three final models. The data were collected from 295 patients over a period of 5 yr. The two patient cohorts are similar, with this second data set comprising 265 (89.8%) women, 190 (64.4%) Caucasians, 52 (17.6%) Afro-Caribbeans and 32 (10.8%) Asians. The number of visits at which these patients presented with active disease having had inactive disease in all systems at the previous visit is given in Table 2 along with the number of these patients who had damage at their first visit.
|
Active mucocutaneous disease
The results of the multivariate analysis of predictors for active mucocutaneous disease are presented in Table 3.
|
A history of mucocutaneous B scores and a history of mucocutaneous A scores both increase a patient's chance of presenting with active mucocutaneous disease. In addition, the significance of the ‘number of mucocutaneous B scores’ indicates that a patient with a greater number of clinic visits with active mucocutaneous disease is more likely to develop active mucocutaneous disease than one with fewer. However, for those patients with a history of either musculoskeletal A or B scores the effect of a history of mucocutaneous A scores is reduced, and for patients with no history of mucocutaneous A scores a history of either musculoskeletal A or B scores has little effect.
Patients with a history of renal A scores have a decreased chance of presenting with active mucocutaneous disease. A history of renal B scores also decreases the chance of a patient presenting with active mucocutaneous disease as long as the patient has no history of musculoskeletal B scores. The longer a patient has been registered at the clinic the less likely they are to present with active mucocutaneous disease.
Active renal disease
The results of the multivariate analysis of predictors for active renal disease are presented in Table 4. A history of renal activity increases a patient's chance of presenting with active renal disease subsequently, and those patients with renal damage have an even greater chance of presenting with active renal disease. In addition, the significance of the ‘number of renal B scores’ indicates that a patient with a greater number of clinic visits with active renal disease is more likely to develop active renal disease than one with fewer visits.
|
A history of musculoskeletal B scores decreases a patient's chance of presenting with active renal disease.
The greater the time since the patient's previous visit the more likely they are to present with active disease. However, the longer the patient goes without presenting with active disease the less likely they are to do so.
Active musculoskeletal disease
The results of the multivariate analysis of predictors for active musculoskeletal disease are presented in Table 5.
|
A history of active musculoskeletal disease increases a patient's chance of presenting subsequently with active musculoskeletal disease. In addition, the significance of the ‘number of musculoskeletal B scores’ indicates that a patient with a greater number of clinic visits with active musculoskeletal disease is more likely to develop active musculoskeletal disease than one with fewer visits.
A history of renal B scores and a history of disease activity in all systems not chosen for detailed analysis decrease a patient's chance of presenting with active musculoskeletal disease.
A history of mucocutaneous A scores increases the chance of patients with a history of musculoskeletal A scores presenting with active musculoskeletal disease. However, for those patients with no history of musculoskeletal A scores the chance of presenting with active musculoskeletal disease is decreased by a history of mucocutaneous A scores.
Patients with both musculoskeletal damage and a history of mucocutaneous B scores have an increased chance of presenting with active musculoskeletal disease. Neither factor alone has any effect.
A patient's chance of presenting with active musculoskeletal disease decreases with the time since they last scored a musculoskeletal A or B, decreases with the time since the first clinic visit but increases with the time since the previous visit.
Validating the models
The three logistic models were validated on the second cohort using the goodness of fit test proposed in [13]. All three models had reasonable fit with P values of 0.103 for the mucocutaneous model, 0.244 for the renal model and 0.166 for the musculoskeletal model. There are some differences, with the main difference being that in the renal model all the effects are smaller if the patient is from the Centre for Rheumatology at the Middlesex Hospital [12]. In addition the protective effect of a renal A in the analysis of active mucocutaneous disease disappears. However, the majority of the findings from the first data set are qualitatively supported by the analysis of this second data set and the differences that do exist may simply reflect differences in patient management. Coefficients from the models fitted using this second data set are given in Appendix 2.
Medication
Variables reflecting the medication a patient was receiving at the previous clinic visit were added to all final models in order to assess whether this had any influence on the observed effects. In all three systems none of the variables significantly affected the outcome or altered the effect of any of the other variables.
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1Appendix 2References |
|---|
This analysis shows that patients with a history of disease activity in any of the mucocutaneous, musculoskeletal or renal systems are more likely to continue to have active disease in that system compared to patients with no previous evidence of disease activity (in that system). In addition, patients with a greater number of clinic visits with active disease (in that system) are generally more likely to develop active disease (in that system) than those with fewer.
This analysis also supports the hypothesis that a possible subset of SLE is represented by patients with renal disease and little else. A history of renal activity appears to decrease a patient's chances of developing both active mucocutaneous and musculoskeletal disease. This analysis also suggests that a patient is less likely to present with renal activity if they have a history of active musculoskeletal disease. A patient's chance of presenting with renal activity is also increased if the patient has renal damage. This particular finding is possibly due to the fact that renal activity is often ‘clinically silent’. That is to say that evidence for renal activity is often only indicated by the presence of protein or small amounts of blood in the urine that are only detectable on dipstick testing. Likewise an increase in blood pressure does not cause symptoms in the early stages. The patient may on occasion, therefore, enter the stage of permanent damage before a subsequent flare in activity. This may also be related to the observation that a longer time between visits increases the chance of active renal disease. This is important clinically as patients need to be reviewed regularly even if they are ‘asymptomatic’. From this analysis, it is therefore apparent that renal disease can occur on its own, with no evidence that disease activity or damage in either of the other two systems increases the chance of active renal disease.
There also appear to be a number of associations between activity in the musculoskeletal and mucocutaneous systems, supporting the suggestion that patients with predominantly musculoskeletal and mucocutaneous disease represent a possible subset of SLE. The analysis suggests that if a patient has a history of mucocutaneous activity and musculoskeletal damage their chance of presenting with active musculoskeletal disease is increased, and that patients with a history of severe mucocutaneous and musculoskeletal disease have an increased chance of developing active musculoskeletal disease. An association between BILAG A scores and an increased risk of damage has been previously shown [14], but this is the first time that we have shown that damage can predict active disease in the same system (both for renal and musculoskeletal disease).
Further associations between activity in these two systems were also identified. It appears that patients with a history of severe (BILAG A scores) mucocutaneous activity are less likely to develop subsequent active mucocutaneous disease if they have a history of musculoskeletal activity. It is also evident that patients who have previously scored a BILAG A in the mucocutaneous system are less likely to develop active musculoskeletal disease as long as they have not previously scored a BILAG A in the musculoskeletal system.
The length of time that a patient has been registered at a clinic is associated with a reduction in both mucocutaneous and musculoskeletal activity. However, a patient's chance of developing active musculoskeletal or renal disease increases with the time since the last visit but decreases with time since the patient was last observed to score an A or B in the respective system. As was previously mentioned, the inclusion of the patient's medication at the previous visit in the models did not affect these findings.
Although statistical models have been used previously in the assessment of disease actually [15, 16] this is the first attempt known to us to develop models for multiorgan/system disease activity in SLE. The model development must therefore be viewed as somewhat exploratory and confirmation of the findings with data from other sources would certainly be useful.
|
The authors declare no conflicts of interest.
|
Appendix 1 |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1Appendix 2References |
|---|
For a typical patient the models used were of the form:
 |
where
 |
x1t, x2t, ... , xkt are explanatory variables observed at times t = 1, ... , ni (ni is the number of times the ith patient has inactive disease in all systems and then a subsequent visit), yit = 1 if the disease is active in the patient at time t and yit = 0 otherwise.
|
Appendix 2 |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1Appendix 2References |
|---|
|
|
|
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1Appendix 2References |
|---|
- Lloyd W, Schur PH. Immune complexes, complement and anti-dna in exacerbations of systemic lupus erythematosus (SLE). Medicine 1981;60:208–17.[Medline]
- Isenberg DA, Garton M, Reichlin MW, Reichlin M. Long term follow up of autoantibody profiles in black female lupus patients and clinical comparison with Caucasian and Asian patients. Br J Rheumatol 1997;36:229–33.
[Abstract/Free Full Text] - Esdaile JM, Joseph L, Abrahamovicz M, Li Y, Danoff D, Clarke AE. Routine immunologic tests in systemic lupus erythematosus: is there a nee for more studies? J Rheumatol 1996;23:1842–4.[ISI][Medline]
- Esdaile JM, Abrahamovicz M, Joseph L, Mackenzie T, Li Y, Danoff D. Laboratory tests as predictors of disease exacerbations in systemic lupus erythematosus. Arthritis Rheum 1996;39:370–8.[ISI][Medline]
- Ho A, Barr SG, Magder LS, Petri M. A decrease in complement is associated with increased renal and haematologic activity in patients with systemic lupus erythematosus. Arthritis Rheum 2001;44:2350–7.[CrossRef][ISI][Medline]
- Hay EM, Bacon PA, Gordon C et al. The BILAG index: a reliable and validated instrument for measuring clinical disease activity in systemic lupus erythematosus. Q J Med 1993;86:447–58.[ISI][Medline]
- Morrow J, Nelson L, Watts R, Isenberg D. Autoimmune rheumatic disease. Oxford: Oxford University Press, 1999;66–7.
- Gladman DD, Goldsmith CH, Urowitz MB, Bacon P, Bombardier C, Isenberg D. Cross cultural validation of 3 disease activity indices in systemic lupus erythematosus. J Rheumatol 1992;19:608–11.[ISI][Medline]
- Gordon C, Sutcliffe N, Skan J, Stoll T, Isenberg DA. Definition and treatment of lupus flares measured by the BILAG index. Rheumatology 2003;42:1372–9.
[Abstract/Free Full Text] - Gladman DD, Urowitz MB, Goldsmith CH. The reliability of the Systemic Lupus International Collaborating Clinics/American College of Rheumatology damage index in patients with systemic lupus erythematosus. Arthritis Rheum 1997;40:809–13.[ISI][Medline]
- Liang K-Y, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika 1986;73:13–22.
[Abstract/Free Full Text] - Allen E. Multi-organ system rheumatological disease: statistical analysis of outcome measures and their interrelationships. PhD Thesis, University of London, 2003:96–141.
- Horton NJ, Bebchuk JD, Jones CL et al. Goodness of fit for GEE: an example with mental health service utilization. Statist Med 1999;18:213–22.
- Stoll T, Sutcliffe N, Mach J, Klaghofer R, Isenberg DA. Analysis of the relationship between disease activity and damage in patients with systemic lupus erythematosus: a 5-yr prospective study. Rheumatology 2004;43:1039–44.
[Abstract/Free Full Text] - Yen JC, Abrahamowicz M, Dobkin PL, Clarke AE, Battista RN, Fortin PR. Determinants of discordance between patients and physicians in their assessment of lupus disease activity. J Rheumatol 2003;30:1967–76.[Medline]
- Chang ER, Abrahamowicz M, Ferland D, Fortin PR, Organ manifestations influence differently the responsiveness of 2 lupus disease activity measures, according to patients' or physicians' evaluation of recent lupus activity. J Rheumatol 2002;29:2350–8.[ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. James, X. Kim-Howard, B. Bruner, M. Jonsson, M. McClain, M. Arbuckle, C. Walker, G. Dennis, J. Merrill, and J. Harley Hydroxychloroquine sulfate treatment is associated with later onset of systemic lupus erythematosus Lupus, June 1, 2007; 16(6): 401 - 409. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||