Discussion
Clinical trials in SLE are challenging and frequently fail to meet their primary endpoint for various potential reasons. Both the heterogeneity of SLE disease manifestations and the small numbers of patients available for recruitment to clinical trials may contribute to this.20 Trials in SLE may also be restrictive in their inclusion criteria with regard to renal disease and, as such, lack a degree of external validity.21 Endpoint definitions have been consistently difficult to agree on, but there is a movement towards composite disease activity scores such as the SLE Responder Index after its successful employment in the phase 3 belimumab trials.22 Major concerns remain, with additional ‘noise’ caused by polypharmacy and traditionally high dose steroid use within SLE populations potentially contributing to trial failure.21 The MASTERPLANS consortium aims to develop early clinical predictive markers in SLE to help inform future trials and personalised medicine studies. In LN trials, several traditional poor prognosis markers are enriched as these patients often have a more severe disease phenotype. Knowledge of and stratification for such markers may improve the conduct of future trials. In clinical practice, it may be possible to employ such markers to inform the treatment strategy used and to improve overall treatment response rates.
Our results found a number of predictors of global lupus and renal-specific responses which are of interest when considering treating patients with SLE and LN. Importantly, predictors of global response at 6 months tended to be different to those that predicted renal outcomes over the same period. Disease activity on BILAG and damage on SDI were associated with global outcomes but were not predictive of renal outcomes. This observation is relevant to future LN trials as balancing non-renal manifestations may influence overall outcomes since trials assess both renal and non-renal changes in their outcome assessments.
LN disease duration of 2–4 years was associated with a decreased likelihood of achieving CRR and PRR at 6 months. This has also been shown by others, with longer lupus disease duration considered a negative predictor of achieving overall low disease activity, although not specifically renal outcomes.23 Longer disease duration may act as a surrogate for a more relapsing-remitting course of LN and also of course may link to some early renal damage that limits a patient’s ability to achieve stringent response targets.
Patients recruited from Latin America had a decreased likelihood of attaining CRR at 6 months compared with our Asian comparator group. Studies have consistently shown that patients from Hispanic backgrounds develop LN early and have more aggressive disease.24 25 This could be explained by socioeconomic factors and variable access to healthcare within the regions, however in a trial setting more consistent provision of therapy would tend to mitigate this. Latin America itself is very ethnically diverse with Caucasians, Mestizo, pure Amerindians and African-Latin Americans all recognised ethnic subgroups.26 Such consistent findings across outcomes do suggest that a complex interaction of factors influence LN outcomes in this region. Our study however lacked power to dissect this out in more detail. While Asian ethnicity is also diverse and is traditionally associated with severe renal disease,27 their response to treatment, long-term renal outcomes and renal survival rates appear to be better, particularly when compared with Hispanic populations.28 In the SLICC inception cohort, we previously found that Asian patients (from South Korea) had less progression to damage over time.29 These results point to potential organ-specific differences in responsiveness among patients from different racial and ethnic backgrounds. The potential prognostic role of ethnicity has also been considered previously in the literature comparing ALMS maintenance and MAINTAIN nephritis trials. Both trials assessed the efficacy of MMF for maintenance therapy, with the former suggesting MMF as superior for the treatment of LN and the latter suggesting no difference. MAINTAIN was a European study with a predominantly Caucasian population, whereas ALMS was an international study with more ethnic diversity (79% and 44% Caucasian, respectively).30 The superiority of MMF in the ALMS study may at least be partially explained by the ethnic background of those enrolled.31
Established damage at baseline was associated with a decreased likelihood of achieving global improvement by 6 months. Higher SDI scores at baseline increase the risk of mortality in patients with SLE.29 Established damage will reflect more severe previous disease and/ or higher chronic steroid exposure and will also be more prevalent in patients with longer disease duration. Activity (BILAG A or B) in haematological and mucocutaneous domains predicted less improvement which supports findings in the EXPLORER trial, where baseline BILAG mucocutaneous involvement was not predictive of treatment response.32 33
Considering haematological involvement, baseline BILAG A or B scores have been demonstrated to predict flares at 24 and 52 weeks34 in the phase III belimumab trials. While the endpoints in this analysis were different, those patients who are going to flare would be less likely to achieve improvement. Patients with higher baseline disease activity are also likely to be harder to treat and may require different therapeutic strategies. An increase in the numerical BILAG was also associated with a decreased likelihood of improvement at 6 months so overall more extensive disease even when using potent immunosuppression in LN is associated with poorer response rates. The ALMS induction trial2 has reported previously the efficacy of MMF and CYC in achieving good BILAG non-renal responses, with particularly promising improvement in BILAG index scores within the mucocutaneous (MMF 84% vs CYC 93%) and musculoskeletal (MMF 91% vs CYC 96%) at 24 weeks.11 This research was evaluating individual disease activity in individual systems but we have demonstrated when considering the patient overall, it is harder to achieve composite non-renal outcomes with only 50.81% achieving improvement at 6 months.
A previous study using this dataset found very few multivariate baseline predictors of renal response and/or renal remission.3 In contrast to the study by Dall’Era et al, the current study was focused on BILAG-based outcomes in this trial and assessed renal responses as well as overall SLE responses. Also, in contrast to Dall’Era et al,3 our renal endpoints of MCR and PRR did not set different response criteria based on whether the patient was nephrotic or not at baseline. Also comparing the ‘renal response’ definition to our equivalent PRR, we used a lower absolute value of urine P/Cr ratio of <100 mg/mmol rather than percentage reduction in proteinuria for subnephrotic patients. Our study therefore complements and adds to this previous analysis by also including overall SLE responses within the trial, which means we were also able to compare and contrast the factors that predict renal and overall SLE responses to show different factors associated with each.
Limitations
ALMS was considered a large global trial at its time but a sample size of 370 still limits our power to identify all important predictors of response in SLE. Trials with larger populations would provide more precision to predictor estimates. We focused on 6-month outcomes in this analysis and while 12-month data was available it was only available for those who showed a level of response at 6 months and that qualified them for re-randomisation. Data beyond 6 months for those not re-randomised was therefore not available.
The predictive performance of the clinical model examined, as shown by the AUROC results, was very modest and implies that any model combing these baseline factors will have a poor ability to predict treatment response. Our variable selection results do however show the relative predictive power of each factor compared with each other and help identify patient characteristics who respond better to conventional therapies. Taken together, our results emphasise the need to identify novel biomarkers that will improve the predictive accuracy for treatment response in patients with SLE over and above the modest performance of clinical factors alone. Urinary biomarkers have recently been demonstrated to predict treatment response to rituximab in LN at 6 months.35 Adding such factors into our models would likely further improve their predictive value. Continuing to identify such biomarkers remains the long-term aim of the MASTERPLANS Consortium.