A data-driven approach finds RNA polymerase III antibody and tendon friction rubs as enrichment tools for early diffuse scleroderma trials

Abstract

Objective

Clinical trials in early diffuse SSc have consistently shown a placebo group response with a declining modified Rodnan skin score (mRSS), with negative outcomes. Our objective was to identify strategies using clinical characteristics or laboratory values to improve trial design.

Methods

We identified early diffuse SSc patients first seen at the University of Pittsburgh from 1980–2015. Eligible patients had ≥3 visits, with at least two mRSS scores within the first year of follow-up. We performed Kaplan–Meier analyses, group-based trajectory analysis of mRSS scores, followed by multivariable regression analysis and classification tree analysis. We applied the results to the abatacept in early diffuse systemic sclerosis (ASSET) trial outcome data.

Results

We identified 403 patients with <18 months, and 514 with <36 months disease duration. The median number of mRSS follow-up scores was 14 (interquartile range 8, 25). All methodologic approaches identified skin thickness progression rate, RNA polymerase III (RNAP3) antibody positivity and presence of tendon friction rubs (TFR) as predictors of mRSS trajectory over 5 years of follow-up, and thereby as potential enrichment variables. When applied to the ASSET data, adjustment for both RNAP3 and TFR demonstrated reduction of the placebo mRSS response, particularly at 6 months. A significant difference in the ACR Composite Response Index in Systemic Sclerosis (CRISS) score was found with adjustment by RNAP3 at 6 months, and TFR or RNAP3 at 12 months.

Conclusion

Adjustment for both RNAP3 and TFR predicts mRSS trajectory and diminished the mRSS decline in ASSET placebo group, and identified significant differences in CRISS. RNAP3, particularly, is a stratification or enrichment approach to improve early diffuse SSc trial design.

Rheumatology key messages.

• Cohort enrichment for RNAP3 antibody status is a useful strategy in early diffuse scleroderma trials.

• Tendon friction rub presence at enrolment is an alternative cohort enrichment strategy.

• Enrichment or stratification with RNAP3 or TFR are particularly advantageous for 24-week endpoints.

Introduction

SSc is a multisystem autoimmune disease. Despite the highest case-specific mortality among the rheumatic diseases [1], the only US Food and Drug Administration (FDA) or European Medicines Agency (EMA) authorized therapies are for SSc-associated lung disease, which affects <50% of patients. Thus, there is no approved therapy available to most patients.

Patients with early dcSSc are at high risk for internal organ complications and mortality [2–5], creating a therapeutic window for intervention. A multitude of therapeutics have been tested in early diffuse SSc, but no publications show a statistically significant improvement in the modified Rodnan skin score (mRSS) [6, 7]. The mRSS has face, criterion and construct validity, with responsiveness to intervention demonstrated [6]. However, a decline in the mRSS in both the placebo and intervention arms, with ensuing negative clinical trials, has been observed. Recently, the ACR Composite Response Index in Systemic Sclerosis (CRISS) [8] has been used as an SSc outcome measure in SSc, although also with largely negative trial results. The ACR-CRISS is composed of five core set measures, the largest statistical contributing factor being the change in mRSS. Thus, failed trials with the mRSS or ACR-CRISS may be due to flawed trial design leading to enrolment of patients whose mRSS will decline based on natural history, thereby creating an mRSS ‘placebo effect’.

The FDA defines enrichment as ‘the prospective use of any patient characteristic to select a study population in which detection of a drug effect (if one is in fact present) is more likely than it would be in an unselected population’ [9]. We have previously published work suggesting that optimizing disease duration inclusion criteria may reduce the number of patients enrolled whose mRSS will decline based on natural history (placebo effect). Other investigators have suggested cohort enrichment strategies utilizing mRSS cut-offs [10] or inflammatory markers [11]. Attempts to enrich clinical trials using the latter two methods have been unsuccessful among recently published trials [12–14].

Serum autoantibodies are associated with phenotypes and prognosis in SSc. Three autoantibodies have been associated with diffuse SSc: anti-RNA polymerase III (RNAP3), anti-Scl-70 and anti-U3RNP [15]. Cohort data have shown that RNAP3+ patients present with higher mRSS scores, and progress to high peak mRSS scores [16]. While few clinical trials have fully analysed participant autoantibodies using gold standard methodology, recent early diffuse SSc trials recruiting in the USA have enrolled a predominance of RNAP3+ patients.

The objective of this project was to identify variables for cohort enrichment to improve early diffuse SSc clinical trial design and analysis. We hypothesized that RNAP3 positivity may be such a variable.

Methods

Identification of potential variables

We used the University of Pittsburgh Scleroderma Center observational cohort. We identified all diffuse SSc who presented for an initial SSc Center visit between 1 January 1980 and 28 February 2015 with <36 months of disease. Patients had ≥3 visits with a mRSS recorded, with at least one of the visits within 1 year of the first Pittsburgh visit. All patients provided written informed consent at the time of cohort enrolment, approved by the University of Pittsburgh Institutional Review Board. Patients were not directly involved in this research, as it used previously collected data. Race/ethnicity were self-reported.

At the first visit the date of onset of the first SSc manifestation (including Raynaud and non-Raynaud characteristic) was recorded. Definitions of organ involvement were as previously published [17]. All mRSS scores was recorded by one of only three clinicians (T.A.M., V.S., R.T.D.). At each visit the interval SSc-specific history, physical examination, objective imaging and SSc-related laboratory studies were recorded.

Autoantibody analysis

Serum from the baseline visit was analysed. First, immunofluorescence was performed for ANA pattern. ACA was identified by its characteristic staining on HEp-2 substrate. Historically, Ouchterlony immunodiffusion has been used to determine anti-PM/Scl and anti-Scl70. The remaining autoantibody identification for anti-RNAP3, anti-U1-RNP, anti-U3RNP, anti-U11RNP, anti-Ku, anti-RUV-BL1/2 and anti-Th/To were performed by immunoprecipitation [18]. This same approach was performed at the University of Pittsburgh on the ASSET trial samples.

Statistical analysis

We created two analysis data sets using <18 months and <36 months of disease duration at SSc Center presentation, given that both have been primary inclusion criteria for clinical trials. Our recent work supports <18 months as the preferred definition [19], and this was thus our primary development dataset.

We assessed the following first visit features as potential prognostic enrichment variables: autoantibody (RNAP3, Scl70, other), presence of palpable tendon or bursal friction rubs (TFR) [20], skin thickness progression rate (STPR) [21], sex, age, ESR (elevated vs normal using age-adjusted values) [22] and DMARD use.

First, we conducted Kaplan–Meier analyses and compared the difference in time to reach peak mRSS during follow-up using the log-rank test for each of these variables.

Second, we performed group-based trajectory analysis [23, 24] to examine longitudinal changes in mRSS scores for the two datasets (<18 months, <36 months). We first determined the number of mRSS trajectory groups using a group-based trajectory for <18 months duration, labelling the trajectory groups based on the mRSS at presentation (low/moderate or high) and if they were a mRSS decline or not (improver/non-improver) as presented below. We then conducted a series of trajectory analyses, i.e. modelling intercept only, linear, quadratic or cubic polynomial terms or varying the number of trajectory groups, until the best-fitting model was obtained as indicated by the Bayesian Information Criterion. We used the posterior probabilities of group membership from each individual to assess the model fit. High probability of membership into a single group represents a good model fit. We repeated the analysis for the <36 months cohort data.

Third, we used multivariable multinomial logistic regression with our candidate variables to identify significant predictors of mRSS trajectory grouping, which could then be potential stratification variables for trial analyses. All above analyses were performed with SAS (version 9.3 SAS Institute Inc., Cary, NC, USA).

Finally, we performed classification tree analyses (CTA) using SPSS to identify nodal factors which defined branch points leading to trajectory group membership. Each of the nodal points represented a potential predictive enrichment variable.

Application of cohort enrichment strategies in a completed interventional clinical trial

We evaluated the functionality of the identified potential enrichment variable using the abatacept in early diffuse cutaneous systemic sclerosis (ASSET) trial data. This dataset was chosen for the disease duration enrolled (<36 months) and primary outcome of change from baseline mRSS (ΔmRSS) at 12 months [25]. The University of Pittsburgh performed autoantibody profiling from stored ASSET samples.

Statistical analysis approach in the ASSET clinical trial data

We used a linear mixed model to include anti-RNAP3 and TFR separately to discern a differential effect on the outcome variable, ΔmRSS. In the original published ASSET analysis, a linear mixed model using ΔmRSS at multiple time points, with baseline mRSS, disease duration, treatment group, visit month, and interaction of treatment group and visit month as predictors [25] was fitted. We obtained the least square means (LSM) and 95% CI by treatment group, and calculated the group difference in LSM at months 6 and 12. Using the same method, we fitted models additionally adjusted for anti-RNAP3 or TFR, respectively, and obtained LSM (95% CI) and group difference in LSM as described. We used quantile regression to fit the CRISS score at month 6 (or month 12) as outcome, with disease duration, treatment group, visit month, and interaction of treatment group and visit month as predictors. We obtained adjusted median and 95% CI for CRISS score at month 6 (or month 12). Similarly, we then fitted models additionally adjusted for anti-RNAP3 or TFR, respectively, and obtained adjusted median (95% CI) and group difference in median. We performed similar analysis for the major statistical components, change in forced vital capacity (FVC) and HAQ-DI, of CRISS as well. All modelling used SAS (version 9.4 SAS Institute Inc., Cary, NC, USA).

Results

We identified 403 patients presenting within 18 months of SSc onset, and 514 within 36 months of SSc onset at the first visit. The mean age at the first Pittsburgh visit was 50 and 49 years, 76% and 74% were female, and 91 and 92% were White, respectively. The median follow-up was >10 years, with a median of 16 and 14 clinic visits, respectively (Table 1).

Table 1.

Pittsburgh first visit (baseline) characteristics

	SSc <18 months	SSc <36 months
	(n = 403)	(n = 514)
Age at first visit, mean (s.d.)	50.0 (13.8)	49.4 (13.8)
Female, n (%)	205 (76)	382 (74)
White, n (%)	370 (92)	468 (91)
Disease duration (years) from first non-Raynaud manifestation, median (IQR)	0.76 (0.55, 1.04)	0.90 (0.59, 1.39)
Follow-up time in years, median (IQR)	10.2 (4.6, 16.3)	10.2 (4.5, 16.3)
Number of clinic visits, median (IQR)	16 (9, 24)	14 (8, 23)
Cutaneous characteristics
mRSS at first visit, mean (s.d.)	25 (11)	25 (11)
Peak mRSS during follow-up, mean (s.d.)	32 (11)	31 (11)
Years to peak mRSS, median (IQR)	1.20 (0.88, 1.76)	1.39 (0.98, 2.17)
STPR, n (%)
Rapid	141 (35)	194 (38)
Intermediate	146 (37)	163 (32)
Slow	113 (28)	154 (30)
Non-cutaneous disease characteristics, n (%)
TFR present	182 (45)	231 (45)
Heartburn plus dysphagia or objective dysmotility	173 (43)	235 (46)
Lung fibrosis on chest imaging	77/334 (23)	103/440 (23)
Renal crisis	37 (9)	50 (10)
Pulmonary hypertension	0 (0)	9 (2)
Autoantibodies, n (%)
Anti-RNAP3	228 (57)	273 (53)
anti-Scl-70	106 (27)	137 (27)

IQR: interquartile range; mRSS: modified Rodnan skin score; STPR: skin thickness progress rate; TFR: tendon friction rub; RNAP3: RNA polymerase III.

The mean baseline mRSS was 25 in both, peaking at a mean of 32 for those presenting within <18 months, and 31 in those presenting within <36 months. Approximately one-third of patients fell into each of the rapid, intermediate or slow STPR categories at first visit as previously defined [21], and 45% had TFRs. In both groups 27% of patients were Scl-70 positive. In those presenting within 18 months of disease, 57% were RNAP3+; of those presenting with 36 months, 53% were RNAP3+.

Kaplan–Meier analysis

A statistically significant difference in time to peak mRSS between the autoantibody groups (P = 0.008), STPR (P < 0.0001) and TFR absence/presence (P = 0.02) was seen in the <36 months presentation cohort (supplementary Fig. S1, available at Rheumatology online). Patients with TFR present, RNAP3+ or rapid/intermediate STPR had a shorter time to peak mRSS. In the <18 months at presentation cohort similar results were found for STPR (P = 0.0003) and TFRs (P = 0.02), but not autoantibody (P = 0.15). The remaining candidate variables were not significant.

Trajectory analysis

Group-based trajectory modelling discovered patient groupings of mRSS trajectories over 5 years of clinical follow-up. Fig. 1 shows mRSS trajectory curves for the <18 months (Fig. 1A) and <36 months (Fig. 1B) cohorts. As described previously [19], the mRSS trajectory patterns appear different in those with early (<18 months) presentation from when longer disease duration (<36 months) is included. In the <18 months cohort, we identified five trajectory groups: (i) the low mRSS-improver (red, 28%), (ii) low mRSS-progressor group (black, 11%), (iii) moderate mRSS-rapid improver (green, 24%), (iv) high mRSS-rapid improver (blue, 21%) and (v) high mRSS-slow improver (yellow, 16%). In the <36 months cohort, a new group with a high mRSS which does not improve appears, comprised of 9% of patients (Fig. 1B, top red line).

Figure 1.

mRSS trajectory patterns for patients.

(A) The five mRSS trajectory groups over 5 years of follow-up in those presenting with <18 months of disease. The corresponding colour % represents the percentage of patients falling into that trajectory. (B) The six mRSS trajectory groups that appear in patients presenting with <36 months of disease. Similarly, the corresponding number depicts the percentage of patients falling into that trajectory. mRSS: modified Rodnan skin score

Regression analysis

We used multivariable multinomial logistic regression modelling (Table 2) to assess for factors predicting mRSS trajectory groupings; the reference group was the low MRSS-improver coloured red in Fig. 1A. The colour of the heading in Table 2 corresponds to the group trajectory colour in Fig. 1A. This revealed that STPR group, TFR presence and RNAP3 positivity were trajectory predictors. Similar results were found for <36 months cohort (supplementary Table S1, available at Rheumatology online), with elevated ESR also a significant predictor of mRSS.

Table 2.

Multivariate analysis for predictors of skin thickness progression trajectories

Parameter	Moderate mRSS-rapid improver,	High mRSS-rapid improver,	Low mRSS progresser,	High mRSS-slow improver,	Overall P-value
	n = 69 (24%)	n = 60 (21%)	n = 31 (11%)	n = 47 (16%)
	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
Age in years					0.42
<35	0.61 (0.18, 2.96)	0.45 (0.10, 1.92)	2.42 (0.48, 12.19)	1.13 (0.25, 5.10)
35–44	0.294 (0.10–0.84)	0.41 (0.13,1.33)	1.48 (0.36, 6.05)	0.88 (0.24, 3.14)
45–54	0.66 (0.247,1.747)	0.90 (0.31, 2.61)	1.50 (0.3, 6.05)	0.75 (0.22, 2.59)
55–64	0.412 (0.14, 1.24)	0.88 (0.28. 2.80)	0.94 (0.192, 4.645)	0.65 (0.15, 2.59)
Female sex	1.208 (0.575, 2.538)	1.53 (0.67, 3.51)	2.64 (0.94, 7.41)	1.45 (0.58, 3.68)	0.43
Non-White	1.02 (0.32, 3.30)	0.79 (0.20, 3.16)	0.22 (0.03, 1.91)	1.55 (0.39. 6.08)	0.57
RNAP3+	3.034 (1.494, 6.161)	2.05 (0.92, 4.51)	2.05 (0.92, 4.51)	2.05 (0.92, 4.51)	0.01
TFR present	1.07 (0.536, 2.153)	3.01 (1.43, 6.35)	3.01 (1.43, 6.35)	3.01 (1.43, 6.35)	0.001
STPR					<0.001
Intermediate STPR	2.825 (1.33, 6.00)	6.73 (2.52, 17.95)	0.94 (0.192, 4.645)	41.27 (5.10, 334.28)
Rapid STPR	2.35 (1.01, 5.46)	14.45 (5.36, 38.98)	2.42 (0.48, 12.19)	98.63 (12.6, 806.97)
Elevated ESR	1.17 (0.53, 2.59)	1.50 (0.62, 3.64)	2.64 (0.94, 7.41)	0.87 (0.33, 2.30)	0.41
Lung fibrosis on imaging	0.47 (0.18, 1.28)	0.39 (0.14, 1.07)	0.22 (0.03, 1.91)	0.50 (0.16, 1.62)	0.37
MMF use	1.75 (0.70, 3.91)	1.51 (0.62, 3.68)	0.75 (0.31, 1.81)	0.94 (0.33, 2.70)	0.51
MTX use	1.15 (0.49, 2.68)	1.95 (0.82, 4.68)	1.48 (0.36, 6.05)	2.93 (1.15, 7.50)	0.14

mRSS: modified Rodnan skin score; OR: odds ratio; RNAP3: RNA polymerase III; TFR: tendon friction rub; STPR: skin thickness progress rate. The bold type is to indicate the significant values.

CTA modelling generates branch points, which we used to identify potential stratification or enrichment variables (supplementary Fig. 2, available at Rheumatology online). In the <18 months group, we identified the branch points of STPR (slow vs rapid/intermediate), followed by TFR, gender, RNAP3+/RNAP3– and age groups (<35, 35–64, >64 years) to predict all mRSS trajectories. The area under the curve (AUC) for the receiver operating characteristic (ROC) was 0.70 (Fig. 2). In the <36 months cohort major branching points were slow vs rapid/intermediate STPR and TFR with an AUC ROC of 0.63 (supplementary Fig. 3, available at Rheumatology online).

Figure 2.

mRSS change in the ASSET trial re-analysed by RNAP3 and TFR presence.

(A) The original ASSET trial analysis comparing abatacept (dotted line) and placebo (solid line). (B) mRSS change stratified by RNAP3 positivity (solid line) and negativity (dotted line). (C) mRSS change adjusted for RNAP3: placebo/RNAP3+ (solid dark), abatacept/RANP3– (solid dotted), abatacept/RNAP3 (solid grey) and abatacept/RNAP3– (dotted grey line). (D) mRSS change stratified by TFR presence (solid) and absence (dotted line). (E) mRSS change adjusted for TFR: placebo/TFR+ (solid dark), abatacept/TFR+ (dotted dark), placebo/TFR absence (solid grey), abatacept/TFR absence (dotted grey). All analyses are adjusted for the stratum (disease duration ≤18 months vs 18–36 months). mRSS: modified Rodnan skin score; ASSET: abatacept in early diffuse systemic sclerosis; RNAP3: RNA polymerase III; TFR: tendon friction rubs

Summary of results

Through multiple modelling methods we identified STPR, presence of TFR and RNAP3 antibody as potential variables for enrichment in early diffuse SSc trial population. As the method to calculate STPR depends on the first evaluation [21], STPR has accuracy and feasibility limitations in the trial setting. We therefore evaluated using TFR presence and RNAP3 positivity as enrichment tools in real-world clinical trial data.

ASSET trial mRSS validation exercise

The ASSET trial population has been previously described [25] and consisted of 88 patients randomized to receive abatacept (n = 44) or placebo (n = 44). Of these 88 patients, 32(36%) had TFRs present at baseline, and 39 of 85 (45%) were RNAP3+. No background immunosuppressive therapy was allowed. Thus, the ASSET trial placebo group demonstrates the natural history of mRSS trajectories in early diffuse SSc patients. Fig. 2 depicts the ΔmRSS between the abatacept and placebo groups from the original analysis (Fig. 2A), as well as stratification by RNAP3+ (Fig. 2B) and TFR (Fig. 2D) independent of treatment group. This reveals significant separation in the LS mean ΔmRSS at 6 months between RNAP+ and RNAP– patients (P = 0.02), and similarly in TFR presence/absence (P = 0.07). This separation persists at 12 months, although it was not statistically significant.

In Fig. 2C, LS mean Δ mRSS is shown stratified by treatment assignment and RNAP3 positivity. What is striking in Fig. 2C is that all groups having a similarly declining mRSS, except for RNAP3+ patients in the placebo group, which have minimal mRSS change. This visually supports and provides face validity to the concept of stratification by RNAP positivity at enrolment in an early diffuse SSc clinical trial. Similarly, Fig. 2E adds TFR presence/absence to treatment assignment. Here a similar pattern appears, with those TFR+ patients in the placebo group having a visually obvious minimal decline in mRSS.

Further analysis is shown in Table 3. In the first section, adjustment for TFR and RNAP3 to the original ASSET trial manuscript is presented. When TFR presence or RNAP3 positivity was adjusted for in the analysis for the primary outcome (change in mRSS from baseline), we observed mitigation of the placebo effect, with a decrease in the mean ΔmRSS observed in the placebo group. This is larger at 6 months than 12 months.

Table 3.

Change in mRSS in the ASSET trial population

Change in mRSS between abatacept and placebo groups at 6 and 12 months with and without adjustment for TFR and RNAP3
		LS median change		Abatacept – placebo	P-value
		Abatacept	Placebo
		(n = 44)	(n = 44)
6 months	Original published analysis	–3.76	–3.00	–0.77	0.59
	Adjusted for TFR	–3.67	–2.20	–1.47	0.32
	Adjusted for RNAP3	–3.81	–2.49	–1.32	0.37
12 months	Original published analysis	–6.24	–4.49	–1.75	0.28
	Adjusted for TFR	–6.13	–4.06	–2.07	0.23
	Adjusted for RNAP3	–6.20	–4.15	–2.04	0.23

Difference in mRSS change (LSM) stratified by TFR presence
		LS median change	P-value
		Difference in mRSS
6 months	TFR+ vs TFR– (abatacept)	1.25 (–2.94, 5.45)	0.56
	TFR+ vs TFR– (placebo)	4.25 (–0.003, 8.50)	0.05
	Abatacept vs placebo (TFR+)	–2.97 (–7.60, 1.66)	0.21
	Abatacept vs placebo (TFR–)	0.02 (–3.63, 3.67)	0.99
12 months	TFR+ vs TFR– (abatacept)	1.56 (–3.10, 6.21)	0.51
	TFR+ vs TFR– (placebo)	2.59 (–2.61, 7.78)	0.33
	Abatacept vs placebo (TFR+)	–2.59 (–8.27, 3.09)	0.37
	Abatacept vs placebo (TFR–)	–1.56 (–5.52, 2.40)	0.44

Difference in mRSS (LSM) changed stratified for RNAP3
		LS median change difference in mRSS	P-value
6 months	RNAP3+ vs RNAP3– (abatacept)	2.85 (–1.36, 7.07)	0.18
	RNAP3+ vs RNAP3– (placebo)	4.52 (0.43, 8.62)	0.03
	Abatacept vs placebo (RNAP3+)	–2.15 (–6.30, 2.00)	0.31
	Abatacept vs placebo (RNAP3–)	–0.49 (–4.52, 3.54)	0.81
12 months	RNAP3+ vs RNAP3– (abatacept)	–0.10 (–4.76, 4.55)	0.97
	RNAP3+ vs RNAP3– (placebo)	4.30 (–0.53, 9.12)	0.08
	Abatacept vs placebo (RNAP3+)	–4.24 (–9.10, 0.61)	0.09
	Abatacept vs placebo (RNAP3–)	0.16 (–4.38, 4.69)	0.95

mRSS: modified Rodnan skin score; ASSET: abatacept in early diffuse systemic sclerosis; RNAP3: RNA polymerase III; TFR: tendon friction rub; LS: least squares; LSM: least square means. The bold type is to indicate the significant values.

In the second and third sections of Table 3, the difference in LSM change in mRSS is examined by TFR and RNAP3 status, with adjusted analysis completed. It is evident that the largest decline in median ΔmRSS occurs between the abatacept and placebo group who are either TFR+ or RNAP3+. This further supports, using another analytic method with clinical trial data, that enrichment for RNAP3+ patient or TFR+ patients may increase the chance of detecting a difference between intervention group outcomes.

ASSET trial analysis of ACR CRISS endpoint

As the ΔmRSS is the most influential component of the ACR CRISS score calculation, we examined the effect of adjustment for RNAP3 and TFR on the observed change in CRISS at 6 and 12 months. This analysis was felt to be important for examining real-world applicability, as the ACR CRISS is the primary outcome measure in most current enrolling early diffuse SSc trials. Table 4 shows the median change in CRISS at 6 and 12 months in the original ASSET analysis. All analyses were stratified by disease duration (<18 months, 18–36 months) per the original trial design.

Table 4.

Change in ACR-CRISS with adjustment for TFR and RNAP3 in the ASSET trial

	CRISS at 6 months		CRISS at 12 months
	Abatacept – placebo Median change (95% CI)	P-value	Abatacept – placebo Median change (95% CI)	P-value
Original published analysis	0.02 (–0.25, 0.30)	0.88	0.62 (0.07, 1.17)	0.03
Adjusted for TFR	0.03 (–0.26, 0.32)	0.86	0.57 (0.09, 1.06)	0.02
Adjusted for RNAP3	0.30 (0.02, 0.59)	0.04	0.63 (0.13,1.13)	0.01

ACR-CRISS: ACR Composite Response Index in Systemic; RNAP3: RNA polymerase III; TFR: tendon friction rub; ASSET: abatacept in early diffuse systemic sclerosis. The bold type is to indicate the significant values.

At 6 months, adjustment for RNAP3 uncovered a significant median change in the CRISS score between the treatment and placebo groups that was not present without adjustment. At 12 months, adjustment for RNAP3 reveals a higher median change in ACR CRISS and more significant P-value the original analysis.

Analysis of the effect of adjustment and stratification approaches for TFR and RNAP3 on change in FVC and HAQ Disability Index (HAQ-DI), subcomponents of the CRISS calculation, are presented in supplementary Tables S2 and S3, available at Rheumatology online. These analyses demonstrate that change in FVC between the abatacept and placebo group at 6 months was –1.4% in RNAP3+ (P = 0.56) and 6.0% in those who were RNAP3– (P = 0.01). A similar difference was seen at 12 months with change of 0.3% (RNAP3+, P = 0.92) and 4.5% (RNAP3–; P = 0.06). A similar pattern of change in HAQ-DI was seen stratified by RNAP3 positivity with difference between abatacept and placebo groups being significantly different, particularly at 12 months. This component analysis suggests that RNAP3+ may be used as an enrichment factor for these components of the CRISS score, and help to account for the apparent effect beyond the mRSS. Results assessing the effect of stratification for TFR and RNAP3 status are in supplementary Table S4, available at Rheumatology online.

Overall, these results support the concept that in trials with ACR CRISS as an endpoint, adjusting for RNAP3 may improve our ability to detect a true difference between the treatment and placebo groups.

Discussion

In our work, multiple analytic methods applied to the Pittsburgh scleroderma cohort consistently identified STPR, TFR presence and RNAP3 positivity as factors that predicted mRSS trajectory pattern in early diffuse SSc. Each of these three variables is appropriate as a prognostic enrichment approach in early diffuse SSc clinical trial design. RNAP3 is an objective test that can be standardized across a trial population while TFR requires examiner experience and skill. STPR was developed to be used at the initial patient presentation, making it the least applicable option in a multicentre clinical trial.

Studying the placebo group behaviour in the ASSET trial data depicts mRSS natural history, as these patients were not on immunosuppression. In both RNAP3– and those without TFRs, we observed a decline in mean mRSS scores from the trial start. The inclusion of RNAP3– and TFR– patients has likely contributed to the observed mRSS placebo response, thereby contributing to negative clinical trial outcomes for previous therapeutic agents.

Interestingly in the ASSET trial placebo group data, and in accordance with observational cohort studies, RNAP3+ patients show a short-interval increase in their mRSS. This concentration in the RNAP3+ patients demonstrating a short-interval increase may be the defining feature that enables RNAP3+ to be an effective stratification technique in trial design specifically at the 6-month trial endpoint.

When applied to a real-world clinical trial data set (ASSET), we confirmed that stratifying by TFR presence or RNAP3 positivity distinguished groups of patients whose mRSS improved over time, irrespective of treatment assignment. Adjusting for TFR presence functioned well, mitigating the placebo effect in mean ΔmRSS from baseline at 6 months, with less effect at 12 months. Adjustment for RNAP3 status also showed mitigation of the placebo group improvement in mRSS at 6 more so than 12 months. Examination identified that the median change was largest between treatment groups for RNAP3+ and TFR+ participants.

Similar results were seen with the ACR CRISS index, where a significant difference in change in ACR CRISS was found at 6 and 12 months when adjusted for RNAP3 positivity.

One limitation of using the ASSET trial is the trial size, although the study recruitment did meet the required number. While the ASSET trial was negative for the primary endpoint, change in mRSS at 1 year, the abatacept group did improve function (HAQ-DI) and met the co-primary outcome of safety [25]. We feel it was our best option for this confirmatory application of this cohort enrichment technique because it was negative for the mRSS outcome, but with promising secondary/co-primary outcomes, raising the distinct possibility that different design techniques may have led to a positive primary outcome.

This extensive work supports the concept that in early diffuse SSc clinical trials, enriching for RNAP3 positivity, or completing a pre-planned stratified analysis, in a trial using the mRSS or the ACR CRISS as a primary outcome measure may improve our ability to detect a drug effect. This is true for 6 or 12 months of follow-up, allowing for a shorter proof of concept or confirmatory trials. Adjustment for TFR also functions well, but appears of less magnitude and is reliant on examiner skin. The STPR variable is not easily applied to clinical trial design. We thus recommend RNAP3+ status as the preferred enrichment or stratification variable.

RNAP3 is positive in approximately one-quarter of all US patients, and 50% of US patients presenting for early diffuse SSc clinical trials, similar to the ASSET trial [25]. RNAP3+ antibody frequency rates are similar in the UK and Canada, although significantly lower in European observational cohort studies [15]. Given the significant portion of early diffuse SSc clinical trial participants who are RNAP3+, enrichment at the time of enrolment is feasible.

The riociguat in patients with early diffuse cutaneous systemic sclerosis (RISE-SSc) trial, which enrolled predominantly European patients, evaluated the effect of riociguat in early diffuse SSc of <18 months duration. This study enrolled a higher percentage of anti-Scl70+ patients (40%), than RNAP3+ patients (22%) [12]. This autoantibody distribution was likely in part attributable to the mRSS cut-off of 22 at enrolment. Based upon the Pittsburgh Scleroderma Center data, the mean mRSS of a RNAP3+ patient presenting with <36 months of disease is 26. Thus, most RNAP+ patients would have been excluded from the RISE-SSc trial. We are uncertain if our enrichment approach would be useful in trials such as RISE-SSc given the mRSS inclusion criterion cut-offs, as our sample size precluded this analysis.

The mRSS is a reliable and well-validated instrument [6]. But, in recent years the failure of diffuse SSc clinical trials has been stated by some to indicate that it is an inadequate outcome measure. We believe that the mRSS should not be discarded completely. Rather, we feel that in this work, combined with our previous analysis optimizing disease duration as an inclusion criterion [19], we have demonstrated that some negative clinical trials could be attributed to the trial design and enrolment populations. Given the mathematical contribution of the mRSS to the calculation of the ACR-CRISS, when the ACR-CRISS is used in future trials as a primary outcome measure, the findings in our two manuscripts still support the need for alternative approaches to early diffuse SSc clinical trial design.

Conclusion

We have identified three potential variables for predictive cohort enrichment in early diffuse SSc trial design. The first, STPR, could not be easily tested in real-world clinical trials, as the necessary components were not collected as part of available clinical trial data. When our potential variables were tested using the ASSET clinical trial data, both RNAP3 and TFR demonstrate mitigation of the placebo effect in mRSS, most notable at a 6-month endpoint. Observation of mRSS change in those with and without, and those RNAP3+ vs RNAP–, suggest a statistically significant difference in mRSS decline at 6 and 12 months of follow-up. These data suggest that prognostic enrichment or stratification for TFR or RNAP3 through patient enrolment (i.e. inclusion criterion) may improve clinical trial design. With TFR, the exam finding can change over time and is dependent on examiner skill. Anti-RNAP3 positivity is an objective laboratory test whose presence will not change over the course of a clinical trial. It can be confirmed using the gold standard technique of immunoprecipitation relatively inexpensively. We therefore believe anti-RNAP3 to be the easiest to use enrichment approach for multicentre clinical trials, and favour a 6-month trial endpoint.

Based on the collective work presented here and in prior papers we make the following recommendations for early diffuse clinical trial design.

• Application to real-world clinical trial data revealed mitigation of the placebo effect in mRSS behaviour for RNAP3+ and TFR presence. This effect was greater at 6 than 12 months, suggesting cohort enrichment for RNAP3+ or TFR may be particularly useful for a 6-month (24-week) endpoint.

• Analysis of ACR-CRISS data supports enriching or stratifying for RNAP3 presence at 6 months, or TFR or RNAP3+ at 12 months, may be an effective strategy to improve our ability to detect a true difference between groups.

• Optimally we should recruit patients of <18 months of disease and enrich or stratify for TFR or RNAP3 with a 6-month endpoint [19]. If unable to do so, then stratified analysis by disease duration should be an a priori approach (as validated in the ASSET trial).

• Pragmatically, if, for feasibility, patients are enrolled up to 36 months of disease, stratify by RNAP3.

Supplementary data

Supplementary data are available at Rheumatology online.

Data availability

The data are available from the corresponding author, (R.T.D.), upon request.

Funding

Research was supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases of the National Institutes of Health under award numbers R01 AR069874 and P50 AR060780 at the University of Pittsburgh. At the University of Michigan, work was supported by NIH/NIAMS K24 AR063120-09.

Disclosure statement: R.T.D. reports personal fees from CSL Behring and Eicos Sciences Inc., outside the submitted work. R.L. reports personal fees from Bristol Meyers Squibb, Boehringer Ingleheim, Formation, Sanofi, Boehringer-Mannheim, Merck, Genentech/Roche, and grants from AstraZenica, Corbus, Formation, Elpidera, Regeneron, Pfizer, Bristol Myers Squibb and Kiniksa, outside the submitted work. V.S. reports personal fees from Boehringer-Ingelheim, Eicos Sciences, Inc., Corbus Pharmaceutical Holdings, CSL Behring, Formation Biologics and Galapagos outside the submitted work. D.K. reports consulting fees from Acceleron, Actelion, Amgen, Boehringer Ingelheim, Chemomab, CSL Behring, Genentech/Roche, Horizon, Paracrine Cell Therapy, Mitsubishi Tanabe Pharma, Prometheus and Theraly; D.K. has stock options in Eicos Sciences.