This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
PROTECT was a randomized, parallel-group, quadruple-masked phase 3 trial evaluating whether sparsentan reduced proteinuria more than irbesartan in patients with immunoglobulin A nephropathy, with longer-term secondary endpoints describing the trajectory of kidney function.
| Feature | PROTECT |
|---|---|
| Official title | A Study of the Effect and Safety of Sparsentan in the Treatment of Patients With IgA Nephropathy |
| Phase | Phase 3 |
| Condition | Immunoglobulin A nephropathy |
| Design | Randomized, parallel-group, quadruple-masked; primary purpose treatment |
| Number of arms | 4 (registry count) |
| Interventions | Sparsentan; irbesartan; dapagliflozin |
| Primary endpoint | Percent change from baseline in urine protein/creatinine (UP/C) at week 36 |
| Enrollment | 406 |
| Status | Active, not recruiting |
| ClinicalTrials.gov | NCT03762850 |
| Lead sponsor | Travere Therapeutics, Inc. (industry) |
2. Clinical Question
The central question was whether sparsentan, compared with the angiotensin receptor blocker irbesartan, would produce a larger reduction in proteinuria, measured as the urine protein/creatinine ratio, after 36 weeks of blinded treatment. Because irbesartan is an active comparator rather than a placebo, the comparison asks about incremental benefit over an established class of therapy, not about benefit over no treatment.
Population
Patients with immunoglobulin A nephropathy enrolled in a randomized, quadruple-masked treatment study.
Intervention
Sparsentan, administered under double-blind conditions.
Comparator
Irbesartan, an active control, administered under double-blind conditions.
Primary question
Is the percent change from baseline in UP/C at week 36 greater with sparsentan than with irbesartan?
3. Trial Design
Sparsentan
- Investigational treatment
- Double-blind administration
- Analyzed by randomized assignment
Irbesartan
- Active comparator
- Double-blind administration
- Analyzed by randomized assignment
4. Randomization, Stratification, and Analysis Populations
The registry description of the primary analysis states that the model included fixed effects for the randomization strata, confirming that randomization was stratified. The strata themselves are not named in the ClinicalTrials.gov record. Including stratification factors as covariates is good practice: it aligns the analysis with the randomization scheme and usually improves precision.
| Analysis population | Definition / role |
|---|---|
| Primary Analysis Set (PAS) | The subset of the Full Analysis Set at the time of the data extraction for the primary analysis; participants analyzed according to randomized treatment assignment. Used for the week-36 UP/C endpoint. |
| Full Analysis Set (FAS) | Used for the eGFR total slope and chronic slope analyses. |
| Safety population | 202 participants at risk in each of the sparsentan and irbesartan groups for the serious adverse event summary. |
The PAS is defined by the timing of data extraction rather than by treatment exposure: it contains those randomized participants whose week-36 data could contribute when the primary analysis was run. Because participants remain in their randomized groups, the analysis keeps the intention-to-treat principle for the population it covers, while the later FAS analyses include the full randomized cohort.
5. Endpoints
| Endpoint | Role | Time frame | Unit / effect measure |
|---|---|---|---|
| Percent change from baseline in urine protein/creatinine (UP/C) at week 36 | Primary | Baseline (Day 1) and week 36 | Percent change; geometric mean ratio |
| Total slope of estimated glomerular filtration rate (eGFR) over a 110-week period | Secondary | Day 1 to week 110 | mL/min/1.73 m²/year; slope difference |
| Annualized slope of eGFR following the initial acute effect of randomized treatment (chronic slope) | Secondary | Week 6 to week 110 post randomization | mL/min/1.73 m²/year; slope difference |
The registry description of the primary endpoint specifies that UP/C was measured from a 24-hour urine sample, that baseline was defined as the last non-missing observation on or before the start of dosing, and that the analysis was carried out on the log scale with back-transformation of the estimates.
6. Primary Endpoint Results: UP/C at Week 36
The posted primary analysis compared sparsentan with irbesartan in the Primary Analysis Set using a mixed-models analysis. The effect measure is a geometric mean ratio (GMR): the ratio of the geometric mean UP/C, relative to baseline, in the sparsentan group to that in the irbesartan group.
Geometric mean ratio, sparsentan vs irbesartan
95% CI (two-sided): 0.51–0.69 · P < 0.0001
Mixed-model repeated-measures analysis of log UP/C change, Primary Analysis Set
| Item | Value |
|---|---|
| Groups compared | Sparsentan vs irbesartan |
| Statistical method | Mixed models analysis (MMRM on log scale, multiple imputation, Rubin's rules) |
| Effect measure | Geometric mean ratio |
| Estimate | 0.59 |
| 95% CI (two-sided) | 0.51 to 0.69 |
| P-value | <0.0001 |
| Analysis population | Primary Analysis Set, analyzed by randomized treatment |
What the estimate means. A GMR of 0.59 means that, at week 36, the model-estimated geometric mean UP/C relative to baseline in the sparsentan group was 59% of that in the irbesartan group, which corresponds to an approximately 41% greater relative reduction in proteinuria with sparsentan than with irbesartan. Because the analysis was done on the log scale, the comparison is multiplicative: it is a ratio of fold-changes, not a difference in percentage points.
What it does not mean. It does not mean that proteinuria fell by 41% in every patient, nor that sparsentan patients had 41% less protein in absolute terms. It is also not the within-group percent change for either arm; those are separate quantities. And it is a statement about a surrogate marker at 36 weeks, not about kidney failure or other clinical events.
Precision. The 95% confidence interval of 0.51 to 0.69 lies entirely below 1.00, and even the upper limit corresponds to an approximately 31% relative reduction. The interval is fairly narrow, indicating that the size of the proteinuria effect is estimated with reasonable precision in this population.
Why the p-value is not the effect size. P < 0.0001 says that data this extreme would be very unlikely if the true GMR were 1.00. It does not say how large the effect is; that information is in the GMR and its interval.
Cautions. The analysis population was the Primary Analysis Set, defined by the data-extraction date, so it may not include every randomized participant. Missing week-36 values were multiply imputed; the result depends on the assumptions of the imputation model, typically that data are missing at random given observed values. The registry lists the hypothesis type as "Other / not stated," so the p-value should be read as a two-sided test of no difference.
7. Secondary Endpoint Results: eGFR Slopes
Two secondary analyses describe the rate of change in kidney function over approximately two years, both in the Full Analysis Set and both using mixed models. The effect measure is the difference in annualized eGFR slope (sparsentan minus irbesartan), in mL/min/1.73 m² per year. A positive difference means that eGFR declined more slowly, on average, with sparsentan.
| Secondary endpoint | Time frame | Slope difference | 95% CI | P-value |
|---|---|---|---|---|
| Total eGFR slope | Day 1 to week 110 | 1 | −0.03 to 1.94 | 0.0582 |
| Chronic eGFR slope | Week 6 to week 110 | 1.1 | 0.07 to 2.12 | 0.0369 |
Total slope
Estimated difference of 1.0 mL/min/1.73 m²/year favoring sparsentan. The 95% CI (−0.03 to 1.94) just includes zero, and P = 0.0582 is above 0.05.
Chronic slope
Estimated difference of 1.1 mL/min/1.73 m²/year. The 95% CI (0.07 to 2.12) excludes zero, and P = 0.0369.
Why two slopes? Drugs that act on renal hemodynamics often cause an early, partly reversible drop in eGFR after starting treatment. The total slope from Day 1 includes this acute effect; the chronic slope starting at week 6 excludes it and describes the longer-term rate of decline. The two estimates are close (1.0 and 1.1), suggesting that the acute phase did not strongly change the overall picture, but the chronic slope has an interval that excludes zero and the total slope does not.
Near the boundary. The total slope interval (−0.03 to 1.94) sits just across zero, and the chronic slope interval (0.07 to 2.12) sits just above it. These are nearly identical findings in terms of point estimate and width; describing one as "positive" and the other as "negative" based solely on which side of 0.05 the p-value falls would exaggerate a small difference. Both intervals are compatible with effects ranging from essentially none to about 2 mL/min/1.73 m² per year.
Multiplicity. The registry does not describe a hierarchical testing procedure for the secondary endpoints. Without a stated alpha allocation, these p-values should be treated as nominal.
8. Safety: Serious Adverse Events
| Group | Participants with serious adverse events | At risk |
|---|---|---|
| Sparsentan | 75 | 202 |
| Irbesartan | 71 | 202 |
Serious adverse event counts were similar in the two groups: 75 of 202 with sparsentan and 71 of 202 with irbesartan. No formal statistical comparison of safety was posted. Counts of participants with any serious event do not distinguish event type, severity, relatedness or duration of exposure, and a trial powered for a proteinuria endpoint is not designed to detect modest differences in safety.
9. Statistical Methodology
Mixed-model repeated measures (MMRM)
The primary analysis used an MMRM. The dependent variable was the change from baseline in UP/C on the log scale, measured at each scheduled visit during the double-blind period. The model included:
- log baseline UP/C as a covariate;
- fixed effects for randomized treatment, time (nominal visit in weeks) and the treatment-by-time interaction;
- fixed effects for the randomization strata;
- participants as a random effect.
The treatment effect at week 36 is read from the treatment-by-time interaction at that visit, then back-transformed from the log scale into a ratio.
The estimated treatment contrast at week 36, Δ, is back-transformed as GMR = exp(Δ). A GMR of 0.59 corresponds to Δ on the log scale of approximately −0.53.
Multiple imputation and Rubin's rules
Missing responses were imputed before analysis using multiple imputation. The MMRM was fitted to each completed dataset, and the treatment effects were combined using Rubin's approach, which averages the estimates and adds a between-imputation component to the variance so that the final confidence interval reflects uncertainty about the missing values.
where Q̂j is the estimate from imputed dataset j, Ū the average within-imputation variance and B the between-imputation variance across m imputations.
eGFR slope models
The two secondary endpoints were analyzed with mixed models in the Full Analysis Set, with the treatment effect expressed as a difference in annualized slope. A slope model summarizes each group's eGFR trajectory as a rate of change per year, using all available measurements from each participant rather than a single end-of-study value.
Analysis by randomized assignment
For the primary analysis, participants in the Primary Analysis Set were analyzed according to randomized treatment assignment, consistent with the intention-to-treat principle within that set.
10. Statistical Methods Explained
Proteinuria measurements are right-skewed and treatment effects on them tend to be proportional: a drug that reduces protein excretion by a given fraction produces larger absolute drops in patients who start higher. Taking logs turns proportional effects into additive ones, makes residuals closer to normal, and allows the result to be reported as a ratio after back-transformation. Adjusting for log baseline UP/C further reduces variability.
It is the ratio of geometric means of the week-36-to-baseline UP/C ratio: sparsentan divided by irbesartan. Values below 1 favor sparsentan. 0.59 implies that the typical post-treatment proteinuria, relative to baseline, was about 41% lower with sparsentan than with irbesartan. A geometric mean is less influenced by extreme values than an arithmetic mean, which suits skewed data.
An MMRM uses measurements from all visits, models their correlation within participants and yields a valid estimate under a missing-at-random assumption. A simple comparison of week-36 values would discard participants with missing week-36 data and ignore information from earlier visits.
An MMRM already handles missing-at-random data implicitly. Explicit multiple imputation before analysis lets the analyst control how missing values are generated, for example using information not in the analysis model, and makes the imputation assumptions transparent. Rubin's rules then propagate the uncertainty from imputation into the final interval.
The total slope answers "what was the net change in kidney function over 110 weeks, including any early drop?" The chronic slope answers "once the initial response had settled, how fast did kidney function decline?" They can give different answers if a treatment has an early hemodynamic effect, and they rely on different amounts of follow-up time.
A p-value above 0.05 means that the data do not exclude a zero difference at the 5% level; it is not evidence that the difference is zero. The interval (−0.03 to 1.94) is mostly positive and nearly identical to that of the chronic slope. Absence of statistical significance is not evidence of absence of an effect.
11. Multiplicity and Endpoint Hierarchy
| Analysis | Role | Interpretation |
|---|---|---|
| UP/C at week 36 | Primary | Confirmatory comparison of sparsentan vs irbesartan |
| Total eGFR slope | Secondary | Supportive; nominal p-value |
| Chronic eGFR slope | Secondary | Supportive; nominal p-value |
| Serious adverse events | Safety | Descriptive; no formal comparison posted |
12. Important Limitations and Interpretation Issues
- Surrogate primary endpoint: proteinuria at 36 weeks is a biomarker. The eGFR slope endpoints provide longer-term information on kidney function, but they are also surrogate measures rather than clinical events such as kidney failure.
- Analysis population for the primary endpoint: the Primary Analysis Set was defined at a data-extraction date and is a subset of the Full Analysis Set, so it may not contain all randomized participants.
- Missing data assumptions: multiple imputation and MMRM rely on the missing-at-random assumption. Informative dropout, for example because of worsening kidney function, would not be fully accounted for.
- Active comparator: the result describes the benefit of sparsentan relative to irbesartan, not relative to placebo.
- Secondary endpoint multiplicity: the registry does not describe how type I error was controlled across secondary endpoints.
- Additional arms: the registry lists four arms and a dapagliflozin intervention that are not part of the posted comparisons, so the posted results do not describe that component.
- Limited reporting: the ClinicalTrials.gov record does not report baseline characteristics, per-arm values, or subgroup analyses in the posted statistical analyses.
13. Why This Trial Matters Statistically
PROTECT is a useful teaching case for continuous, skewed biomarker endpoints analyzed with repeated measures, and for the interpretation of slope-based kidney function endpoints.
| Concept | How it appears in PROTECT |
|---|---|
| Log transformation | UP/C analyzed on the log scale; results back-transformed to a geometric mean ratio |
| Mixed-effects model | MMRM for UP/C; mixed models for eGFR slopes |
| Covariate adjustment | Log baseline UP/C and randomization strata in the model |
| Multiple imputation | Missing responses imputed; Rubin's rules used to combine estimates |
| Confidence intervals | Narrow interval for the GMR; intervals straddling or just excluding zero for eGFR slopes |
| P-values | Contrast between P = 0.0582 and P = 0.0369 for nearly identical slope estimates |
| Analysis populations | Primary Analysis Set versus Full Analysis Set |
| Active comparator | Incremental effect over irbesartan |
14. Related Tutorials
Learn more about the methods used in this trial:
15. Related Calculators
16. Sources
- ClinicalTrials.gov: NCT03762850 — A Study of the Effect and Safety of Sparsentan in the Treatment of Patients With IgA Nephropathy.
- PubMed: PMID 41428405
- PubMed: PMID 38299639
- PubMed: PMID 37931634
- PubMed: PMID 37254256
- PubMed: PMID 37015244
Explore more trial analyses and methods
Each trial page connects its endpoints and methods to tutorials and calculators that explain the underlying statistics.
17. Record Summary
PROTECT compared sparsentan with irbesartan in IgA nephropathy. On the primary endpoint, the geometric mean ratio for UP/C change at week 36 was 0.59 (95% CI 0.51–0.69; P < 0.0001), estimated with a log-scale MMRM after multiple imputation. The eGFR slope differences favored sparsentan by 1.0 (total slope; 95% CI −0.03 to 1.94; P = 0.0582) and 1.1 mL/min/1.73 m²/year (chronic slope; 95% CI 0.07 to 2.12; P = 0.0369). Serious adverse events occurred in 75 of 202 and 71 of 202 participants, respectively. The most informative reading combines the ratio-scale proteinuria effect, the width and position of each confidence interval, and the design features (active comparator, analysis sets, imputation) that shape its interpretation.