The use of Day 14 endpoint models in aGvHD clinical trials

The overall response rate (ORR) of clinical symptoms at Day 28 is a validated replacement for non-relapse mortality (NRM) and has been accepted as the primary endpoint for clinical trials of acute graft-versus-host disease (GvHD) treatment.¹ Physicians typically modify immunosuppression earlier than Day 28 for patients who do not respond to treatment; therefore, measuring ORR at Day 28 may not be as effective.¹ A further limitation of using Day 28 ORR is that it does not account for the distinct treatment goals required for patients with high- and low-risk GvHD.¹

Here, we summarize an article by Spyros et al.¹ published in Transplantation and Cellular Therapy, which evaluates Day 14 endpoint models for acute GvHD (aGvHD) as predictors of long-term outcomes.

Methods¹

• Patients who underwent hematopoietic cell transplantation between January 2015 and December 2021, and had data and samples in the Mount Sinai Acute GvHD International Consortium (MAGIC) database and biorepository were evaluated.
  • Data were collected and reviewed using a prospective-specimen collection, retrospective-blinded-evaluation (PRoBE) design.
  • Patients with Grade 1─4 aGvHD, who had not relapsed, and were treated with 0.25 mg/kg/day of systemic steroids or other immunosuppressive agents were included in the analysis.
• A recursive partitioning classification tree algorithm was used to develop a new predictive model (the Day 14 Mount Sinai model), which accounts for the grade of GvHD at treatment initiation and at Day 14 to classify patients into favorable and unfavorable groups, with 12-month NRM as the outcome variable.
• The Day 14 MAGIC algorithm probability (MAP) score was included in the Day 14 Mount Sinai model to create a Day 14 MAGIC model, using Day 14 MAP biomarker score (low ≤ 0.290 vs high > 0.290).
• Responses or non-responses at Day 28 were categorized according to standard clinical criteria.
• Quantitative comparisons were made of the different response models for censored event times with competing risks.

Key findings¹

• In total, 1,144 patients, from 23 MAGIC sites, were included in this study.
• The Mount Sinai model correctly identified 57% of NRM events in the unfavorable group compared with 48% identified by the Day 28 standard criteria model in the non-response group.
• With the Day 14 MAGIC model, three distinct response populations were identified which were good, intermediate, and poor, with escalating NRM (8%, 35%, and 76%, respectively).
  • With Day 28 standard criteria, partial and complete responders experienced overlapping NRM.
• The Day 14 MAGIC model successfully stratified patients with Grade 2 aGvHD at onset that remained Grade 2 at Day 14.
• Overall, 12 months following GvHD treatment, the Day 14 MAGIC model was more accurate, as measured by area under the curve vs the Day 28 standard clinical response and the Day 14 Mount Sinai model (Figure 1).

Figure 1.  Quantitative comparison of Day 28 and Day 14 models for competing risks with NRM as outcome* 

AUC, area under the curve; GvHD, graft-versus-host disease; NRM, non-relapse mortality; MAGIC, Mount Sinai Acute GvHD International Consortium.
*Data from Spyrou, et al.¹

• The Day 14 MAGIC model displayed the most favorable sensitivity, Youden’s index, and negative predictive values among the models when 12-month NRM was considered as the endpoint (Table 1).

Table 1. Model characteristics at the 12-month post-treatment NRM endpoint*