Data-driven grading of acute graft-versus-host disease

Despite advances in allogeneic hematopoietic cell transplantation, acute graft-versus-host disease (aGVHD) remains its leading complication, yet with heterogeneous outcomes. Here, we analyzed aGVHD phenotypes and clinical classifications in depth in large, multicenter cohorts involving 3019 patients and addressed prevailing gaps by developing data-driven models. We compared, tested and verified these along with all conventional classifications in independent cohorts and found that data-driven grading outperformed conventional grading in Akaike information criterion and concordance index metrics. Data-driven classifications refined aGVHD assessment with up to 12 severity grades, which were associated with distinct nonrelapse mortality (NRM) and confirmed the key role of intestinal aGVHD. We developed an online calculator for physicians to implement principal component-derived grading (PC1). These results provide substantial insight into the evaluation of aGVHD phenotypes and multiorgan involvement, which relegates the exclusive reporting of overall aGVHD severity grades in transplant registries and clinical trials. Data-driven aGVHD grading provides an expandable platform to refine classification and transplant risk assessment.

outcome association (e.g. Figure 6).Detailed list of the involved phenotypes are provided in the supplementary tables 3-13.
Analysis of redistributed patients: Some grading systems categorized a high number of aGVHD phenotypes in a single grade.Within these categories (e.g. in MAGIC grade III), we identified patients that were attributed to distinct aGVHD severity grades in other aGVHD grading systems, which we coined redistributed patients (e.g.MAGIC grade III patients that were classified by PC1 as grade II).These redistributed patients were analyzed separately from the remaining patients of the same severity grade and compared for differences in clinical outcome to dissect heterogeneity within selected large classification categories.(cumulative dosage: ATG 60 mg/kg).Before patient discharge, intravenous ciclosporin was substituted orally.Inpatients had daily medical visits until discharge around day+30.
are presented as density plots.Patient numbers (n) of each subgroup are indicated right to each box.Higher n in each subgroup is shown by greater surface coverage.Density of aGVHD target organ combinations is indicated from light green to dark blue.b Target organ stage correlation matrix (Spearman) of the training cohort shows the distribution of single variables skin, liver and GI and their respective interactions.Range from -1.0 to +1.0, dark blue indicates full overlap.c Pair plot and d Target organ stage correlation matrix (Spearman) of ⅓ split cohort (n=996).Analysis, labels and colors as in a-b.Supplementary Figure 2. Internal Validation and additional characteristics of PC1-based grading development, training cohort (n=2319) a Plotting of PC1 and overall survival (OS, days from HCT without censoring) reveals lower OS with increasing PC1.Each dot representing one patient with aGVHD.Colored dots representing Consensus grade I-IV (I=yellow, II=green, III=blue, IV=violet) indicate overlap between Consensus grades I & II and II & III on the PC1 axis.b Plotting of PC1 stages and OS (months from HCT, censoring < 12 months has not been considered in this representation, patients surviving > 12 months were censored at the 12 months dot) with a power function derived from maximal OS confirms lower OS with increasing PC1 within the first 12 months after HCT.Colors as in B. c 500-fold bootstrapping of PCA on training cohort sample (each sample with size n=1546).The boxplot shows the eigenvalues of each PC (median, orange middle line; 25th, 75th percentile, box).The two lines outside of the box represent the 5th and 95th percentile (whiskers).Single points outside the whiskers are outliers.Source data are provided as a source data file.d Color-coded plotting of PC1 stages against aGVHD organ involvement (combinations: 1:skin, red, 2: liver, yellow 3: GI, lime 4: skin and liver, green 5: skin and GI, blue 6; liver and GI, violet 7: skin and liver and GI purple).The circle size corresponds to the n of patients in each category.training cohort (n=2319) on two-dimensional data (tSNE1 and t-SNE2) with each dot representing a single patient.Patients are colored according to their corresponding severity grading (grade I, light green; grade II, dark green; grade III, teal, grade IV, blue).a PC1 4 grades, b Hierarchical clustering 4 grades, c K-means 4 grades.d MAGIC 4 grades, e Consensus 4 grades, f IBMTR 4 grades.Supplementary Figure 5. Visualization of ascending DBSCAN cluster numbers on UMAP output and OS association, training cohort (n=2319) Nonlinear dimensionality reduction with UMAP applied on the training cohort (n=2319) revealing two-dimensional data (UMAP1 and t-UMAP2) with each dot representing a single patient.The UMAP output was clustered using DBSCAN and patients colored according to the corresponding DBSCAN cluster.Explorative Kaplan-Meier analysis was performed to illustrate the spread of the clusters.a DBSCAN with 4 clusters.Colors representing DBSCAN cluster aGVHD severity grades I-IV (grade I, light green; grade II, dark green; grade III, teal, grade IV, blue) b Kaplan-Meier OS curve with 95% confidence interval (CI) of 4 clusters.Strata were compared with the two-sided log-rank test.c 8 clusters.Colors representing DBSCAN cluster aGVHD severity grades I-VIII.d Kaplan-Meier OS curve with 95% CI of 8 clusters.Strata were compared with the two-sided log-rank test.e 12 clusters.Colors representing DBSCAN cluster aGVHD severity grades I-XII.f Kaplan-Meier OS curve with 95% CI of 12 clusters.Strata were compared with the two-sided log-rank test.includes age).Horizontal bars represent 95% CI.P values were computed using the Wald test.The hazard ratio (HR) is a measure of the ratio of the hazard between two groups.A value of 1 is the reference.Source data are provided as a source data file.grades and a common reference cohort of patients with aGVHD grade 0 (same reference for both conventional and data-driven classifications).P values were computed using the Wald test.Source data are provided as a source data file.a PC1 4 grades b HClust c K-means 4 grades d MAGIC e Consensus f IBMTR.Error bars show 95% confidence interval.
Description of institutional transplant protocols and practices: The studied patients received HCT for treatment of hematological malignancies.Donor search was performed by the respective transplant office at each institution in collaboration with donor-databases.Donors were HLA-matched related donors (MRD), mismatched-related and haploidentical related donors (MMRD, 10/10 HLA-A-, -B, -C, -DRB1, -DQB1 matched unrelated donors (MUD), or mismatched unrelated donors (MMUD).HLA-DPB1 was not considered for donor-recipient matching.The conditioning regimen was chosen by the treating physician considering diagnosis, age, comorbidities and donor constellation.Patients received the same early supportive and follow-up care.With the beginning of the conditioning regimen until discharge, all in patients were treated in reverse isolation single rooms with high-efficacy particle air filtration.Antiviral prophylaxis during neutropenia consisted of intravenous aciclovir at 250 mg three times daily.Antifungal prophylaxis consisted of oral posaconazole at 200 mg three times daily from day+1 with a minimal duration until day +100.Colony stimulating factors were not routinely applied.As pneumocystis-jirovecii pneumonia prophylaxis patients received either monthly pentamidin inhalation or oral cotrimoxazol at 960 mg three times per week from day +30.Only, irradiated red blood cell and platelet transfusions and in-line leukocyte-filtered products were used.Patients in both training and test cohorts received a pharmacological aGVHD prophylaxis including calcineurin inhibitors, most frequently in combination with either methotrexate or mycophenolate mofetil.Patients with higher GVHD risk (in particular thosewith MUD and MMUD) were assigned to additional in vivo T cell depletion using anti-T-Lymphocyte globulin (ATG).In Essen, Heidelberg and Berlin, the preferred calcineurin inhibitor based GVHD prophylaxis was ciclosporin plus methotrexate (MTX), while in Hamburg and Hannover it was the combination of ciclosporin and mycophenolate mofetil (MMF).Ciclosporin was applied intravenously at 3 mg/kg bodyweight starting from day -1 before HCT e.g. in combination with 15 mg/m2 MTX on day +1 und 10 mg/m2 MTX on days +3, +6 and +11 after HCT.Normal early ciclosporin target blood levels (range, 150-250 ng/ml) controlled three times weekly.MMF was started intravenously on day 0 at 30mg/kg bodyweight/day split into 2-3 daily doses.Following drug oralisation, MMF was continued orally depending on the protocol e.g. until day +28 for MRD/MUD or until day +35 (MMUD, MMRD).Polyvalent rabbit-ATG was applied at a dosage of 10 mg/kg bodyweight on days -4, -3 and -2 (cumulative dosage: ATG 30 mg/kg) or at a dosage of 20 mg/kg bodyweight on days -4, -3 and -2

Table 2 :
CMV, cytomegatovirus; D, donor serostatus; R, recipient serostatus; RIC, reduced intensity conditioning; TBI, total body irradiation; ATG, anti-thymocyte globulin.P values were calculated using chi-square test; t-test was used for age.Core outcomes of training and test cohort