Prognostic scores were calculated for each patient using the definitions and following the methodology described in the original publications.3-8 The KPS and HCT-CI were evaluated as part of our clinical practice, executed by expert allo-HCT physicians. The DRI, EBMT, rPAM and EASIX scores were calculated retrospectively using data from clinical charts. The PAM score was calculated using the simplified version published in 2015 [revised PAM score (rPAM score)]7 and the EASIX score was calculated based on the information provided from bloodwork collected during pre-transplant assessments (between days −30 and −7 before allo-HCT).

Between January 2014 and February 2021, 205 consecutive adults with hematological malignancies underwent their first peripheral blood allo-HCT combined with PTCy-based GvHD prophylaxis at our Institution. Eligibility criteria for allo-HCT are detailed in the Supplementary Material. Clinical information was collected retrospectively and updated in January 2022. This study was approved by the Ethics Committee of the Hospital Clinic de Barcelona, Spain and was performed in accordance with standards set by the Declaration of Helsinki.

Myeloablative conditioning (MAC) regimens generally combined high-dose of busulfan (3.2 mg/kg/day intravenously (IV) × four days) or 12 Gy of total body irradiation (TBI) with fludarabine (30 mg/m2/day IV × four days). The majority of reduced-intensity conditioning (RIC) regimens contained lower doses of busulfan (3.2 mg/Kg/day IV × three days), or 8 Gy of TBI combined with standard doses of fludarabine.

Between January 2014 and May 2015, the Institutional standard GvHD prophylaxis consisted of PTCy 50 mg/kg/day administered on days +3 and +4, followed by mycophenolate mofetil (MMF) from day +5 to day +28, and tacrolimus (TK) started on day +5 at a dose of 0.04/kg/day IV, maintained therapeutically until day +90, and tapered down progressively up to day +120, in the absence of GvHD. Since June 2015, the standard GvHD prophylaxis out of the haplo-HCT setting was modified to PTCY and TK (PTCY-TK). Information related to graft source, infectious prophylaxis and supportive care is reported in the Supplementary Material.

The overall survival (OS) and non-relapse mortality (NRM) were considered the main outcome variables. The main explanatory variables were KPS, HCT-CI, DRI, EBMT score, rPAM and EASIX. Patients were grouped into two risk groups of high- and low-risk strata. This classification was made based on the optimal cutoff value for each score and for the main outcome variable: OS. The cutoff value was calculated using maximally selected rank statistics. In the case of EASIX and following standard practices in the literature, the original value of the index was transformed to a base-2 logarithm value (log2-EASIX) to perform the statistical analysis.

The OS was calculated using the Kaplan-Meier estimator method and the NRM was estimated using the cumulative incidence method, considering relapse as a competing event. The score prediction ability for the OS and NRM was first explored during the entire post-transplant follow-up across the entire cohort and in each of the scores independently, using the concordance index (C-index). This index confirms the predictive power of the evaluated score for values superior to 0.5 with the perfect discrimination of the value of 1.19 The prognostic indices were compared using the reclassification calibration statistic goodness-of-fit applied to reclassified categories. This method compares the proportions of patients whose estimated risk shifts in the correct and wrong directions based on a χ2 goodness-of-fit test within reclassified categories for each score. In this setting, a p-value > 0.05 indicates a good fit.20 Diagnostic accuracy measures (sensitivity, specificity and predictive values) were calculated at 2 years.

The impact of the explanatory variables on the OS and NRM and the prediction ability of the five scores were analyzed using the regression analysis and C-index, respectively, and censoring the post-transplant follow-up at the following time points after the allo-HCT: 3 months, 6 months, 1 year and 2 years. The score discrimination was additionally evaluated using the inverse probability of censoring weighting estimation of cumulative time-dependent receiver operating characteristic (ROC) curves and, from them, calculating the area under the curve (AUC) at each time-point. The perfect discrimination corresponds to an AUC of 1.0, meaning that the predicted risk for all individuals who developed the outcome was higher than that for all individuals who did not experience the outcome. An AUC of 0.5 is indicative of an equal probability of risk assigned to individuals who experience the outcome and to individuals who did not, i.e., the prediction is purely random.21 The p-values were two-sided. For statistical analyses, p < 0.05 was considered to indicate a statistically significant result. Statistical analysis was performed using the R.3.6.1 software (R Foundation for Statistical Computing, Vienna, Austria, http://www.R-project.org).

As shown in Table 1, the patients’ median age was 52 years (range: 18 - 70), acute myeloid leukemia (38.0 %) was the most prevalent baseline diagnosis, 51.2 % of adults underwent the RIC allo-HCT and 72.2 % of patients received grafts from unrelated donors (UDs), while 19.5 % received from haploidentical donors.

Of the 205 patients included, 202 (98.5 %) were primary engrafted. Three (1.4 %) patients had moderate-severe sinusoidal obstruction syndrome (SOS) and 9 (4.4 %), thrombotic microangiopathy (TMA). The cumulative incidences of grades II to IV and grades III to IV acute GvHD at day +100 were 24.4 % (95 %CI 18.7–30.5) and 4.9 % (95 %CI 2.5–8.4), respectively, while those of moderate/severe chronic GvHD at 2 years was 7.6 (95 % CI 4.2–12.2). With a median follow-up among surviving patients of 35.9 months, 54 (26.3 %) patients relapsed and 66 (32.2 %) died. The main causes of death were relapse (34 patients, 16.6 %) and infection (13 patients, 6.3 %). The estimated 2-year OS, NRM, relapse-free survival and cumulative incidence of relapse were, respectively, 69.9 % (95 %CI 62.8–76.0), 16.1 % (95 %CI 11.4–21.6), 57.5 % (50.1 %–64.2 %) and 26.3 % (95 %CI 20.3–32.8).

Based on the result obtained from the five clinical risk-scoring indices, patients were categorized into two risk groups (high-risk and low-risk) based on the optimal cutoff value for the OS prediction determined using maximally selected rank statistics. As shown in Table 1, according to the KPS, 49 (23.9 %) patients were classified into the high-risk group (KPS values: 70–80 %). According to the HCT-CI, the number of high-risk (HCT-CI: > 3) patients was 45 (22.0 %). Regarding the DRI, 46 (22.4 %) patients were classified into the high-risk group (DRI: high or very high risk) and, according to the EBMT score, 68 (33.8 %) adults were considered high-risk (EBMT score: ≥ 4). The optimal cutoff value of the rPAM score for the prediction of the OS was ≥ 21.0 (Supplementary Figure 1) and 59 (28.8 %) adults were classified into the high-risk group. Lastly, 38 (18.5 %) patients with log-2 EASIX ≥ 1.900 (pre–logarithmic transformed EASIX equivalent value ≥ 3.800) were classified into the high-risk group (Supplementary Figure 2).

As shown in Table 2 and Figure 1, the classifications of patients in the high- and low-risk strata according to the KPS, HCT-CI, DRI and EASIX indices for the prediction of the OS were statistically significant during the entire post-transplant period. In terms of hazard ratios (HRs) and according to the KPS, high-risk adults were 2.11 (p = 0.004) times more likely to die secondary to any cause than low-risk adults. According to the DRI, patients classified as high-risk were 2.04 (p = 0.005) times more likely to die than low-risk patients. The estimated HR for mortality of high-risk adults according to the HCT-CI index (score > 3) was 1.95 (p = 0.012) and the probability of mortality was 2.15 times superior for patients with an EASIX value ≥ 3.800 (log2-EASIX ≥ 1.900). According to the C-index value, the ranking of indices in terms of the predictive power of overall mortality, from high to low, was as follows: first the KPS (C-value 60 %), DRI (C-value 58 %), EASIX (C-value 58 %), HCT-CI (C-value 56 %), rPAM (C-value 55 %) and EBMT score (C-value 54 %).

Figure 1.

Overall Survival (First Row) and Non-Relapse Mortality (Second Row) Based on the Clinical Scoring System Distribution.

(0.32MB).

The discrimination capacity for the OS at 100 days, 6 months, 1 year, and 2 years is summarized in Table 3 and Figure 2. The KPS and EASIX maintained an appropriate discrimination during the four-time points evaluated after the allo-HCT, with values of the AUC-statistic higher than 55 %. The discriminative ability of the DRI and HCT-CI was low at day +100 (AUC values inferior to 50 %) and increased progressively during the post-transplant period, with a peak of discriminative capacity at 2 years (AUC equal to 61.1 % and 61.8 %, respectively). The rPAM showed an acceptable discrimination capacity during the first year after the allo-HCT (AUC increasing from 55.9 % at day +100 to 58.7 % and 58.2 % at 6 months and 1 year), decreasing to as low as 54.6 % at 2 years. These results were consistent with those obtained in the Cox regression analysis as the criterion for statistical significance and the C-value. Lastly, the predictive capacity for the OS of the EBMT score was neither demonstrated during the entire post-transplant period nor at the different time points after the allo-HCT.

Table 3.

Prediction ability of the six indices for overall survival predicted at different time-points after allo-HCT.

Legend: The prediction ability of each score was analyzed in the entire cohort and independently. Each time period accounts for the sum of events documented from day 0 to day +100, day 180, day +365 and day +730. Green color: Good prediction (C-Index ≥ 58 %), Yellow color: Acceptable prediction (C-Index > 55 % - < 58 %), Red color: Inaccurate prediction (C-Index ≤ 55 %). *ROC curve: Null hypothesis: true area = 0.5.

Figure 2.

Overall Survival and Non-Relapse Mortality Predictive Ability of the Six Indices at Different Time-Points after Allo-HCT.

(0.52MB).

Sensitivity and specificity, together with the positive predictive values (PPVs) and negative predictive values (NPVs) were calculated for the prediction of the 2-year OS for each score independently (Table 3). The estimated sensitivity of each score was inferior to 40 %, except for the EBMT score, which was 44.1 %. The EASIX score, followed by the KPS and HCT-CI, had the highest PPV values (50 %, 46.9 % and 46.6 %, respectively) and the rPAM and the EBMT scores, the lowest (36.2 % and 36 %, respectively). The EASIX, HCT-CI, KPS and DRI were the more specific predictive tools, with proportions superior to 80 % (87 %, 83.6 %, 82.9 % and 82.3 %, respectively) and all these four scales presented an NPV higher than 75 % (76.6 %, 76.8 %, 77.5 % and 76.8 %, respectively) (Table 4).

Table 4.

Sensitivity, positive predictive value, specificity and negative predictive value of each index for the prediction of 2-year overall survival.

Legend: If a predictive tool is less sensitive, risk of mortality can be overestimated (Type I error). If a test less is specific, risk of mortality can be underestimated (Type II error); for these reasons, the present table explores the sensitivity, predictive positive value, specificity and predictive negative value of the selected five prognostic scales for the prediction of mortality at 2 years.

Sensitivity: ability of a screening test to detect a true positive. It refers to the proportion of those who have the condition that received a positive result on this test. Calculation: number of true positives / number of true positives + number of false negatives. Formula:% Sensitivity= [a/(a + b)] × 100. Positive Predictive Value (PPV): probability that people with a positive screening test result indeed do have the condition of interest. Calculation: Truly classified as high-risk/Total predicted as high-risk. Formula: % PPV= [a/(a + d)] × 100.

Specificity: ability of a screening test to detect a true negative. It refers to the proportion of those who do not have the condition that received a negative result on this test. Calculation: number of true negatives/number of true negatives + number of false positives. Formula:% Specificity= [c/(c + d)] × 100. Positive Predictive Negative (PPN): probability that people with a negative screening test result indeed do not have the condition of interest. Calculation: Truly classified as high-risk/Total predicted as high-risk. Formula: % PPN= [c/(c + b)] × 100.

Needs to be checked!!!.

The predictive ability of the five scores was additionally calculated for the prediction of the NRM across the entire cohort and in each of the scores independently. The study cohort was stratified into the same two groups defined for the prediction of the main outcome variable, the OS. In the case of the EASIX and rPAM, the cutoff values for the prediction of the NRM were re-calibrated and were found to be close to those defined for the prediction of the OS. For this reason, as well as to increase the clinical application of the conclusions obtained from the present study, the cutoff values for the EASIX and rPAM were maintained at 3.800 (log2-EASIX 1.900) and 21.0, respectively (Supplementary Figures 1 and 2).

As shown in Table 2 and Figure 1, the KPS and EASIX grouped patients into higher and lower risk strata, supporting their use for risk classification for the prediction of the NRM and during the entire post-transplant period. According to the KPS, high-risk adults were 3.24 (p < 0.001) times more likely to die secondary to transplant-related toxicity than low-risk adults. The estimated HR for transplant-related mortality of high-risk adults, according to the EASIX (log2-EASIX > 1.900), was 3.32 (p < 0.001). Additionally, high-risk adults, according to the DRI, were found to present a non-significant trend to a higher NRM, compared to low-risk patients (HR 1.90, p = 0.079). According to the C-index, the ranking of indices in terms of the predictive power of the NRM from high to low was the KPS (65 %), EASIX (61 %), rPAM (57 %), DRI (56 %) EBMT score (55 %) and HCT-CI (53 %).

The KPS and EASIX maintained an appropriate predictive capacity of NRM during the four time points evaluated during the first 2 years after the allo-HCT and with values of the AUC-statistic higher than the cutoff value of 55 % (Table 5, Figure 2). The KPS was the most accurate prognostic index for the NRM, with an estimated median of AUC ranges > 60 % during the post-transplant period and with a peak of prediction at day +100 and +180 (AUC 71.8 % and 66.2 %, respectively). The rPAM showed an acceptable discrimination capacity at day +180 (AUC of 55.9 %) and at 2 years (56.6 % at 2 years) after the allo-HCT for the prediction of the NRM. In addition, the DRI, HCT-CI and EBMT score were not appropriate predictors for the NRM during the post-transplant period. These results were consistent with those obtained when using the cumulative incidence regression analysis as the criterion for the statistical significance and C-Index.

Table 5.

Prediction ability of the six indices for non-relapse mortality predicted at different time-points after allo-HCT.

Legend: The prediction ability of each score was analyzed in the entire cohort and independently. Each time period accounts for the sum of events documented from day 0 to day +100, day 180, day +365 and day +730. Green color: Good prediction (C-Index ≥ 58 %), Yellow color: Acceptable prediction (C-Index > 55 % - < 58 %), Red color: Inaccurate prediction (C-Index ≤ 55 %).