A simple method to predict response to immunosuppressive therapy in patients with myelodysplastic syndrome

Myelodysplastic syndrome (MDS) is a clonal hematopoietic disorder characterized by bone marrow failure, marrow dysplasia, and a tendency to evolve to acute leukemia. MDS shares some of the features of acquired aplastic anemia (AA) and up to 30% of patients with MDS respond to immunosuppressive treatment with cyclosporine (CSA) or antithymocyte globulin (ATG) with sustained increases in red blood cell, granulocyte, and platelet production.^1-5  The class II histocompatibility antigen DR15 (serologic split of HLA-DR2) is reported to be overrepresented in patients with AA.^6-9  We noted a similar overrepresentation of HLA-DR15 in patients with MDS of the refractory anemia subtype.¹⁰  Furthermore, HLA-DR15, younger age, and a shorter duration of red cell transfusion dependence (RCTD) were pretreatment characteristics correlating significantly with a response to immunosuppression in multivariate analysis.¹⁰  In this study we have used these pretreatment variables to devise a simple method to predict the probability of response to immunosuppression in MDS. To validate the predicted probability of response (PPR), we assessed the correlation of the PPR with observed responses in a separate treatment cohort (the validation cohort). In multivariate analysis the PPR was the single best predictor of response.

The minimum criteria used to diagnose MDS were as published.¹⁰  Characteristics of patients in the cohort used to identify pretreatment variables correlating with response (training cohort, n = 82) have previously been published.¹⁰  The validation cohort consisted of 23 patients with MDS enrolled on the National Heart, Lung and Blood Institute (NHLBI) Institutional Review Board–approved protocol 00-H-0169 for combination ATG and CSA to treat the cytopenia of MDS (17 with refractory anemia [RA], 2 with refractory anemia with ring sideroblasts [RARS], 4 with refractory anemia with excess blasts [RAEB]). Twenty-one patients were receiving regular red cell transfusions for symptomatic anemia. Many of these patients were also thrombocytopenic (< 20 × 10⁹/L) or neutropenic (< 0.5 × 10⁹/L) or both (Table 1). Of the 2 patients not receiving regular red cell transfusions, one met the eligibility criteria because of transfusion-dependent thrombocytopenia, and the other had neutropenia with recurrent infections. On comparison of the training and validation cohorts, nonstatistically significant differences were noted in age, the proportion of patients with hypocellular marrows, and the proportion with paroxysmal nocturnal hemoglobinuria (PNH) by flow cytometry (Table 1).

In the training cohort, 69 patients with MDS were treated with one course of ATG 40 mg/kg/d intravenously over 4 to 6 hours for 4 days and 13 patients were treated with CSA starting at 5 to 12 mg/kg/d, with dose adjustments to maintain therapeutic levels, and continued indefinitely if a response was noted by 6 months.¹⁰  The validation cohort received a combination of ATG and CSA at the same doses with ATG starting at day 0 and CSA at day 14.

For transfusion-dependent patients, response was defined as transfusion independence for red blood cells or platelets (ability to maintain a hemoglobin > 8 g/dL and a platelet count > 20 × 10⁹/L without symptomatic indications for transfusions for a period of at least 6 weeks regardless of subsequent events). For nontransfusion-dependent patients, response was defined as the normalization of the absolute neutrophil count.

The following logistic regression equation with response as the dependent variable was derived from analysis of the training cohort: log_e p/1–p = 8.3 – 0.156 (age in years) – 0.148 (duration of RCTD in months) + 2.14 (HLA-DRB1 15 status: 1 for present, 0 for absent), where p is the probability of a response. The c-index (concordance probability) to measure association of predicted probabilities and associated responses was 0.933 (values of `c' near 0.5 indicate that the model is no better than a coin-flip; values near 1.0 indicate superior performance of the model).

To generate a simplified prediction system integrating the 3 pretreatment variables of age, duration of RCTD, and HLA-DR15, we first converted the units for duration of RCTD to months. Then, to obtain approximately whole number values for the coefficients for age and duration of RCTD we multiplied both sides of the logistic regression equation by 6.67. By rounding to whole numbers the coefficients for age and RCTD in the logistic regression equation, the sum of the patient's age in years and duration of RCTD in months is readily calculated. This sum, depending on the patient's DR15 status, indicates the PPR (Table 2). In the training cohort, the mean ± SD PPR among actual responders was 0.7 ± 0.3 and among nonresponders 0.12 ± 0.18 (distributions were normal). Based on this observation, a PPR range of 0.41 to 1.0 (41%-100%) is designated “high PPR” and a PPR range of 0 to 0.4 (0%-40%)is designated “low PPR” (Table 2). In the training cohort, this choice of cutoffs for PPR categorization generated the largest c-index (0.85) in logistic regression.

The “freq”(χ² and Fisher tests), “ttest” (t test), “anova” (Kruskal-Wallis test), and “logistic” (logistic regression) procedures in SAS (SAS Institute, Cary, NC) version 8.1 were used to analyze for differences between pretreatment variables in cohorts and to assess the association between variables and response.

Of the 23 patients in the validation cohort, 7 met criteria for response to the combination of ATG and CSA. All responding patients were red cell transfusion dependent before treatment and met response criteria for red cell transfusion independence. Of the 7 responders, 5 were pancytopenic and 2 bicytopenic before treatment; all responses were pancellular. In one responder, the duration of transfusion independence was 10 weeks; the remaining 6 patients had durable responses with the median transfusion-free survival not reached at the 2-year follow-up.

Based on Table 2, 14 of the 23 patients had a low and 9 a high PPR. In univariate analysis, PPR (low, high), age, and PNH detected by flow cytometry were significantly associated (P < .05) with a response to immunosuppression (Table 3). In multivariate analyses using the 3 variables significantly associated in univariate analysis (PPR, age, and presence of PNH), only the PPR (which integrates age, duration of RCTD, and DR15 status) significantly correlated with response (Table 3).

The observed responses (ORs) were compared to the PPR (low, high). Responses to immunosuppression occurred in 6 of 9 patients with a high PPR and 1 of 14 patients with a low PPR (Table 4; Fisher exact test, P = .005).

HLA-DR15 may help identify that subset of MDS patients with immune-mediated marrow failure, a finding we have discussed previously.¹⁰  Reasons for the association of age and duration of RCTD with response to immunosuppression are not known. Speculatively, prolonged RCTD may indicate prolonged immune-mediated marrow damage and stem cell depletion. Similarly, older patients may have fewer hematopoietic stem cells.¹¹  Stem cell depletion might limit recovery of normal stem cells after immunosuppression. Other pretreatment variables of predictive value (eg, identification of PNH clones) that could improve the scoring system may be identified through analysis of a larger training cohort.

In conclusion, the proposed scoring system, using HLA-DR15, age, and duration of RCTD, provides a useful method to identify those MDS patients most likely to respond to immunosuppression. These pretreatment variables are also relevant to the design and evaluation of immunosuppression trials for MDS.