Response Predicting Factors to Recombinant Human Erythropoietin in Cancer Patients Undergoing Platinum-Based Chemotherapy

Transcription

1 2408 Response Predicting Factors to Recombinant Human Erythropoietin in Cancer Patients Undergoing Platinum-Based Chemotherapy Manolo González-Barón 1 Amdio Ordóñez 1 Rosa Franquesa 2 Manuel Constenla 3 Joaquin Montalar 4 Frederic Gili 5 Carlos Camps 6 José Felix Sancho 7 Pedro Pérez-Cachot 8 1 Hospital La Paz, Servicio de Oncología Médica, Madrid, Spain. 2 Hospital General de Vic, Servicio de Oncología Médica, Barcelona, Spain. 3 Hospital Montecelo, Servicio de Oncología Médica, Pontevedra, Spain. 4 Hospital La Fe, Servicio de Oncología Médica, Valencia, Spain. 5 Consorci de l Hospital de la Creu Roja, Servicio de Oncología, Barcelona, Spain. 6 Hospital General Universitario de Valencia, Servicio de Oncología, Valencia, Spain. 7 Hospital Militar Central Gómez Ulla, Servicio de Oncología, Madrid, Spain. 8 Hospital Militar Vigil de Quiñones, Servicio de Oncología, Seville, Spain. Presented at the 23rd ESMO Congress, November 6 10, 1998, Athens, Greece and at the ASCO meeting, May 16 19, 1998, Los Angeles, CA. The authors thank M.A. Arcusa, M.D. (Hospital de Terrasa), M. Boleda, M.D. (Hospital de San Camil), R. Márquez, M.D. (Hospital Clínico de Málaga), and M. Feyjóo, M.D., for their contributions to this study. Address for reprints: Manolo González-Barón, Hospital La Paz, Servicio de Oncología Médica, Paseo de Castellana 261, Madrid, Spain; Fax: ; oncopaz@ene.es Received November 13, 2001; revision received May 24, 2002; accepted July 2, BACKGROUND. The response to epoetin- treatment is hard to predict in cancer patients receiving chemotherapy. METHODS. One hundred and seventeen patients were enrolled in this observational study. They had a hemoglobin (Hb) level less than or equal to 10.5 g/dl, were receiving platinum chemotherapy with three cycles pending, and they did not have an iron deficiency or hemolysis. Epoetin- was administered subcutaneously three times a week at a dose of 150 IU/kg. Ninety patients were examined. RESULTS. Response was defined as an increase in Hb of at least 2 g/dl during the treatment period. The response rate was 63.3%. The following data were compared between responders and nonresponders at the onset of treatment and after 2 and 4 weeks of epoetin therapy: Hb, reticulocytes, serum iron, ferritin, transferrin, transferrin saturation index, and endogenous erythropoietin levels. At baseline, these variables were similar for responders and nonresponders; after 2 weeks, responders showed higher Hb (P 0.001) and transferrin levels (P 0.042) and reticulocyte counts (P 0.003); after 4 weeks, only the Hb level showed a significant difference (P ). Changes from baseline in Hb level after 2 and 4 weeks correlated significantly (P 0.01) with response. The change in Hb level at Week 4 was the best predictor. A change in Hb level of less than 0.5 g/dl was associated with a lack of response (predictive power, 71%); a change in Hb greater than or equal to 0.5 g/dl was associated with response (predictive power, 89%). CONCLUSIONS. Response to epoetin- treatment in cancer patients receiving platinum chemotherapy can be predicted from changes in Hb level after 4 weeks of therapy. Cancer 2002;95: American Cancer Society. DOI /cncr KEYWORDS: platinum chemotherapy, epoetin-, hemoglobin level, response rate. Anemia is the most frequent hematologic disorder among cancer patients, 1 and is generally classified as anemia of the chronic illnesses. Several factors are associated with its origin, including deficiencies in endogenous erythropoietin (EPO) and production in response to a reduction in hemoglobin (Hb) levels compared with patients with iron-deficient anemia. 2 The hypoproliferative state of anemia among cancer patients is related to reduced EPO production or reduced response by the bone marrow. 3 As several studies have shown, the treatment of anemia in cancer patients with recombinant human EPO (epoetin- ) is effective in reducing both the degree of anemia and the need for transfusion in a wide range of tumor types. 4 7 According to currently available data, the percentage of patients responding to treatment with EPO varies depending on which response criterion is used. In the study by Abels 2002 American Cancer Society

2 Predicting Factors to Erythropoietin/González-Barón et al et al., 8 a response (defined as an increase in the hematocrit [Hct] value of six or more percentage points) was observed in 48.4% of the patients, whereas Ludwig et al. 7 identified a response (defined as an increase in Hb of 2 g/dl) in 50% of their patients. Bearing these data in mind, the question remains: Which patient has the greatest chance of responding to the treatment? To provide an answer, Ludwig et al. 9 designed an algorithm to predict the response to treatment with EPO. This algorithm is based on the levels of endogenous EPO and increases in Hb level after 2 weeks of treatment. As an alternative, the ferritin level after 2 weeks of treatment was also shown to be a good response predictor. On the basis of these results, Henry et al. 10 observed that the Hb level after 4 weeks of treatment also had a certain predictive power. However, in both the Ludwig et al. and the Henry et al. studies, the overwhelming majority of cancer patients were not receiving chemotherapy, and of those who were, not all their treatments included platinum. Therefore, these algorithms may not be useful in standard clinical practice (cancer patients with anemia receiving treatment with platinum). For example, the baseline levels of endogenous EPO are subject to changes in patients receiving chemotherapy or radiotherapy. 11 There is no reliable way to predict the response to treatment with epoetin- in patients receiving chemotherapy. 12 The current study analyzed factors that might predict the response to epoetin- under real treatment conditions. This was an open-label, observational, multicenter study of a series of cases. The aims were to identify factors that would predict response after 2 and 4 weeks of treatment with epoetin- and to develop a response algorithm on the basis of these factors. MATERIALS AND METHODS The study included patients who were candidates for treatment with epoetin- who fulfilled the following criteria: age 18 years or older; diagnosis of cancer; on chemotherapy treatment with platinum with three further cycles to be received; Hb levels equal to or below 10.5 g/dl at the beginning of treatment with epoetin; no evidence of uncontrolled high blood pressure, iron-deficient anemia, clinically relevant hemorrhage, or active hemolysis; no packed cell blood transfusion in the week before study entry; and no previous treatment with recombinant human EPO. During the period from 1997 to 1998, this study involved 17 Spanish hospitals and 117 patients, 90 of whom were eligible for analysis (i.e., patients who, once included, continued treatment for at least 4 weeks). The following data were compiled before the first dose of EPO: type of tumor, TNM classification, time since the diagnosis of cancer, existence of bone marrow infiltration, and treatment with cisplatin or carboplatin. If the patient had received previous radiotherapy, total percentage of the body irradiated was also noted. The baseline data required on blood parameters were levels of Hb, Hct, reticulocytes (optional), ferritin, iron, transferrin, and transferrin saturation index. Levels of endogenous EPO were also determined (optional) and the process was centralized at a single laboratory. The same data were recorded after 2 and 4 weeks of treatment with epoetin-. Erythropoietin treatment was given in accordance with the data sheet specifications for the product. The dose of epoetin- was 150 IU/kg, given subcutaneously 3 times a week from entry in the study to 1 month after the end of chemotherapy, unless the physician in charge discontinued treatment sooner. No dose adjustments were made at the fourth week of treatment according to the response obtained. Patients who experienced an increase of at least 2 g/dl in Hb levels during the treatment period were considered to be responders to epoetin-. Study Design The study was nonexperimental and prospective and involved a series of cases. It is an appropriate design for identifying factors that influenced the response to the treatment. Sample Size The Ludwig et al. 9 study identified prognostic factors with a predictive value of about 80%. Because this estimate is based on a total sample of 80 patients, it is subject to a sampling error of 9% (at a 95% confidence level). Our study involved an initial sample of 120 patients to reduce this sampling error and reproduce the results of previous studies with relative accuracy. With the final sample of 90 patients eligible for analysis, we identified prognostic factors with a predictive power of 82%. This estimate is subject to a sampling error of 8%, with a confidence interval of 95%. Therefore, the sample of patients obtained reduces the sampling error of previous studies. An intention-to-treat analysis was performed with all patients who remained on treatment without interruption for at least 4 weeks. In the event of premature interruption or blood transfusion, the last available value was carried forward to 1 month after the end of chemotherapy. The final status of the patient (responder or nonresponder) consisted of the status when the interruption or blood transfusion occurred.

3 2410 CANCER December 1, 2002 / Volume 95 / Number 11 Statistical Analysis The analysis to identify predictive factors consisted of several processes. First, the significant differences between responders and nonresponders were compared with initial patient characteristics. This same analysis was performed with each parameter based on the hematologic data. The chi-square or Fisher exact test, as appropriate, was used to compare significant differences related to categorical variables. For quantitative variables, the Kolmogorov Smirnov test was used to establish the advisability of using nonparametric tests. The Student t test or the Mann Whitney nonparametric test was used according to the result. Second, having established that there were no significant differences between groups as far as initial patient characteristics were concerned, a first attempt was made to find possible predictive factors by correlating data with response and using discriminatory step-wise analysis. The data were correlated with response using biserial correlation. The criterion used in the discriminatory analysis to select variables was based on Wilks lambda. Finally, logistic regression was applied to the data to isolate the factors with the greatest predictive power. These results were used to create models that allowed cutoff points to be established to propose rules for predicting response or nonresponse to treatment with epoetin alfa. The maximum verisimilitude method was used in these analyses. In the regression analysis, the goodness of fit of the models obtained was referred at all times to the available sample of cases. To verify the consistency of the predictive power of the resulting model, six independent random samples extracted from the total number of patients were used, each with a sample fraction of 50% (45 patients). RESULTS Twenty-seven patients were excluded from the analysis for the following reasons: 11 patients due to blood transfusion in the first 4 weeks, 5 patients due to death caused by the malignant disease, 4 patients with fewer than three cycles of chemotherapy pending, and 7 patients for whom no follow-up data for the first 4 weeks were available. There was a 63.3% response rate to epoetin- (defined as an increase of at least 2 g/dl in Hb levels during the treatment period, with no transfusions). The patients basic characteristics are shown in Table 1. None of the variables differed significantly between the two groups of patients as defined by response or nonresponse to treatment with epoetin-. No significant differences were found for the baseline TABLE 1 Baseline Characteristics of Patients Baseline characteristics Responders (n 57) Nonresponders (n 33) Total (n 90) Type of tumor Lung Ovarian Head and neck Bone sarcoma Ewing sarcoma 2 2 Cervix Miscellaneous Mean age (yrs) 55.9 (20 86) 52.9 (18 74) 54.8 (18 86) Gender (%) Men 24 (42.1) 15 (45.5) 39 (43.3) Women 33 (57.9) 18 (54.5) 51 (56.7) Bone marrow infiltration (%) 0 (0) 1 (1.8) 1 (1.1) Previous treatment (%) Surgery 34 (59.6) 16 (48.5) 50 (55.6) Chemotherapy 36 (63.2) 16 (48.5) 52 (57.8) Radiotherapy 2 (3.5) 4 (12.1) 6 (6.6) Mean percent irradiated Previous transfusions (%) 6 (10.5) 1 (3) 7 (7.8) Average time since diagnosis (mos) Current treatment (%) Cisplatin 41 (71.9) 25 (75.8) 66 (73.3) Carboplatin 16 (28.1) 8 (24.2) 24 (26.7) Average no. of cycles received Average total dose received (mg) Cisplatin Carboplatin , , data (Table 2). Differences at the second and fourth weeks are shown in Table 2. The absolute values for Hct (P 0.003), Hb (P 0.001), reticulocytes (P 0.003), and transferrin (P 0.042) after 2 weeks of treatment, and the absolute Hct (P ) and Hb (P ) values at 4 weeks, showed significant differences between the two groups. After 2 and 4 weeks of treatment, the variables that best differentiated between the groups were change in Hct (P ) and Hb (P ) values during both periods. There was also a significant difference in the change observed in transferrin level after 2 weeks (P 0.013; Table 3). Possible Prognostic Factors Analysis of the correlations showed that only six variables correlated significantly with response (P

4 Predicting Factors to Erythropoietin/González-Barón et al TABLE 2 Differences between Groups at Baseline and at the Second and Fourth Week (Response, Nonresponse) At baseline Second week Fourth week Responders Nonresponders P a Responders Nonresponders P a Responders Nonresponders P a Hematology Mean red blood cells ( ) ( ) ns ( ) ( ) ns ( ) ( ) Mean hematocrit (%) 28.2 (20 33) 29.1 ( ) ns 30.6 ( ) 27.9 ( ) ( ) 29.6 ( ) Mean hemoglobin (g/dl) 9.3 ( ) 9.6 ( ) ns 10.1 (6 12.2) 9.2 ( ) ( ) 9.6 ( ) Mean reticulocytes (%) 2.7 ( ) 2.4 (0.1 8) ns 3.4 (0.1 7) 2.1 ( ) ( ) 2.66 ( ) ns Biochemistry Mean serum iron 79.7 (7 450) (10 470) ns 89.1 (6 830) (14 410) ns 83.8 (6 400) 98.1 (9 450) ns Mean ferritin (37 838) (13 962) ns (5 846) (7 819) ns 245 (8 901) 347 (9 999) ns Mean transferrin (79 479) (46 655) ns ( ) (80 343) (80 420) (7 410) ns Mean transferrin saturation index 39.5 ( ) 26.1 (6 56) ns 29.9 (7.3 76) 29.9 (5 76) ns 31.1 (6 84) 29.6 ( ) ns Mean endogenous EPO (mu/ml) 69.1 ( ) 84.0 ( ) ns (33 567) ( ) ns ( ) ( ) ns EPO: erythopoietin. a ns: not significant (P 0.05). TABLE 3 Differences between Groups (Response vs. Nonresponse) Characteristics Baseline values a P values at second week a P values at fourth week a Absolute change after Absolute change after 2 weeks (P) a 4 weeks (P) a Hematology Red blood cells ns Hematocrit (%) ns Hemoglobin (g/dl) ns Reticulocytes (%) ns ns ns ns ns Biochemistry Serum iron ns ns ns ns ns Ferritin ns ns ns ns ns Transferrin ns ns ns Transferrin saturation index ns ns ns ns ns Endogenous EPO ns ns ns ns ns EPO: erythropoietin. a ns not significant (P 0.05) ): change in the number of red blood cells (RBCs) in the fourth week (r ), change in Hct level in the fourth week (r ), change in Hb level in the fourth week (r ), change in the number of RBCs in the second week (r ), change in Hct level in the second week (r ), change in Hb level in the second week (r ), change in Hct level in the fourth week (r ), and change in Hb level in the fourth week (r ). These variables were related solely to the values for Hct and Hb (absolute values in the fourth week and changes observed after 2 and 4 weeks of treatment with epoetin). Of these six variables, the changes in the levels of Hb and Hct showed a correlation of over 0.5. Discriminatory analysis of the results from the correlation analysis confirmed that the variables with significant power to differentiate between responders and nonresponders were absolute Hb values (in the fourth week) and changes observed at 2 and 4 weeks. In addition, the best discriminatory function was constructed with a single variable: the change in Hb level after 4 weeks. The results of the classification matrix pointed to the possibility of using this variable as a prognostic fac-

5 2412 CANCER December 1, 2002 / Volume 95 / Number 11 tor. The discriminatory analysis also showed that the absolute value for Hb level after 4 weeks was a better prognostic factor than the change observed after 2 weeks, in terms of the variance explained. This result was consistent with the high correlation between the change in Hb level after 4 weeks of treatment and the absolute value of Hb at that time (r 0.837). Logistic Regression Multiple logistic regression yielded the same results as discriminatory analysis: the change in Hb values after 4 weeks of treatment was the best prognostic factor to explain the response. In the absence of this variable, the absolute value after 4 weeks was also used as a predictive factor. Conversely, the combination of the absolute value and the change observed in the Hb level after 4 weeks as a first-order interaction was not a significant variable in predicting response. Therefore, the adjusted model identified the change in Hb level after 4 weeks of treatment as the best indicator to explain response to treatment with epoetin. Response Algorithm The model adjusted by logistic regression indicated a change in Hb level after 4 weeks equal to or greater than 0.5 g/dl, as opposed to less than 0.5, as a cutoff point to predict response or nonresponse. The response algorithm estimated response better than nonresponse. The power to predict response to treatment (when the rise in Hb level was 0.5) was 89.1%, whereas the power to predict nonresponse (when the Hb level was 0.5) was 71%. In general terms, the predictive power of the algorithm was high. To verify the consistency of the proposed algorithm, six strictly random samples from the total number of patients studied were simulated. Each sample comprised 50% of the total number of patients. The model adjusted as a result of logistic regression analysis was applied to these samples. Values for Endogenous EPO Table 4 shows the distribution of cases according to the interval of EPO values. There were no significant differences between groups for any of these variables. The correlation with response is extremely low: EPO baseline (r ), EPO second week (r ), EPO fourth week (r ), absolute change in EPO second week (r ), absolute change in EPO fourth week (r ), increase in EPO between the second and fourth week (r ). TABLE 4 Distribution of Cases According to the Interval of Values in mu/ml (Using as a Reference the 100 mu/ml Value Given by Ludwig et al. 9 as a Predictor) Baseline (%) Second week (%) Fourth week (%) Nonresponders (n 23) (n 21) (n 19) or over Responders (n 39) (n 33) (n 28) or over Treatment Times for Patients For all patients, median treatment time was 14 weeks (range, 4 28 weeks). For responders, median treatment time was 14 weeks (range, 4 28 weeks). For nonresponders, median treatment time was 15 weeks (range, 4 26 weeks). Adverse Reactions and Transfusions An adverse reaction occurred in only one patient. This patient started treatment with epoetin- on June 18, On September 10, 1997, the patient suffered moderately severe deep venous thrombosis, possibly related to the study drug. No action was taken with regard to the drug. The symptoms remitted without consequences. DISCUSSION The results of this study add to our current knowledge of factors that predict the response to epoetin-. The factors identified in this study as predictive of the response to epoetin- may be considered useful once treatment has begun (i.e., after 2 or 4 weeks of treatment). The Ludwig et al. 9 study is probably one of the most important studies of predictive factors that can be identified after 2 weeks of treatment. Its prediction algorithm is based fundamentally on changes in Hb and endogenous EPO levels after 2 weeks of treatment, with ferritin levels at 2 weeks as an alternative. However, it is important to point out that the population of patients used to develop this algorithm comprised patients with solid and hematologic tumors who received no concomitant chemotherapy. The criterion for response was an increase of at least 2 g/dl in Hb levels after 12 weeks of treatment. In the current study, a change in Hb level after 4 weeks (increase of at least 0.5 g/dl) was the best indicator of response to treatment with epoetin-. To interpret adequately the response algorithm, the following factors must be considered: the 1) patients had solid tumors and

6 Predicting Factors to Erythropoietin/González-Barón et al were undergoing chemotherapy with platinum and 2) the response to epoetin was defined as an increase in Hb level of at least 2 g/dl during the treatment period. These characteristics of the population and the response criterion may explain why neither the data nor the response algorithm differed from those in the study by Ludwig et al. 9 It is not possible to replicate the EPO curve in the study by Ludwig et al. 9 because no association with response was reported. One of the findings of the current study is that the changes in endogenous EPO did not constitute a factor that predicted response. The absence of differences reflected, in part, the small amount of cases (by the second week, no data were available for 42% of the responders and 36% of the nonresponders). Moreover, the values we obtained are not consistent with those given by Ludwig et al. 9 Ludwig et al. 9 reported that EPO levels remained relatively stable in responders but increased in nonresponders during treatment with EPO. Their average and mean values are consistent with our data between the second and fourth weeks, but not with the baseline values for our patients. This was the reason that led us to analyze the changes between the second and fourth weeks, although this comparison yielded no findings of interest. Conversely,the percentage of our patients with values lower than 100 mu/ml was higher in the group of nonresponders, both in the second and fourth weeks. In the algorithm proposed by Ludwig et al., 9 response was predicted only when EPO was less than 100 mu/ml. We note, however, that with a sample size of 80 patients, a significant correlation (P 0.01) was detectable between EPO levels after the second and fourth weeks and response to treatment. As a result, it is difficult to establish a relationship between EPO levels and response with the data currently available to us. Studies of patients with solid tumors undergoing chemotherapy failed to confirm the predictive value of endogenous EPO levels, 13 which probably explains why we did not find any correlation between this variable and the response to treatment. In a large study of cancer patients, Glaspy et al. 14 showed that the increase in Hb after 4 weeks on epoetin treatment was a useful predictor of final Hb increase. In their study, 75.1% of the patients who achieved an increase in Hb greater than or equal to 1.0 g/dl from baseline to Week 4 had an increase of at least 2.0 g/dl by the end of epoetin treatment. Our response algorithm was similar, although our cutoff value to predict response is 0.5 g/dl. The power to predict nonresponse (when the increase in Hb is 0.5g/dL) estimated in our response algorithm is 71%. We expect to see a response in Hb values in a certain number of patients. However, we have not been able to identify likely responders. When we consider the impact of anemia on the quality of life of cancer patients, the best option at this point is to decide with the patient whether to continue/discontinue epoetin treatment. The availability of a response algorithm for patients receiving treatment undoubtedly improves and targets the use of epoetin- for personalized treatment. In this connection, identifying baseline response factors is of considerable interest to identify those patients who will potentially respond to treatment before treatment is begun. Therefore, the need remains for well designed studies to increase our knowledge of response to treatment in patients with cancer and to allow us to extend the data that we currently possess. REFERENCES 1. Johnson R., Roodman G. Hematologic manifestations of malignancy. Dis Mon. 1989;35: Miller CB, Jones RJ, Plantadosi S, et al. Decreased erythropoietin response in patients with the anemia of cancer. N Engl J Med. 1990;332: Saba Hussain I. Anemia in cancer patients: introduction and overview. Cancer Control. 1998;5(Suppl): Henry DH, Abels RI. Recombinant human erythropoietin in the treatment of cancer and chemotherapy induced anemia. Results of double-blind and open-label follow-up studies. Semin Oncol. 1994;2(Suppl 3): Abels R. Erythropoietin for anemia in cancer patients. Eur J Cancer. 1993;29(Suppl 2):S2 S8. 6. Platanias LC, Miller CB, Mick R, et al. Treatment of chemotherapy-induced anemia with recombinant human erythropoietin in cancer patients. J Clin Oncol. 1991;9: Ludwig H, Fritz E, Leitgeb C, et al. Erythropoietin treatment for chronic anemia of selected hematological malignancies and solid tumors. Ann Oncol. 1994;73: Abels RI, Larholt KM, Krantz KD, et al. Recombinant human erythropoietin for the treatment of the anemia of cancer. In: Murphy MJ. Blood cell growth factors: their present and future use in hematology and oncology. Proceedings of the Beijing Symposium. Dayton, OH: Alpha Medical Press, 1991: Ludwig H, Fritz E, Leitgeb C, et al. Prediction of response to erythropoietin in chronic anemia of cancer. Blood. 1994;84: Henry D, Abels R, Larholt K. Prediction of response to recombinant human erythropoietin therapy in cancer patients. Blood. 1995;85: Spivak JL. The physiologic basis for the pharmacologic use of recombinant erythropoietin. Cancer Control. 1998;5 (Suppl): Thatcher N. Predicting the response to r-huepo in anaemic cancer patients undergoing chemotherapy. Erythropoiesis: new dimensions in the treatment of anaemia. 1995;6: Beguin Y. Prediction of response to optimize outcome of treatment with erythropoietin. Semin Oncol. 1998;25 (Suppl 7): Glaspy J, Bukowski R, Steinberg D, at al. Impact of therapy with epoetin alfa on clinical outcomes in patients with nonmyeloid malignancies during cancer chemotherapy in community oncology practice. J ClinOncol. 1997;15: