logistic regression model of fotemustine toxicity combining independent phase ii studies

1980

Logistic Regression Model of Fotemustine Toxicity Combining Independent Phase II Studies

Eric Raymond, M.D.'

Corinne Haon, Ph.D.2

Catherine Boaziz, M.D .~

Maylis Coste, Ph.o.3

' Service deMedecine Interne-Oncologie, H8p- ital Saint-Antoine, Paris, France.

Processus Statistique, Versailles, France.

lnstitut de Recherches lnternationales Servier (IRIS), Courbevoie, France.

Presented in part as an abstract in the Proceed- ings of the 1995 Annual Meeting of the Ameri- can Society of Clinical Oncology, Los Angeles, CA, May 20-23, 1995.

The authors grateful acknowledge Mrs. C. Henry for her help in preparing the article, and special acknowledgments to Dr. D. Moccati and Dr. B Giroux for their critical reading of the article.

Address for reprints: Dr. Eric Raymond, Insti- tute for Drug Development, 14960 Omicron, San Antonio, TX 78245-3217.

Received May 6, 1996; revision received July 8, 1996; accepted July 8, 1996.

BACKGROUND. To optimize fotemustine chemotherapy, the authors considered how to combine independent Phase 11 trials to predict the risk of first occurrence of severe toxicity as a function of initial patient characteristics. METHODS. Clinical data from six Phase I1 trials were collected. Of the 478 patients enrolled, 442 (malelfemale, 2591183; age range, 15-81 years) were evaluable for toxicity (1384 cycles of chemotherapy), including 221 with malignant melanomas, 138, with primary brain tumors, 29 with lung carcinomas, 8 with head and neck carcinomas, and 46 with miscellaneous cancers. The influence of age, sex, performance status, type of tumor, number and location of metastases, and previous treatment by chemotherapy andlor radiotherapy was studied. The logistic regression method was applied to predict occurrence of leukopenia, thrombocytopenia, anemia, digestive tract, and/or hepatic toxicity. RESULTS. Univariate analysis showed that predictive factors for hematologic toxicity were age (> 50 years), type of tumor (brain < melanoma < others), number of rnetastatic sites (> 3), location of metastases (nonvisceral), and previous chemotherapy. Performance status and previous radiotherapy did not affect the toxicity of fotemustine. Nausea and vomiting were predictable based on the type of tumor (head and neck < lung < brain < melanoma < others), the number of metastatic sites (> three), and visceral metastases. Hepatic disorders occurred preferentially in patients with hepatic metastases and more than three metastatic sites. Individual risk. of hematologic and hepatic toxicity for patients with melanoma and primary brain tumors was predicted using logistic regression models. CONCLUSIONS. By combining clinical data from independent Phase 11 trials, the logistic model developed could predict the probability of fotemustine hematologic and hepatic toxicity. Cancer 1996; 78:1980-7. 0 1996 American Cancer Society.

KEYWORDS fotemustine, nitrosourea, multifactorial analysis, logistic regression method, chemotherapy, toxicity, quality of life.

oxicity of chemotherapy remains an important factor in the care, T management, cost of treatment, and quality of life of cancer patients with a particularly poor prognosis. However, information pro- vided by independent small Phase 1/11 trials are often unusable to define high risk patients and for most cytotoxic drugs, including nitrosoureas, there is little correlation between pharmacokinetic or phar- macodynamic parameters and the main toxicity.'-3 Therefore, new methods are warranted to predict patient at risk for common toxicity.

Fotemustine (diethyl 1- [3-(2 chloroethyl) 3 nitrosoureido] ethyl phosphonate) is a novel alkylating agent characterized by the grafting of a phosphonoalanine group onto the nitrosourea radical with conse- quent high lipophilia and improved diffusion through the cell mem- brane and the blood-brain barrier.4 Since 1984, fotemustine has been extensively studied in patients with malignant me lan~rna .~ - '~ Its activity has been widely assessed in multicentric European Phase I1

0 1996 American Cancer Society

A Logistic Model of Fotemustine ToxicitylRaymond et al. 1981

studies with an overall response rate of 24% interestingly preserved in cerebral metastases. Activity has also been observed in some supratentorial malignant gliomas and activity has been described in advanced non small cell lung carcinoma^.'^-'^ Fotemustine chemotherapy is known to be well tolerated on an outpatient basis. The main toxicity of fotemustine was my- elosuppression with delayed reversible thrombocytopenia (40.3%) and leukopenia (46.3%), which were cumulative and dose-related as observed with other nitrosoureas. Gastrointestinal toxicity was infrequent (7.7%). Mild transient increases in serum alanine aminotransferase, aspartate aminotransferase, alkaline phosphatase, and bilirubin values were reported in 29% of patients, suggesting moderate hepatotoxicity. Although fotemustine toxicity has been graded, the characteristics of patients who are at a greater risk have not been determined using common criteria. Therefore, guidelines to identify the individual risk of toxicity on treatment need to be established. Until now, the number of patients included in Phase I1 trials has usually been too small to perform multifactorial analysis and obtain individual prognostic factors re- garding toxicity.

Over the last decade, the logistic regression method has become a useful alternative to the linear regression procedure in cases in which the outcome variable is discrete, taking on two or more possible value^.'^-'^ This method has been proposed in data analysis concerned in describing the relationship between an outcome variable (such as toxicity) and one or more clinical explanatory variables. It has been used in the estimation of individual attributable risks of single agent chemotherapy t o x i ~ i t y . ~ ~ ~ " - ~ ~ To identify clinical factors associated with fotemustine toxicity and to predict a patient's individual probability of first occurrence of severe toxicity, an analysis combining single agent Phase I1 trials was performed and subse- quently elaborated predictive models using the multi- ple logistic regression method was elaborated.

MATERIALS AND METHODS Patient Characteristics Clinical trials performed with the same dose and schedule comprised of a weekly induction of intrave- nous fotemustine at a dose of 100 mg/m' infused over 1 hour for 3 consecutive weeks followed by a 4- or 5-week rest period were selected. This schedule was recommended in the Phase I study published by Khayat et al.4 Individual clinical and laboratory records of 478 patients were obtained from 6 Phase I1 trials. This represent greater than 90% of available data performed in Europe using single agent fotemustine. Thirty-six patients were nonevaluable in this analysis because information relating to clinical and laboratory

parameters and toxicity was incomplete. Statistical analysis was performed on the remaining 442 patients (1384 cycles of chemotherapy). Informed consent had been obtained from patients included in these Phase I1 studies. Minimal inclusion criteria used in the 6 trials were: histologic evidence of advanced carcinoma; measurable and/or evaluable evidence of progressive disease when entering the study; no specific therapy given within the 4 weeks prior to entry into the study or within 6 weeks if the patient had previously received a nitrosourea; a life expectancy of at least 8 week;; Karnofsky performance status greater than 50%; no other acute illness; adequate cardiac, renal (creatinine < 100 ,umol/L), hepatic (< 20% above upper limit of normal for each parameter), and hematologic func- tions (neutrophil count > 1500/mm3, platelets > 100,000/mm3, hemoglobin > 10 g/dL); and no concomitant hepatotoxic, hematotoxic, or cytotoxic treatment.

Three trials including advanced malignant melanomas (155, 30, and 26 evaluable patients, respectively), 2 trials including nonoperable advanced primary glioblas- tomas (63 and 59 evaluable patients, respectively), and 1 trial combining different types of advanced malignancies (109 evaluable patients) were analyzed. The main characteristics of evaluable patients are summarized in Table 1. There were 259 men and 183 women, with a median age of 54 years (range, 15 to 81 years); most had malignant melanoma (50%) or a primary malignant brain tumor (31%). The tumoral lesions were grouped according to metastatic site and organized into a hierarchy of three categories based on the predominant metastatic site: visceral (including hepatic and/or lung metastases, iso- lated, or associated with nonvisceral or cerebral lesions), nonvisceral (cutaneous, subcutaneous, lymph nodes, bone, or other sites except for cerebral sites) or cerebral. Patients with malignant primary brain tumors (glioblas- tomas), as expected, had no metastases (31%), whereas others had 1 or 2 sites in 50% of cases and more than 2 sites in 19% of the cases. Most patients had received prior treatment; 185 (42%) had received radiotherapy and 210 (47.5%) had been treated by 1 (137 patients) or more (73 patients) chemotherapy regimens (mean number of cycles: 4; range, 1 to 99 cycles). Fifty-one patients (24%) had previously received a nitrosourea. In keeping with the inclusion criteria, hematologic, renal, and liver parameters were normal at inclusion and no major disease other than cancer was present. Some patients with brain tumors received concomitant treatment including anticonvulsant therapy with phenobarbital and antiedema therapy with steroids.

Toxicity Toxicity was assessed before each administration and during the rest period by clinical and laboratory exam-

1982 CANCER November 1, 1996 I Volume 78 / Number 9

TABLE 1 Patient Characteristics

Age (yrs) Sex

Male Female

90-100% 7040% 50-60%

Karnofsky performance status

> 50% (value unspecified) Types of carcinoma

Lung Head and neck Breast Colorectal Renal carcinoma Sarcoma Melanoma Malignant brain tumors Hematologic Ovarian carcinoma Gastric carcinoma Ampuloma

No. of metastatic sites 0 1-2 2 3

Brain Visceral Nonvisceral

No Yes Unknown

No Yes 1 line regimen 2 line regimens Mean number of courses

Predominant metastatic sites

Previous radiation therapy

Previous chemotherapy

54 (range, 15-81)

259 (58.6%) 183 (41.456)

190 (43%) 162 (36%) 66 (15%) 24 (5%)

29 (6%) 8 (2701 7 (2%) 9 (2%) 6 (l%l 10 (2%) 221 (50%) 138 (31%) 11 (2%) 1 1 I

138 (31%) Brain tumors 223 (50%) 81 (19%)

128 115 82

254 (57%) 185 (42%) 3 (1%)

232 (52.5%) 210 (47.5%) 137 73 4 (range, 1-99)

inations (i.e., complete blood count and platelets, blood urea nitrogen, creatinine, transaminases, alkaline phosphatase, and bilirubin). Toxicity was graded using the World Health Organization (WHO) criteria. Leukopenia, anemia, thrombocytopenia, nausea and vomiting, and liver function test abnormalities were considered. Because a modification of the dose after the first occurrence of toxicity was likely to affect the probability of occurrence of further toxicity, it was de- cided to consider only the first occurrence of severe (Grade 3 and 4) toxicity in this study.

Selection of Explanatory Variables The following potential explanatory variables were selected: age, sex, performance status, pathology, site of primary tumor, number and location of metastatic

sites, and previous treatments, i.e., radiotherapy and chemotherapy (number of regimens and respective courses). According to selection criteria in Phase I1 trials, all patients had laboratory parameters within the normal range at entry to the study, thereby excluding the possibility of using any abnormal biologic parameters as explanatory variables of toxicity. The clinical data display and contingency tables allowed organization of the covariates into several classes based on clinical observation and sample sizes: age (5 50 years/> 50 years), sex (maleIfemale), performance status (90- 100%/70-80% /50- 60%), pathology (primary brain tu- morslmelanomallung + head and necklothers), number of metastatic sites (none/ 1 or 2 / 2 3), type of metastatic site (brain/visceral/non visceral), previous radiotherapy (yedno), and previous chemotherapy (nonell regimen up to 6 courses/more than 1 regimen or more than 6 courses). Because primary brain tumors gener- ally do not metastasize, when the number of metastases and metastatic sites were analyzed as covariates, primary brain tumors were removed to avoid instability of the model. The logistic method was applied to compare none and mild toxicity (Grade 1 and 2) to severe toxicity.

The Logistic Regression Method The goal was to find the best fitted model to describe the relationship between a discrete outcome variable and a set of independent explanatory variables called covariates. The tools used were the same as for any model building technique used in statistics. A logistic regression model may be distinguished from a linear regression model by the fact that the outcome variable in logistic regression is binary or dichotomous. This difference between logistic and linear regression is re- flected both in the choice of a parametric model and in the assumptions. The methods employed in logistic regression analysis follow the same general principles as those used in linear r e g r e s s i ~ n . ~ ” ~ ~ In the current study, the outcome variable was toxicity, whereas the explanatory variables were defined by the baseline values of the clinical parameters.

The logistic regression method expressed the probability (P) to observe a toxicity according to the explanatory variables, as follows:

P = exp(u)/[l + exp(u)]

in which u is a linear function (u = bo + blx, + b2x2 + * . * + bixi + e s * + b,x,) of explanatory (xi) variables and b, (i = 1, . . . , p) the logistic model parameters. A more convenient of the model expression is based on the logit transformation on P, as follows: Log [P/(1 - PI1 = u.

The logit is a linear function of the explanatory variables. The parameters of the logistic model are

A Logistic Model of Foternustine Toxicity/Raymond et al. 1983

estimated using the maximum likelihood method. The relationship between the outcome variable and each covariate was appraised by an odds ratio. When a covariate of interest has two levels with the level of interest and the reference level coded as x, = 1 and 0 respec- lively, the odds ratio is the ratio of the probability of toxicity at the interested level over the probability of toxicity at the reference level. The odds ratio can be estimated by the exponential of the estimator of bi. The selection criteria of explanatory variables were based on the likelihood ratio test (G-test). The modelization was performed by a two-step procedure. The first step was comprised of univariate logistic regression for each covariate that enabled elimination of variables with a G-test value greater than 0.25. The second step was comprised of a multivariate analysis to select a subset of independent variables that could explain the outcome variables. The selection of explanatory variables followed a stepwise forward inclusion then backward elimination; the threshold covariate value was 0.10 to enter a variable in the model and 0.15 to remove this variable from the model. The suitable way to summarize the results of a fitted logistic model is via a classification table. This table is the result of cross-classifymg the outcome variable with a dichotomous variable whose values are derived from the estimated logistic probabilities. To obtain the derived dichotomous variable, a cutoff point must be defined and each estimated probability compared with the cutoff point. If the estimated probability ex- ceeds the cutoff point value, a value of 1 is attributed; otherwise, it is equal to 0. The cutoff point value that maximizes the overall rate of correct classification pro- duces a threshold value that may be used to predict the group membership of a new patient.

Using the model, it was then possible for each group of patients to calculate the probability of toxicity. The graphs obtained show the estimated probability of toxicity for several groups of patients. The size of the spot on the graph was proportional to the number of patients included in the group. If a probability of toxicity was greater than the threshold value, a patient with the same baseline characteristics presented a risk of toxicity. Three goodness-of-fit tests were per- f ~ r m e d : ~ ’ chi-square test (comparison between the predicted and the observed frequencies), Hosmer- Lemeshow’s test (comparison between the predicted and the observed frequencies with a calculation based on the predicted probability), and C.C. Brown’s test (which verifies that the logistic regression is the right model for the multifactorial toxicity analysis). For the three tests, a high P value confirmed lack of evidence of a poor fit.

The software used included PC-Focus,@ BMDP,@

and SYSTATO on COMPAQ@ (PC-386/25) and TOSHI- BA@ (T4400SX) hardware (MsDOS 5.00@).

RESULTS Severe leukopenia, thrombocytopenia, and anemia occurred in 38%, 37%, and 14% of patients, respectively. Few patients presented with severe gastrointestinal toxicity (9%) or disordered liver function tests (12%). Toxicity rates are shown in Table 2 .

Hematologic toxicity occurred after the first induction chemotherapy. It was investigated whether all dose modifications were performed after the first occurrence of leukopenia. It could then be assumed that all the patients received full doses of fotemustine when the first occurrence of leukopenia was considered for statistical analysis. Planned doses of fotemustine were given to 128 patients (95%) and 35 (72%) patients, respectively, before the first occurrence of thrombocytopenia and anemia. The number of patients in the group receiving decreased doses was too small to allow separated statistical analysis. Planned doses of fotemustine were given to 340 patients (80%) and 200 patients (52%), respectively, before the first occurrence of digestive and hepatic toxicity. Because no dose-related relationships has been reported with nitrosoureas for digestive and hepatic toxicity, individual dose level has not been considered as a covariate in the multifactorial analysis of toxicity.

Univariate Analysis Univariate analysis showed that predictive factors for hematologic toxicity were age (older than 50 years), type of tumor (brain < melanoma < other malignancies), number of metastatic sites (> 3 sites), metastatic site (nonvisceral), and previous chemotherapy. Nau- sea and vomiting could be predicted by type of tumor (head and neck < lungs < brain tumors < melanoma < other malignancies), number of metastatic sites (more than three sites), and visceral metastases. He- patic disorders occurred preferentially in patients with hepatic metastases and more than three metastatic sites. Sex, performance status, and previous radiation therapy did not show any relationship to toxicity. These variables were not therefore taken into account in the multifactorial analysis.

Multivariate Analysis Multivariate analysis showed that the significant prognostic factors for severe hematologic toxicity were an age older than 50 years, the type of tumor (brain < melanoma < others) and previous chemotherapy. An age older than 50 years increased the risk of leukopenia (odds ratio = 2.7) and thrombocytopenia (odds ratio = 1.6) but did not increase the risk of anemia. Patients with brain tumors had a lower risk of hemato-

1984 CANCER November 1, 1996 / Volume 78 I Number 9

TABLE 2 Toxicity According to the World Health Organization Classification

Toxicity NO. Grade 0 Grade 1 Grade 2 Grade 3 Grade 4

Leukopenia 353 93 (26%) 48 (14%) 75 (21%) 85 (24%) 52 (15%) Thrombocytopenia 357 98 (27.5%) 56 (16%) 68 (19%) 84 (23.5%) 51 (14%) Anemia 351 182 (52%) 71 (20%) 51 (14%) 34 (10%) 13 (4%) Nauseahomiting 425 207 149%) 74 117%) 106 (25%) 36 (8.5%) 2 10.5%) Hepatic toxicity 385 202 (52%) 76 (20%) 60 (16%) 22 (6%) 25 (6%)

logic toxicity compared with all other types of tumors (odds ratio range, 3.5-5.7). In all cases, previous chemotherapy increased the risk of severe hematologic toxicity (odds ratio range, 2-3.5), but there was no correlation between the number of cycles or the number of previous regimens of chemotherapy and hematologic toxicity.

The risk of gastrointestinal and hepatic toxicity depended only on the number and type of metastatic site. Head and neck tumors and lung carcinomas were associated with a significantly lower risk of nausea and vomiting (odds ratio = 2.3). Patients with liver metastases had a significantly higher risk of hepatic toxicity (odds ratio = 5).

Modelization of Toxicity A well adjusted modelization was obtained for thrombocytopenia and hepatic toxicity. Models to predict anemia and leukopenia were also obtained with low goodness-of-fit test values (Fig. 1). This current study failed to modelize digestive toxicity.

Age older than 50 years, prior chemotherapy, and type of tumor were the variables selected by the multistep logistic regression model to predict hematologic toxicity. Interestingly, the number of courses of chemotherapy did not increase the risk of heinatologic toxicity. Although previous chemotherapy with nitrosoureas was associated with an increased risk of hematologic toxicity compared with no previous chemotherapy, the risk was not significantly different from that of other types of chemotherapy.

Thrombocytopenia Patients with primary brain tumors had a low risk of thrombocytopenia regardless of their age or prior therapy. A significantly higher risk of thrombocytopenia was observed for patients with melanoma aged older than 50 years and previously treated with chemotherapy. The risk of toxicity was high for patients with other tumors except for patients aged 50 years or younger who were previously untreated.

Leukopenia All patients with primary brain tumors had a low risk of leukopenia even if they were elderly or had previously

received chemotherapy. As was the case for thrombocytopenia, patients with an increased risk of leukopenia were those aged older than 50 years and previously treated with chemotherapy. Prior chemotherapy was the most important explanatory variable predicting leukopenia. Patients with a high risk of leukopenia were again those with tumors other than melanoma or primary brain tumors, except for previously untreated patients aged younger than 50.

Anemia Groups of patients at low risk of severe anemia were those with primary brain tumors and those with previously untreated melanoma. Patients who had previously been treated (for melanoma or other tumors) had a high risk of anemia.

Hepatic toxicity The model for hepatic toxicity showed that the most important variable for predicting hepatic toxicity was the presence of hepatic metastases. Patients with primary brain tumors had a lower risk of hepatic toxicity.

Subgroup analysis A separate analysis performed in patients with melanoma showed that predictive parameters of leukopenia and thombocytopenia were an age of 50 years and previous chemotherapy. Similar analysis performed in patients with primary brain tumors confirmed that in this group, previous chemotherapy and advanced age did not affect the risk of hematologic toxicity.

DISCUSSION Toxicity of chemotherapy may reduce clinical drug activity through dose reduction or treatment delay. Moreover, it alters the patient’s quality of life, and may significantly increase the cost of treatment. The current understanding of drug-related toxicity is obtained after assessment of dose-limiting toxicity and cumulative toxicity from Phase I and Phase I1 trials. However, the small numbers of patients included in these prospective trials have prevented any relevant multifactorial analysis of toxicity to clearly define the profile of patients at higher risk. In some retrospective

A Logistic Model of Fotemustine Toxicity/Raymond et al. 1985

1 0

0.9

0.8

0.7

0.6

0.5

0.4

1 .o

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0

I I 1 I I I 4 D -. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ._

. . . . . . . . . . . . . . . . . . . . . . . . . . .

- . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .......-

-, .................. . . . . . . . . . . . . . . . . . . . . . . . . .-

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .-

. . . . . - . . T . " ' '-

t i

0.1

I .]...+ I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . --

! l ! l , , l I , , , I I I I I I I

e a c f h k g q m d i l h o r n j p

1 1 Cut-off point .................. .~~ .... t ....................... ~

-. . . . . . L . . . . . . . . . . . . . . . . . a " ' € I L I

A

cut-off point

0.1

10

0.9

0.8

0.7

0.6

0.5

0.4

0.3

02

0.1

0.0

I -. . . . . . . . . . . . . . . . . . . . . . . . . . . ........................

8 ,

....................................................................

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.....................................................................

. . . . . . . . ..... ....... t ..............................

I I I 1 1

ah ef gh cd kl mn i j qr op

GROUPS OF PATIENTS CROUPS OF PATIENTS

C

cut-off point

03 c 02 c T T I4

GROUPS OF PATIENTS GROUPSOF PATIEWTS

FIGURE 1. Predicted probability of fotemustine toxicity. The size of the spots is proportional to the number of patients. Standard deviation values are symbolized by vertical lines. An increased risk of toxicity could be predicted if the probability of toxicity is above the cutoff point (dashed line). (A) Risk of leukopenia (cutoff point = 0.43; chi-square value 8.9, P = 0.7; Hosmer-Lemeshow value 2.5, P = 0.9; C.C. Brown's test value 1.9, P = 0.38). (B) Risk of thrornbocytopenia (cutoff point = 0.40; chi-square value 23.6, P = 0.02; Hosrner-Lemeshow value 14.1, P = 0.08; C.C. Brown's test value 12.1, P = 0.002). ( C ) Risk of anemia (cutoff point = 0.16; chi-square value 17.2, P = 0.19; Hosrner-Lerneshow value 4.8, P = 0.7; C.C. Brown's test value 4.5, P = 0.1). (D) Risk of hepatic toxicity (cutoff point = 0.12; chi-square value 10.7, P = 0.005; Homer-Lemeshow value 9.3, P = 0.026; C.C. Brown's test value 10.6, P = 0.005). Groups were defined as follows:

Brain Hematologic toxicity tumors Melanoma Other tumors

Age 5 501no previous chemotherapy Age > 501no previous chemotherapy Age 5 501s 1 previous regimen of chemotherapy, s 6 courses Age > 501s 1 previous regimen of chemotherapy, s 6 courses Age 5 5012 2 previous regimens of chemotherapy or > 6 courses

d e f Age > 5012 2 previous regimens of chemotherapy or > 6 courses

Hepatic toxicity No metastases 1-2 metastatic sites t 3 metastatic sites

Primary brain tumors S 0 0 Brain metastases 0 t U Liver metastases 0 V W Other metastatic sites 0 X Y

1986 CANCER November 1, 1996 / Volume 78 / Number 9

analyses, the toxicity of alkylating agents appears to be related to particular clinical categories such as older age, poor general state of health, and previous chemotherapy or radi~therapy.~ ' ,~~

Fotemustine is a nitrosourea mainly used in the treatment of patients with advanced melanoma or primary brain tumors. Phase I1 studies revealed that its toxicity was mild and similar to that observed with other nitrosoureas. Although previous s t~d ie s~ ,* ,~ ' showed that hematologic toxicity of fotemustine is delayed, dose-related, and cumulative, affecting both leukocytes and platelets, the risk of severe myelosup- pression was unpredictable after the first infusion, re- quiring monitoring of hematologic parameters. It was speculated that if the individual risk of hematologic toxicity could be predicted before a patient receives chemotherapy, the use of hematopoietic growth factors to prevent hematologic toxicity could be opti- mized and the cost of treatment reduced. Apart from hematologic toxicity, fotemustine is a well tolerated outpatient chemotherapy with only mild gastrointestinal and hepatic t o x i ~ i t y . ~ . ~ - ' ~ , ~ ~ In the current study, patients treated with the same monotherapeutic Phase I1 regimen were selected and studied to learn if most received the planed dose of chemotherapy before the first occurrence of hematologic toxicity. Therefore, it was possible to conduce further analysis without including dose levels as covariates for hematologic toxicity.

The models in the current study showed that initial patient characteristics predicted individual probabilities of biologic hepatic disorders, leukopenia, and thrombocytopenia. Interestingly, it appears that patients with primary brain tumors have a low risk of hematologic toxicity whatever their age and previous treatments. It was hypothesized that concomitant ste- roid or anticonvulsant treatments commonly used in brain tumors may modify the pharmacokinetics of nitrosoureas and result in a low toxicity rate. However, because the data concerning concomitant drug therapy were not fully registered in Phase 11, this specula- tion could not be confirmed. Moreover, further pharmacokinetics analysis of fotemustine in combination with steroids are warranted to ensure that the reduced toxicity will not decrease antitumor efficacy. For patients with melanoma, the most important variables predicting hematologic toxicity are an age older than 50 years and previous chemotherapy. The impact of alkylating agents on hematopoietic stem cells is well known and previous studies have demonstrated that aging andlor previous chemotherapy are the most important factors predicting the hematologic toxicity of alkylating agents3 This study allows quantification of the risk of hematologic toxicity and helps to define groups of patients according to their risk. The reason

why the number of previous regimens of chemotherapy did not increase the risk of toxicity remains un- clear. Patient records were analyzed and no specific characteristics were found. In fact, only a small number of patients with melanoma received more than two previous regimens of chemotherapy. The small size of this group could explain the low rate of toxicity. Although patients with a high risk of toxicity with single agent fotemustine are probably exposed to a similar or higher risk if fotemustine is administered in combination, it should be emphasized that the study does not provide direct information on the toxicity of fotemustine in combination with other cytotoxic drugs (i.e., dacarbazine and cisplatin) or immunotherapy (i.e., interferon and interleukin-2). Moreover, results must be cautiously interpreted for patients with cancers other than melanoma or primary brain tumors, because of the small numbers of patients included in the study. This study shows that the logistic regression method may be used to estimate the risk of fotemustine toxicity. It is hoped that this approach may be extended to new alkylating agents in conjunction with proper use of pharmacokinetics to obtain accurate prediction of toxicity.

1.

2.

3.

4.

5.

6.

7.

8.

9.

Chalmers TC, Levin H, Sacks HS, Reitman D, Berrier J, Na- galingam R. Meta-analysis of clinical trials as a scientific discipline: control of bias and comparison with large coop- erative trials. Stat Med 1987;6315-25. Berlin JA, Laird NM, Sacks HS, Chalmers TC. A comparison of statistical methods for combining event rates from clinical trials. Stat Med 1989;8:141-51. Colvin M, Chabner BA. Alkylating agents. In: Chabner BA, Collins JM, editors. Cancer chemotherapy principles and practice. 2nd edition. Philadelphia: J.B. Lippincott, 1990:

Khayat D, Lokiec F, Bizzari JP, Weil M, Meeus L, Jacquillat C. Phase I clinical study of the new amino acid-linked nitrosourea, S 10036, administered on a weekly schedule. Cancer Res 1987;47:6782-5. Aapro MS. Advances in systemic treatment of malignant melanoma. Eur J Cancer 1993;29:613-7. Retsas S, Bayliss M, Sheikh N, Chong SL, Bafaloukos D. Che- motherapy of malignant melanoma: the European experi- ence. Rev Med Interne 1990; 11:51-8. Jacquillat C, Khayat D, Banzet P, Weil M, Fumoleau P, Bizzari JP. Final report of the French multicenter phase I1 study of the nitrosourea fotemustine i i I 153 patients with disseminated malignant melanoma including patients with cerebral metastasis. Cancer 1990;66:1873-8. Khayat D, Cour V, Bizzari JP, Aigner K, Bore1 C, Jacquillat C. Fotemustine (S 10036) in the intra-arterial treatment of liver metastasis from malignant melanoma. A phase I1 study. Am J Clin Oncol 1991; 14:400-4. Khayat D, Avril MF, Gerard B, Bertrand P, Bizzari JP, Caur V. Fotemustine: an overview of its clinical activity in disseminated malignant melanoma. Melanoma Res 1992;2:147-51.

276-313.

A Logistic Model of Fotemustine Toxicity/Raymond et al. 1987

10. Khayat D, Bizzari JP, Frenay M, Naman H, Weil M, Jacquillat C. Interim report of phase I1 study of new nitrosourea S 10036 in disseminated malignant melanoma. / Nut1 Cancer Inst 1988; 17:1407-8.

11. Aapro M, Becquart D, Kleeberg U, Wils J, Michel J, Gerard B. International phase I1 trial of fotemustine in advanced malignant melanoma. Invest New Drugs 1989; 9:436.

12. Jacquillat C, Khayat D, Banzet P, Weil M, Fumoleau P, Bizzari JP. Multicentric study of the nitrosourea fotemustine (S 10036) in advanced malignant melanoma (AMM) including patients with cerebral metastases. In: Jacquillat C, Weil M, Khayat D, editors. Neo-adjuvant chemotherapy. Paris: John Libbey Eurotext Lid, 1988:787-90.

13. Jacquillat C, Khayat D, Banzet P, Weil M, Avril MF, Bizzari LP. Chemotherapy by fotemustine in cerebral metastases of disseminated malignant melanoma. Cancer Chemother Pharmacol 1990;25:263-6.

14. Grisoli F, Namer M, Bourdin S, Poisson M, Derlon JM, Giroux B. Fotemustine (S 10036) in the treatment of malignant $0-

mas in adults. Invest New Drugs 1989;9:431. 15. Khayat D, Giroux B, Berille J, Cour V, Gerard B, Bizzari JP.

Fotemustine in the treatment of brain primary tumors and metastases. Cancer Invest 1994; 12414-20.

16. Frenay M, Giroux B, Khoury S, Derlon JM, Namer M. Phase I1 study of fotemustine in recurrent supratentorial malignant gliomas. Eur / Cancer 1991;27:852-6.

17. Le Chevalier T, Monnier A, Douillard JY, Pujol JL, Riviere A, Bizzari JP. Phase I1 study of fotemustine in the treatment of inoperable non small cell lung carcinoma (NSCLC): final report. Eur / Cancer 1993; 1:5161.

18. Frenay M, Biron P, Poisson M, Giroux B. Tumor exposure to fotemustine in malignant glioma. The higher the better? E u r / Cancer 1993; 1:5189.

19. Efron B. The efficiency of logistic regression compared to normal discriminant analysis. / A m Stat Assoc 1975; 70:892- 8.

20. Moriguchi S, Hayashi Y, Nose Y, Maehara Y, Korenaga D, Sugimachi K. A comparison of the logistic regression and the Cox proportional hazard models in retrospective studies on the prognosis of patients with gastric cancer. /Surg Oncol

21. Drescher K, Schill W. Attributable risk estimation from case- 1993;52:9- 13.

control datavia logistic regression. Biometrics 1991;47:1247- 56.

22. Moses LE, Shapiro D, Littenberg B. Combining independent studies of a diagnostic test into summary ROC curve: data- analytic approaches and some additional considerations. Stat Med 1993; 12:1293-316.

23. Rao VR, Levy K, Lustik M. Logistic regression of inhalation toxicities of perchlorethylene: application in noncancer risk assessment. Regut Toxicol Pharmacol 1993; 18233-47.

24. Preston Martin S, Yu MC, Benton B, Henderson BE. N-Ni- troso compounds and childhood brain tumors: a case control study. Cancer Res 1982;42:5240-5.

25. Coughlin SS, Nass CC, Pickle LW, Trock B, Bunin G. Regres- sion methods for estimating attributable risk in population- based case-control studies: a comparison of additive and multiplicative models. A m / Epidemiol 1991; 133:305- 13.

26. Brazer SR, Pancotto FS, Long TT, Harrell FE, Lee KL, Pryor DB. Using ordinal logistic regression to estimate the likelihood of colorectal neoplasia. / Clin Epidemiol1991; 44:1263- 70.

27. Hosmer DW, Lemeshow S. Applied logistic regression. 1st edition. New York: John Wiley & Sons, 1994:l-80.

28. Mc Cullagh P. Regression models for ordinal data. / R Stat

29. Gordon NH, Willson JK. Using toxicity grades in the design and analysis of cancer phase I clinical trials. Stat Med

30. Chevret S. The continual reassessment method in cancer phase I clinical trial: a simulation study. Stat Med

31. Hosmer DW, Lemeshow S. Goodness-of-fit tests for the mul- tiple logistic regression model. Commun Statist PartA Theor Meth 1980; 10:1043-69.

32. Baldwin JG. Hematopoietic function in the elderly. Arch In- tern Med 1988; 1482544-6.

33. Newel1 DR, Gore ME. Toxicity of alkylating agents: clinical characteristics and pharmacokinetic determinants. In: Powis G, Hacker MP, editors. The toxicity of anticancer drugs. New York Pergamon Press, 1991:44-62.

34. Calabresi F, Aapro M, Becquart D, Dirix L, Wils J, Gerard B. Multicenter phase I1 trial of the single agent fotemustine in patients with advanced malignant melanoma. Ann Oncol

SOC 1980; 42: 109-42.

1992; 112063-75.

1993; 12:1093-108.

1991; 2~377-8.

logistic regression model of fotemustine toxicity combining independent phase ii studies

Documents