Survival and Prognostic Stratification of 670 Patients
`With Advanced Renal Cell Carcinoma
`
`By Robert J. Motzer, Madhu Mazumdar, Jennifer Bacik, William Berg, Alison Amsterdam, and Joseph Ferrara
`
`Purpose: To identify prognostic factors and a model
`predictive for survival in patients with metastatic renal-
`cell carcinoma (RCC).
`PatientsandMethods: The relationship between pre-
`treatment clinical features and survival was studied in
`670 patients with advanced RCC treated in 24 Memorial
`Sloan-Kettering Cancer Center clinical trials between
`1975 and 1996. Clinical features were first examined
`univariately. A stepwise modeling approach based on
`Cox proportional hazards regression was then used to
`form a multivariate model. The predictive performance
`of the model was internally validated through a two-
`step nonparametric bootstrapping process.
`Results: The median survival time was 10 months
`(95% confidence interval [CI], 9 to 11 months). Fifty-
`seven of 670 patients remain alive, and the median
`follow-up time for survivors was 33 months. Pretreat-
`ment features associated with a shorter survival in the
`multivariate analysis were low Karnofsky performance
`status (F80%), high serum lactate dehydrogenase (G 1.5
`times upper limit of normal), low hemoglobin (F lower
`
`RENAL CELL CARCINOMA (RCC) is the most com-
`
`mon tumor arising in the kidney, affecting approxi-
`mately 30,000 individuals each year in the United States.1,2
`The outlook for patients with distant metastases is poor, with
`a 5-year survival rate of less than 10% for patients present-
`ing with stage IV disease.1,2 This reflects the lack of effective
`systemic therapy for patients with metastases. RCC is
`resistant to chemotherapy and hormonal therapy because no
`agent consistently achieves a response in more than 10% of
`patients.3 Immunotherapy, ie, interleukin-2 and interferon
`alpha, achieves responses in 10% to 20% of patients.1 However,
`the low response rate, toxicity associated with high-dose regi-
`mens,4 and few long-term survivors after treatment with inter-
`feron-alpha or interleukin-2 provide the rationale for clinical
`trials as a priority for management of patients with this disease.
`
`From the Genitourinary Oncology Service, Division of Solid Tumor
`Oncology, Department of Biostatistics and Epidemiology, Memorial
`Sloan-Kettering Cancer Center; and Department of Medicine, Cornell
`University Medical College, New York, NY.
`Submitted January 13, 1999; accepted April 12, 1999.
`Supported in part by National Institutes of Health grants no.
`CM-57732 and CA-05826.
`Address reprint requests to Robert J. Motzer, MD, Memorial Sloan-
`Kettering Cancer Center, 1275 York Ave, New York, NY 10021.
`娀 1999 by American Society of Clinical Oncology.
`0732-183X/99/1708-2530
`
`limit of normal), high ‘‘corrected’’ serum calcium (G 10
`mg/dL), and absence of prior nephrectomy. These were
`used as risk factors to categorize patients into three
`different groups. The median time to death in the 25% of
`patients with zero risk factors (favorable-risk) was 20
`months. Fifty-three percent of the patients had one or
`two risk factors (intermediate-risk), and the median sur-
`vival time in this group was 10 months. Patients with three
`or more risk factors (poor-risk), who comprised 22% of the
`patients, had a median survival time of 4 months.
`Conclusions: Five prognostic factors for predicting
`survival were identified and used to categorize patients
`with metastatic RCC into three risk groups, for which the
`median survival times were separated by 6 months or
`more. These risk categories can be used in clinical trial
`design and interpretation and in patient management.
`The low long-term survival rate emphasizes the priority
`of clinical investigation to identify more effective therapy.
`J Clin Oncol 17:2530-2540. 娀 1999 by American
`SocietyofClinicalOncology.
`
`Determining prognostic factors of survival for patients
`with advanced RCC would be valuable in directing therapy
`and interpreting results of clinical trials. Clinical trials in
`RCC frequently use biologic agents where responses may be
`delayed for 3 months or more after the institution of
`therapy,5 and prospective assessment of patient survival is
`necessary to determine appropriate eligibility. Response
`proportions to interferon-alpha, interleukin-2, or combina-
`tion programs vary considerably among phase II trials,6
`implying patient selection is an important factor in achieving
`a favorable treatment outcome. Clinical trials that include
`survival as an end point must account for prognostic factors
`to assure that treatment groups are comparable so that the
`proper interpretation of trial outcome can be ascertained.
`Also, an assessment of patient survival benefits both patient
`and physician in clinical management.
`Published analyses of prognostic factors performed in a
`multivariate analysis have been limited in both the number
`of series and the number of patients studied.7-12 To define
`pretreatment features predictive of survival, we performed a
`retrospective study on 670 patients with advanced RCC
`treated in successive clinical trials at the Memorial Sloan-
`Kettering Cancer Center (MSKCC). The results were exam-
`ined by multivariate analysis, and a model was developed to
`stratify patients according to risk.
`
`2530
`
`JournalofClinicalOncology,Vol 17, No 8 (August), 1999: pp 2530-2540
`
`West-Ward Exhibit 1035
`Motzer 1999
`Page 001
`
`

`

`RCC SURVIVAL AND PROGNOSTIC STRATIFICATION
`
`2531
`
`PATIENTS AND METHODS
`
`Table 2. Patient Characteristics
`
`Patients
`All patients were treated on MSKCC Institutional Review Board–
`approved clinical trials conducted between September 1975 and July
`1996. The patients were identified through registration on 24 consecu-
`tive MSKCC clinical
`trials;
`the specific eligibility and treatment
`programs have previously been described (Table 1).13-33 Eligibility
`details are described in individual reports but all included histologic
`confirmation of RCC, stage IV disease with presence of measurable
`lesions, adequate Karnofsky performance status, lack of severe comor-
`bid conditions, and adequate hematologic, renal, and hepatic function.
`Patients entered onto more than one clinical trial were evaluated for
`this study at the time of entry on their first MSKCC trial. Routine
`studies at the time of clinical trial entry included the following: detailed
`history and physical examination, complete blood count, prothrombin
`and partial thromboplastin times, creatinine, total bilirubin, alkaline
`phosphatase, AST lactate dehydrogenase, blood urea nitrogen, calcium,
`total protein, albumin, and imaging studies to assess measurable
`disease. The majority of patients had a computerized tomography scan
`of the abdomen and chest to assess extent of disease. Response to
`treatment, time to progression after systemic therapy, and survival and
`current status were recorded.
`
`Table 1. Composition of MSKCC Retrospective Study
`
`Protocol
`Reference
`
`Agent(s)
`
`No. of
`Patients
`
`14
`15
`16
`17
`18
`*
`19
`20
`13
`13
`22
`21
`13
`23
`24
`26
`27
`28
`29
`30
`31
`25
`32†
`33
`
`Vindesine
`Methyl GAG
`Flutamide
`4-epi-doxorubicin
`10-deaza-aminopterin
`AAFC*
`Bisantrene
`4-demethoxydaunorubicin
`Interferon-␣
`Interferon-␣
`Elliptinium
`N-methyl-formamide
`Interferon-␣ ⫹/⫺ vinblastine
`Trimetrexate
`Interleukin-2
`Didemnin
`Interleukin-2 plus interferon-␣
`Suramin
`Vinblastine
`Topotecan
`Liposomal doxorubicin
`Interferon-␣ plus 13-cis-retinoic acid
`Interferon-␣ ⫹/⫺ 13-cis-retinoic acid
`13-cis-retinoic acid
`
`18
`29
`23
`10
`12
`2
`18
`17
`36
`58
`9
`14
`51
`14
`68
`20
`34
`21
`23
`15
`11
`40
`109
`18
`
`Accrual Dates
`
`9/75-5/77
`11/79-5/80
`6/80-2/82
`7/80-9/81
`7/80-7/83
`2/81-5/81
`5/81-10/81
`2/82-10/82
`3/82-4/83
`7/83-5/84
`9/83-1/84
`4/84-4/85
`6/84-3/86
`9/86-9/87
`9/87-3/89
`2/88-9/89
`7/89-8/90
`8/90-6/91
`6/91-10/93
`12/91-6/92
`9/92-2/94
`1/93-4/94
`4/94-7/96
`6/94-2/95
`
`Abbreviations: AAFC, 2⬘2-anhydro-1-B-D-arabino-F-fluorocytosine; Methyl
`GAG, methylglyoxal bis(guanylhydrazone)dihydrochloride.
`*Trial unpublished.
`†Only patients treated at MSKCC included; patients treated by Eastern
`Cooperative Oncology Group were used as an external validation set and are
`described in a separate publication.
`
`Characteristic
`
`No. of patients
`Sex
`Male, %
`Female, %
`Age, years
`Median
`Range
`Range of diagnosis dates
`Karnofsky performance status, %
`ⱕ 60
`70
`80
`90
`Prior therapy, %
`Nephrectomy
`Radiation therapy
`Immunotherapy
`Chemo- or hormonal therapy
`No. of metastatic sites, %
`Renal primary or local recurrence
`only
`
`1
`2
`3
`ⱖ 4
`Sites of metastatic disease, %
`Lung
`Mediastinum
`Retroperitoneal lymph nodes
`Bone
`Liver
`Median baseline laboratory param-
`eters
`Albumin, normal 4.0-5.7 g/dL
`Alkaline phosphatase, normal
`0-115 U/L
`Calcium, normal 8.5-10.5 mg/dL
`Corrected calcium, normal ⬍ 10
`md/dL
`Hemoglobin, normal ⬎ 13 g/dL
`(M); ⬎ 11.5 g/dL (F)
`Lactate dehydrogenase, normal
`⬍ 200 U/L
`
`No. of
`Patients
`
`670
`
`450
`220
`
`%
`
`Range
`
`67
`33
`
`58
`18-82
`6/15/57-6/3/96
`
`7
`22
`32
`39
`
`65
`22
`8
`10
`
`3
`36
`38
`16
`7
`
`72
`20
`20
`26
`19
`
`46
`146
`211
`264
`
`434
`150
`56
`65
`
`19
`242
`253
`110
`46
`
`483
`135
`134
`176
`130
`
`4
`
`108
`9.7
`
`9.3
`
`12.3
`
`189
`
`2.3-5.3
`
`37-1248
`6.8-14.6
`
`6.2-14.2
`
`5.2-18
`
`59-5380
`
`Survival Analysis
`The end point of interest was survival time, defined as the time from
`treatment initiation to the death date or last follow-up date. Clinical
`features examined included number and sites of metastases (lung,
`mediastinum, bone,
`liver, and retroperitoneum), Karnofsky perfor-
`mance status, prior treatment (radiation, chemotherapy, and immuno-
`therapy), prior nephrectomy, the time interval from diagnosis to the start
`of treatment, and selected baseline biochemical features. The biochemi-
`cal features were based on a previous analysis and consisted of
`hemoglobin, serum albumin, alkaline phosphatase, lactate dehydroge-
`nase, and total calcium concentrations.34 To separate out the effects of
`
`West-Ward Exhibit 1035
`Motzer 1999
`Page 002
`
`

`

`2532
`
`MOTZER ET AL
`
`Fig 1. Overall survival (670 pa-
`tients, 57 alive). Vertical lines indi-
`cate last follow-up.
`
`protein binding and assess free calcium, an adjustment formula was
`used: ‘‘Corrected’’ calcium ⫽ total calcium ⫺ 0.707 [albumin-3.4].35
`The ‘‘corrected’’ calcium value was used in the survival analyses.
`Survival distributions were estimated using the Kaplan-Meier meth-
`od.36 The relationship between survival and each of the variables was
`analyzed using the log-rank test37 for categorical variables and a score
`test based on Cox proportional hazards regression model38 for continu-
`ous variables. Bivariate relationships among the variables were ex-
`plored to better understand how the variables interacted and how these
`interactions related to survival. There were few missing values for any
`of the variables (no more than 2%), and in all analyses, case deletion
`was used to handle the missing values. When necessary, a logarithmic
`transformation was used to reduce skewness.
`Two types of exploratory plots were used to display the functional
`relationship between continuous covariates (eg, lactate dehydrogenase
`and hemoglobin) and patient survival. The first was the running median
`survival time plot,39 which divided the covariate values into overlap-
`ping intervals, calculated the Kaplan-Meier–based median survival time
`for corresponding patients, and plotted these median survival times
`against the midpoint of the intervals. The second was the predictive
`failure time plot,40 which plotted the predicted median survival time
`based on a Cox regression model against each of the observed covariate
`values. These two plots are more descriptive of the relationship between
`a continuous covariate and survival time than a Kaplan-Meier plot.
`They allow the risk of death to vary according to the value of the
`covariate instead of assuming that all individuals in one group are at an
`equivalent risk of death.
`
`Multivariate Model
`Using a significant relationship with survival as criteria for including
`a variable in the stepwise modeling procedure, seven variables were
`retained and entered into a multivariate model. Because this retrospec-
`
`tive study included patients in clinical trials from 1975 through 1996,
`whose treatment included both cytotoxic therapy and immunotherapy, a
`stratified Cox proportional hazards model41 was used to account for
`differences in the year of treatment and the type of therapy. This model
`states that the hazard or risk of death at time t for a patient in strata j with
`variables x ⫽ (x1j, x2j, . . . , xpj) is
`
`␭j(t,x) ⫽ ␭oj (t) exp(␤1x1j ⫹ ␤2x2j ⫹ . . .⫹ ␤ pxpj)
`
`where ␭0j(t) is the baseline hazard function for strata j and ␤1, ␤2, . . . ,
`␤p are the regression coefficients. According to this model, when the
`regression coefficient is positive, then the risk of death increases with
`higher values of the variable. When the regression coefficient
`is
`negative, the risk of death decreases with higher values of the variable.
`Using a stepwise modeling algorithm with a .15 significance level for
`entering and removing explanatory variables, the independent risk
`factors were determined and the model was formed.
`Because it was desired to dichotomize the continuous variables
`chosen in the modeling for ease of clinical use, a minimum P value
`approach as well as the above exploratory plots were used to perform a
`cut point analysis.42 In the minimum P value approach, selected values
`of the prognostic factor are examined as candidates for the cut point.
`The value is chosen that best separates patient outcomes according to a
`maximum ␹2 statistic and minimum P value or a maximum relative risk.
`The P value is adjusted to account for the problem of multiple testing.
`The running median survival time plot and predicted failure time plot
`were used to restrict a region for the cut point search. Laboratory
`information about biologic cut points coupled with the information
`from the statistical techniques guided the decision about which cut point
`to use for each of the variables. It was verified that the relationship
`between survival and the prognostic factor remained significant when
`the variable was dichotomized.
`
`West-Ward Exhibit 1035
`Motzer 1999
`Page 003
`
`

`

`RCC SURVIVAL AND PROGNOSTIC STRATIFICATION
`
`2533
`
`Table 3. Univariate Survival Analysis of Number and Sites of Metastases and Prior Therapy
`
`%
`
`% Alive
`
`Median Survival
`
`CI
`
`␹2
`
`P
`
`Risk Ratio
`
`Clinical features of metastatic disease
`Lung metastases
`Yes
`No
`Mediastinum metastases
`Yes
`No
`Retroperitoneal metastases
`Yes
`No
`Bone metastases
`Yes
`No
`Hepatic metastases
`Yes
`No
`Total no. of metastatic sites
`0 or 1
`ⱖ 2
`Prior therapy
`Prior radiation
`Yes
`No
`Prior immunotherapy
`Yes
`No
`Prior chemotherapy
`Yes
`No
`Prior nephrectomy
`Yes
`No
`Interval from initial diagnosis to treatment, years
`⬍ 1
`ⱖ 1
`
`⬍ 2
`ⱖ 2
`
`72
`28
`
`20
`80
`
`20
`80
`
`26
`74
`
`19
`81
`
`39
`61
`
`22
`78
`
`8
`92
`
`10
`90
`
`65
`35
`
`63
`37
`
`85
`25
`
`8
`9
`
`6
`9
`
`7
`9
`
`7
`9
`
`5
`9
`
`10
`8
`
`3
`10
`
`4
`9
`
`8
`9
`
`11
`4
`
`6
`13
`
`6
`15
`
`9.9
`10.6
`
`11.6
`9.5
`
`8.5
`10.5
`
`9.0
`10.3
`
`7.4
`10.7
`
`10.7
`9.4
`
`8.2
`10.7
`
`8.2
`10.3
`
`5.8
`10.6
`
`11.3
`8.3
`
`8.5
`13.8
`
`8.8
`15.1
`
`8.8-11.0
`8.5-13.1
`
`9.4-14.5
`8.7-10.7
`
`7.7-10.4
`9.4-11.6
`
`7.8-11.4
`9.2-11.5
`
`5.5-8.7
`9.6-11.8
`
`9.2-13.0
`8.4-10.9
`
`7.6-9.5
`9.5-11.9
`
`6.2-12.1
`9.2-11.1
`
`4.2-8.0
`9.5-11.5
`
`9.5-12.7
`6.9-10.0
`
`7.6-9.4
`11.8-16.4
`
`7.9-9.8
`12.0-18.9
`
`1.79
`
`.181
`
`0.28
`
`.596
`
`1.50
`
`.221
`
`2.42
`
`.120
`
`9.00
`
`.003
`
`3.98
`
`.046
`
`7.59
`
`.0059
`
`0.30
`
`.5863
`
`13.49
`
`.0002
`
`30.40
`
`.0001
`
`33.74
`
`.0001
`
`30.28
`
`.0001
`
`1.1
`
`0.9
`
`1.1
`
`1.2
`
`1.4
`
`1.2
`
`1.3
`
`1.1
`
`1.6
`
`1.6
`
`1.6
`
`1.7
`
`The categorical counterparts of the risk factors determined in the
`model were used to assign each patient to one of three risk groups: those
`with zero risk factors (favorable-risk), those with one or two (intermedi-
`ate-risk), and those with three or more (poor-risk). Survival curves for
`each of these groups were estimated, and the groups were compared
`using the log-rank test.
`
`Validation of Model by Bootstrap Technique
`The predictive performance of the model was internally validated
`through a two-step nonparametric bootstrapping process.43 In the
`bootstrap procedure, the original set of data of size N becomes a parent
`population from which samples of size N are randomly drawn with
`replacement.
`In the first step of
`internal validation,
`the boot-
`strapping technique was used for variable selection. Two hundred
`bootstrap samples were created, and a stepwise procedure was applied
`to each sample using the same significance level for entering and
`removing a variable as in the original model. From this analysis,
`
`the percentage of samples for which each variable was included in
`the model from the 200 samples was calculated. Percent inclusion was
`used to determine the prognostic importance of a variable because it
`was expected that a prognostically important variable would be
`included in the model for a majority of the bootstrap samples. A model
`was formulated that included all variables whose percent inclusion was
`greater than or equal to 65%.44 The models obtained from the step-
`wise modeling algorithm and the bootstrapping technique were com-
`pared.
`In the second internal validation step, the bootstrap was used for
`parameter estimation. Three hundred bootstrap samples were created,
`and, for each of the samples, the model with the five final variables was
`refit and the regression parameters and risk ratios were estimated. The
`sample mean and SD of the 300 risk ratios for each parameter were
`computed and used to formulate confidence intervals about the risk
`ratio. These estimates were compared with those quantities obtained in
`the final Cox model.
`
`West-Ward Exhibit 1035
`Motzer 1999
`Page 004
`
`

`

`2534
`
`MOTZER ET AL
`
`Table 4. Univariate Survival Analysis of Performance Status and Biochemical Parameters
`
`Continuous Form
`
`Parameter Estimate
`
`Karnofsky performance status
`Albumin
`Alkaline phosphatase
`Hemoglobin
`Lactate dehydrogenase
`Calcium
`Corrected calcium
`
`⫺0.0458
`⫺0.798
`0.002
`⫺0.253
`0.001
`0.092
`0.373
`
`P
`
`.0001
`.0001
`.0001
`.0001
`.0001
`.1274
`.0001
`
`Cut Point Used
`
`⬍ 80
`4 g/dL
`88/115 U/L*
`13 g/dL (M)/11.5 g/dL (F)
`300 U/L†
`9 or 11 mg/dL‡
`10 mg/dL
`
`Categorical Form
`
`␹2
`
`73.62
`82.05
`25.42
`88.13
`105.14
`28.69
`37.59
`
`Risk Ratio
`
`95% CI
`
`2.15
`2.12
`1.51
`2.19
`3.32
`1.77
`1.98
`
`1.80-2.55
`1.80-2.50
`1.29-1.78
`1.86-1.78
`2.64-4.18
`1.44-2.18
`1.59-2.46
`
`*Eighty-eight units per liter used for patients ⱕ 55 years old at start of treatment and 115 U/L for patients ⬎ 55 years old.
`†LDH categorized as 1.5 times upper limit of normal.
`‡High-risk group defined as ⬍ 9 or⬎ 11 mg/dL.
`
`RESULTS
`Patient Characteristics and Treatment
`The median age of the patient group was 58 years; 67%
`were male (Table 2). Sixty-five percent had undergone a
`prior nephrectomy, 61% had two or more sites of metastases,
`22% had received prior radiation therapy, and 18% had
`received prior immunotherapy or cytotoxic chemotherapy.
`Thirty-seven percent of patients had an interval from
`diagnosis to treatment of 1 year or more. Six hundred eight
`patients (91%) were treated at MSKCC, whereas 62 (9%)
`were treated at an outside hospital on an MSKCC trial. Treatment
`consisted of immunotherapy in 396 patients (59%) and chemo-
`therapy (or hormonal therapy) in 274 patients (41%) (Table 1).
`With regard to immunotherapy, 294 patients were treated with
`interferon alpha, 68 patients with interleukin-2a, and 34 patients
`with a combination program. The overall response rate for
`the 670 patients was 12.5%, which included 10 complete
`responses and 41 partial responses.
`
`Survival Distribution
`The median overall survival time was 10 months (95%
`confidence interval [CI], 9 to 11 months) (Fig 1). Fifty-seven
`(8%) of the 670 patients remained alive and the median
`follow-up time for the survivors was 33 months (range, 0.9
`to 187 months). The percentage of patients surviving at 1
`year was 42%; the 2- and 3-year survival percentages were
`20% and 11%, respectively.
`
`Univariate Survival Analysis
`Factors considered in the univariate analyses included
`number and site of metastases, prior therapy, Karnofsky
`performance status, and baseline biochemical parameters
`(Tables 3 and 4). Factors associated with an adverse
`prognosis included presence of hepatic metastasis, two or
`more sites of metastases, a Karnofsky performance status
`less than 80, prior radiation or chemotherapy, lack of prior
`
`nephrectomy, and a time interval from disease diagnosis to
`treatment of less than 1 year. The median survival time according
`to Karnofsky performance status was 2.7 months for 60%,
`6.1 months for 70%, 10.6 months for 80%, and 14.4 months
`for 90% (P ⬍ .0001).
`The first two columns of Table 4 list parameter estimates
`and P values for testing the association of each biochemical
`parameter (in its continuous form) with survival. The
`negative regression coefficients on Karnofsky performance
`status, serum albumin, and hemoglobin concentrations indi-
`cate that, as the values of these three covariates increased,
`the risk of death decreased. The positive regression coeffi-
`cients on the other variables indicate that the risk of death
`increased as the value of the covariate increased. The
`biochemical parameters found to be significant for an
`adverse prognosis included low serum albumin, elevated
`serum alkaline phosphatase, low hemoglobin, an elevated
`serum lactate dehydrogenase level, and a high corrected
`serum calcium level. For lactate dehydrogenase, a logarith-
`mic transformation was used to reduce skewness.
`The effect on survival of the treatment year and program
`was evaluated (Table 5). Patients were classified according
`to treatment with immunotherapy, ie, interferon alpha and/or
`interleukin-2a, versus chemotherapy (cytotoxics or hor-
`monal therapy) and according to when they received treat-
`ment (1975 to 1980, 1981 to 1990, 1991 to 1996). Survival
`
`Table 5. Effect of Agent and Year of Treatment
`
`No. of
`Patients
`
`No. of
`Patients Alive
`
`Median Survival
`(months)
`
`CI
`(months)
`
`Agent
`IFN␣/IL-2
`Chemotherapy
`Year of treatment
`1975-1980
`1981-1990
`1991-1996
`
`396
`274
`
`66
`370
`234
`
`48
`9
`
`1
`20
`36
`
`12.9
`6.3
`
`4.2
`9.4
`13.2
`
`11.5-14.6
`5.1-7.6
`
`3.3-5.7
`8.1-10.7
`11.3-15.2
`
`Abbreviations: IFN␣, interferon alfa; IL-2, interleukin-2.
`
`West-Ward Exhibit 1035
`Motzer 1999
`Page 005
`
`

`

`RCC SURVIVAL AND PROGNOSTIC STRATIFICATION
`
`2535
`
`(A) Running median sur-
`Fig 2.
`vival
`time plot and (B) predicted
`failure time plot for serum lactate
`dehydrogenase concentration.
`
`for patients treated with immunotherapy
`was greater
`(P ⬍ .0001) and for patients treated in more recent years
`(P ⬍ .0001). To account for these effects and to develop a
`prognostic model based on pretreatment features, type and
`year of treatment were included as strata in the multivariate
`survival analysis.
`
`Multivariate Survival Analysis
`The seven variables included in the multivariate analysis
`were hemoglobin, serum lactate dehydrogenase, corrected
`calcium, prior nephrectomy, Karnofsky performance status,
`hepatic metastases, and the interval from diagnosis to
`treatment. Using a .15 significance level for entering and
`
`West-Ward Exhibit 1035
`Motzer 1999
`Page 006
`
`

`

`2536
`
`MOTZER ET AL
`
`Fig 3. Minimum P value ap-
`proach for lactate dehydrogenase.
`
`it was determined that
`removing explanatory variables,
`hemoglobin, lactate dehydrogenase, corrected calcium, ne-
`phrectomy, and Karnofsky performance status were indepen-
`dent risk factors predicting survival.
`
`Cut Point Analysis
`For hemoglobin and corrected calcium, the cut point
`suggested by the minimum P value approach matched the
`upper/lower limit of normal. For these variables, the running
`median survival time plot and predicted failure time plot
`were continuously increasing or decreasing (monotonic) and
`did not indicate any obvious cut point. For lactate dehydro-
`
`Table 6. Results of Multivariate Analysis
`
`Parameter
`Estimate
`
`SE
`
`␹2
`
`P
`
`Risk
`Ratio
`
`95% CI
`
`Lactate dehydroge-
`nase
`Hemoglobin
`Corrected calcium
`Karnofsky perfor-
`mance status
`Prior nephrectomy
`
`No. of
`Risk
`Factors
`
`0
`1 or 2
`3, 4, or 5
`
`% of
`Patients
`
`25
`53
`22
`
`0.9019
`0.5439
`0.5268
`
`0.1230 53.74 .0001 2.46 1.94-3.14
`0.0897 36.75 .0001 1.72 1.45-2.05
`0.1147 21.11 .0001 1.69 1.35-2.12
`
`0.4050
`0.2992
`
`0.0967 17.56 .0001 1.50 1.24-1.81
`0.0908 10.87 .001
`1.35 1.13-1.61
`
`% of
`Patients
`Alive
`
`18
`7
`0.7
`
`Median
`Survival
`(months)
`
`19.9
`10.3
`3.9
`
`95% CI
`
`17.1-27.9
`8.9-11.4
`3.4-5.0
`
`1-Year
`Survival
`(%)
`
`3-Year
`Survival
`(%)
`
`71
`42
`12
`
`31
`7
`0
`
`genase, these two plots suggested that its relationship to
`survival was essentially monotonically decreasing, but that a
`cut point search could be restricted to values between 100
`and 500 U/L (Figs 2A and 2B). The graph from the minimum
`P value approach showed a peak at 319 U/L with a
`maximum ␹2 value of 153.8 and a minimum P value of
`⬍ .0001 (Fig 3). The adjusted P value remained significant;
`therefore, serum lactate dehydrogenase was categorized at
`the value of 300 U/L (1.5 times upper limit of the normal
`value).
`The last three columns of Table 4 list the cut points chosen
`for each of the continuous variables along with the results of
`the univariate survival analysis for the dichotomized ver-
`sions of the variables. The magnitudes of the ␹2 and the
`corresponding risk ratio illustrate the magnitude of the
`covariate’s effect on survival. For example, the risk ratio for
`lactate dehydrogenase was 3.3. This indicates that a patient
`with a lactate dehydrogenase value greater than 300 U/L was
`3.3 times more likely to die than a patient with a value less
`than 300 U/L. All dichotomized biochemical parameters
`were significant at the .0001 level.
`
`Risk Groups
`Low Karnofsky performance status (⬍ 80%), high lactate
`dehydrogenase (⬎ 1.5 times upper limit of normal), low
`serum hemoglobin (⬍ lower limit of normal), high corrected
`serum calcium (⬎ 10 mg/dL), and absence of nephrectomy
`were risk factors. Using the dichotomized versions of these
`
`West-Ward Exhibit 1035
`Motzer 1999
`Page 007
`
`

`

`RCC SURVIVAL AND PROGNOSTIC STRATIFICATION
`
`2537
`
`Fig 4. Survival stratified accord-
`ing to risk group (n ⫽ 656; 14 pa-
`tients missing one or more of the five
`risk factors were excluded). Vertical
`lines indicate last follow-up.
`
`variables, a Cox model was fit (Table 6). Each patient was
`then assigned to one of three risk groups: those with no risk
`factors (favorable-risk), those with one or two risk factors
`(intermediate-risk), and those with three or more risk factors
`(poor-risk).
`The median time to death in 25% of patients deemed
`favorable-risk was 20 months, and the 1-, 2-, and 3-year
`survival rates were 71%, 45%, and 31%, respectively.
`Fifty-three percent of the patients were in the intermediate-
`risk group. The median survival time for this group was 10
`months, with 1-, 2-, and 3-year survival rates of 42%, 17%,
`and 7%, respectively. In contrast, the poor-risk group, which
`comprised 22% of the patients, had a median survival time
`of 4 months, with 1-, 2-, and 3-year survival rates of 12%,
`3%, and 0%. Only one poor-risk patient remained alive at
`the time of last follow-up. There was a significant difference
`in the survival profiles of the three risk groups (P ⬍ .0001)
`(Fig 4).
`
`Table 7. Percent Inclusion of Each Variable in Variable Selection Step
`of Bootstrap Validation
`
`HgB
`
`100
`
`LDH
`
`100
`
`Corrected
`Calcium
`
`100
`
`Nephrectomy
`
`67
`
`KPS
`
`89
`
`Hepatic
`Metastases
`
`40.5
`
`Time from
`Diagnosis
`to Treatment
`
`55
`
`Abbreviations: HgB, hemoglobin; LDH, lactate dehydrogenase; KPS, Karno-
`fsky performance status.
`
`Type and year of treatment were included as strata in the
`multivariate survival analysis. When patients were stratified
`according to risk, the median survival time was greater in
`each of the three risk groups for patients treated in more
`recent years versus those treated earlier. The median survival
`time was also greater for patients treated with immuno-
`therapy versus those treated with chemotherapy. For patients
`treated with immunotherapy (interferon and/or interleukin-
`2), the median survival times for favorable-risk, intermediate-
`risk, and poor-risk patients were 26 months, 12 months, and
`6 months, respectively.
`
`Bootstrap Validation
`In the bootstrap procedure, the original set of data of size
`N becomes a parent population from which samples of size
`N are randomly drawn with replacement. The bootstrap
`samples are then treated as if they come from the true
`distribution of advanced RCC patients, and inferences about
`the risk ratios for each covariate are based on the empirical
`
`Table 8. Results of Parameter Estimation Step of Bootstrap Validation
`
`Risk Ratio
`
`SE
`
`Lactate dehydrogenase
`Hemoglobin
`Corrected calcium
`Karnofsky performance status
`Prior nephrectomy
`
`2.52
`1.76
`1.70
`1.53
`1.35
`
`0.3879
`0.1690
`0.1989
`0.1656
`0.1375
`
`95% CI
`
`1.76-3.28
`1.43-2.09
`1.32-2.09
`1.20-1.85
`1.08-1.62
`
`West-Ward Exhibit 1035
`Motzer 1999
`Page 008
`
`

`

`2538
`
`distribution of the risk ratios. For the first step of validation,
`five of the seven variables had a percent inclusion greater
`than 65% (Table 7). These were hemoglobin, lactate dehydro-
`genase, corrected calcium, prior nephrectomy, and Karnof-
`sky performance status. The results of this model selection
`technique confirmed the variables chosen in the original
`modeling procedure.
`In the second step of validation, a risk ratio with a 95%
`confidence interval was estimated for each covariate in the
`final model. Risk ratios (Table 8) were similar to those
`obtained in the original multivariate model (Table 6). For
`example, the risk ratio for Karnofsky performance status
`from the bootstrap procedure was 1.53 (1.20 to 1.85),
`whereas in the original model it was 1.50 (1.24 to 1.81). The
`results of these two steps provide evidence of the predictive
`ability of the final model.
`
`DISCUSSION
`This study resulted in a model based on five pretreatment
`clinical features that predicted survival for patients with
`advanced RCC. Risk factors associated with a shorter
`survival period were low Karnofsky performance status
`(⬍ 80%), high lactate dehydrogenase (⬎ 1.5 times upper
`limit of normal), low serum hemoglobin (⬍ lower limit of
`normal), high corrected serum calcium (⬎ 10 mg/dL), and
`absence of prior nephrectomy. These risk factors were used
`to stratify patients into three different groups. Three-year survival
`percentages for the favorable-risk (no risk factors), intermediate-
`risk (one or two risk factors), and poor-risk (three or more risk
`factors) groups were 31%, 7%, and 0%, respectively.
`Validation was performed by the bootstrap method.43
`Repeated sampling of the original data with replacement
`allowed independent samples of RCC patients to be gener-
`
`MOTZER ET AL
`
`ated from which the predictive accuracy of the model was
`assessed. In addition, we have applied the prognostic model
`to an external data set
`taken from a trial by Eastern
`Cooperative Oncology Group.32 The external group was
`composed of 175 patients treated on a randomized trial of
`interferon-alpha with or without 13-cis-retinoic acid. In this
`group, the median survival times of favorable-, intermediate-,
`and poor-risk patients were 29, 14, and 4 months, respectively.
`There are few reports of prognostic factors studied by
`multivariate analysis in patients with metastatic RCC.7-11,45-50
`The prognostic factors vary among the studies but consistently
`include performance status, nephrectomy, and a measure of
`extent of disease. A summary of multivariate analyses resulting
`in criteria for risk stratification is listed in Table 9.7-12
`A study by Elson et al7 contained a number of patients
`similar to the retrospective study in this article. The popula-
`tion was composed of 610 patients treated with chemother-
`apy on phase II trials between 1975 and 1984.7 The lack of
`patients treated with immunotherapy in this analysis7 is seen
`by some as a present-day limitation.9 The model stratified
`patients into five categories with a difference in median
`survival time of as little as 1.3 months between groups and
`included subjective criteria of ‘‘weight loss in previous 6
`months’’ as a component. Today the patient population is
`different from that of Elson et al’s7 study, reflecting improve-
`ment in imaging techniques and selection factors used to
`choose patients with RCC exclusively for phase II trials of
`cytotoxic agents. The median survival time for all patients
`treated in that series was 5.6 months,7 compared with 10
`months in the present series. Also, the median survival time
`in the most favorable risk group was 12.8 months, which
`comprised 18% of the entire group,7 compared with a
`
`Table 9. Multivariate Analyses of Prognostic Factors for Survival in Patients With Advanced Renal Cell Carcinoma
`
`Author(s)
`
`Year
`
`No. of Patients
`
`Treatment
`
`Elson et al7
`
`1988
`
`610
`
`Chemotherapy
`
`DeForges8
`
`Palmer et al9
`
`Jones et al10
`
`Fossa et al12
`Lopez-Hannineh et al11
`
`1988
`
`1992
`
`1993
`
`1994
`1996
`
`134
`
`327
`
`387
`
`295
`215
`
`Chemotherapy and interferon-␣
`
`Interleukin-2
`
`Interleukin-2
`
`Interferon ⫹ chemotherapy
`Interleukin-2 with/without interferon-␣, 5-FU
`
`Present series
`
`1998
`
`670
`
`Interferon-␣, interleukin-2, chemotherapy
`
`Abbreviation: FU, fluorourac