Article Tools
ORIGINAL REPORTS
Phase I and Clinical Pharmacology
Article Tools
Patient Selection for Oncology Phase I Trials: A Multi-Institutional Study of Prognostic Factors
Abstract
The appropriate selection of patients for early clinical trials presents a major challenge. Previous analyses focusing on this problem were limited by small size and by interpractice heterogeneity. This study aims to define prognostic factors to guide risk-benefit assessments by using a large patient database from multiple phase I trials.
Data were collected from 2,182 eligible patients treated in phase I trials between 2005 and 2007 in 14 European institutions. We derived and validated independent prognostic factors for 90-day mortality by using multivariate logistic regression analysis.
The 90-day mortality was 16.5% with a drug-related death rate of 0.4%. Trial discontinuation within 3 weeks occurred in 14% of patients primarily because of disease progression. Eight different prognostic variables for 90-day mortality were validated: performance status (PS), albumin, lactate dehydrogenase, alkaline phosphatase, number of metastatic sites, clinical tumor growth rate, lymphocytes, and WBC. Two different models of prognostic scores for 90-day mortality were generated by using these factors, including or excluding PS; both achieved specificities of more than 85% and sensitivities of approximately 50% when using a score cutoff of 5 or higher. These models were not superior to the previously published Royal Marsden Hospital score in their ability to predict 90-day mortality.
Patient selection using any of these prognostic scores will reduce non–drug-related 90-day mortality among patients enrolled in phase I trials by 50%. However, this can be achieved only by an overall reduction in recruitment to phase I studies of 20%, more than half of whom would in fact have survived beyond 90 days.
The appropriate selection of patients with advanced cancer for phase I clinical trials has always represented a unique challenge because of the difficult underlying risk-benefit assessments.1 Therapeutic utility is not a conventional primary end point of these dose- and toxicity-finding studies,2 and potential patients are vulnerable because of the presence of progressive disease and the lack of standard treatment options.3–5 Several institutional studies3,6–16 have previously attempted to address these selection issues but were severely limited because of small patient numbers and larger interpractice heterogeneity. These include the previously validated3,6 (n = 290) Royal Marsden Hospital (RMH) score, which comprises three variables: serum albumin, number of metastatic sites, and lactate dehydrogenase (LDH).
To overcome previous limitations, we conducted a large multi-institutional, international study based on individual patient case records. The aims of this study included an analysis of treatment antitumor activity and the identification of objective clinicopathologic factors that determine prognosis. Our expectation was that these factors could then to provide guidance on patient selection for phase I trials to minimize early trial attrition. Patients often discontinue clinical trial participation during the mandatory evaluation period (usually the first 3 to 4 weeks on study) because of non–drug-related events, and they typically derive minimal or no benefit from participation in such studies. These patients would need to be replaced for a formal evaluation of safety and toxicity. Hence, the cost of conducting the clinical trial is inadvertently increased with delays incurred in dose-escalation decision making.1,7,17
A wide variety of potential parameters were assessed in this study. Although an expected survival beyond 90 days is a common eligibility criterion in oncology phase I trials, the accuracy of current predictions is poor.7–9,13 Thus, we selected 90-day mortality as the primary end point for this prognostic modeling study.
This is a retrospective multicenter study comprising all patients treated consecutively in phase I trials (including phases IA and IB) from January 2005 to December 2007 within the European Drug Development Network (EDDN), an international consortium involving 14 oncology drug development units. The primary aim was to generate and validate a prognostic model for 90-day mortality by using a pan-European database. Secondary aims included an estimation of early (within the first 21 days) trial discontinuation, a description of risk factors associated with patient-related early trial discontinuation, and an analysis of current overall response and nonprogression rates. All patients who met the general eligibility criteria were included for descriptive analysis. These criteria included histologic confirmation of cancer, age 18 years or older, and receipt of at least one dose of study medication. Patients also met the relevant eligibility criteria for each study. Another criterion necessary for inclusion in the 90-day mortality prognostic modeling was the availability of at least 21 of the 23 parameters included in the calculation; patients censored for 90-day mortality were ineligible. For overall survival (OS) analysis, only patients taking part in their first phase I trial were considered. All patients included in this analysis gave their informed consent to take part in phase I trials and were approved by local institutional review boards, which also granted their approval for this analysis.
Standard clinicopathologic parameters were collected at study entry. In addition, we calculated the time per treatment index (TPTi), a log ratio of the time interval between diagnosis of advanced/metastatic cancer and phase I trial entry over the number of lines of systemic treatment, by using [log (Time + 1/treatments + 1)] (Appendix, online only).18 Other data collected included time to trial discontinuation and causes and tumor responses classified according to Response Evaluation Criteria in Solid Tumors (RECIST).19
On the basis of the Fisher-Irwin test, we estimated that 848 patients were required to validate a prognostic variable (dichotomized by the median), which associates with a minimum increase of 1.5-fold in the risk of 90-day mortality (if considering a 90-day mortality in a phase I population of 15%, an alpha error of 5%, and a beta error of 20%). In addition, two different models were defined a priori: model A consisted of 23 baseline variables in all eligible patients; model B excluded patients with Eastern Cooperative Oncology Group performance status (ECOG PS) 2 and did not include PS as a variable in the model (Fig 1A).
Fig 1. (A) Flowchart and (B) tumor type distribution for study patients (pts). ECOG PS, Eastern Cooperative Oncology Group performance score.
A posteriori we found that 1,845 and 1,780 patients were eligible for 90-day survival analysis in models A and B, respectively. For each model, patients were randomly divided into two groups: one for derivation of and the other for validation of the prognostic model. For the derivation of the prognostic model, all variables with a significance level below 0.10 in the univariate analysis were considered in a multivariate analysis that used logistic regression, and if their P value in the derivation set was less than .05, they were further tested in the validation set.
To simplify the application of prognostic factors validated as continuous variables, we transformed them into binomial variables. All cutoffs for the conversion of these continuous variables into binary categories were chosen on the basis of considerations of the standard reference values (normal range v low or increased) or according to the median values. The only exception to this was the TPTi; despite a median of 20 weeks, the cutoff of 24 weeks (approximately 6 months) was chosen because it is a commonly considered relevant interval in clinical practice and trials. (An extended description of the statistical methods is provided in the Appendix).
We collected data from 2,232 phase I trial patients treated in 14 participating oncology drug development units. Fifty patients were found to be ineligible; analyses were carried out by using data from the remaining 2,182 patients (Fig 1A). Baseline clinicopathologic characteristics of patients included in this study are listed in Table 1. Briefly, the ratio of males to females was 1.3, and the median age was 55 years (range, 19 to 86 years). Patients in this study received a median of two (range, zero to 11) prior lines of systemic treatments for advanced disease. The main tumor systems were GI (31%), genitourinary (12%), and gynecologic (12%), as shown in Figure 1B. Data collection cutoff for this analysis was September 2009 when the median follow-up was 50 weeks.
|
| Characteristic | No. | % |
|---|---|---|
| Sex | ||
| Male | 1,227 | 56.20 |
| Female | 955 | 43.80 |
| Age, years | ||
| Median | 55 | |
| Range | 19-86 | |
| ECOG performance score | ||
| 0 | 884 | 40.50 |
| 1 | 1,227 | 56.20 |
| 2 | 71 | 3.30 |
| Disease at study entry | ||
| Progressive | 2,128 | 97.50 |
| Stable | 45 | 2.10 |
| Unknown/not specified | 9 | 0.40 |
| Time from advanced disease to phase I, months | ||
| Median | 15 | |
| Range | 0-71 | |
| No. of metastatic sites | ||
| Median | 2 | |
| Range | 0-7 | |
| Site of metastases | ||
| Liver | 1,052 | 48.20 |
| Lymph nodes | 1,044 | 47.80 |
| Lung | 989 | 45.80 |
| Bone | 392 | 18.00 |
| Serum LDH × ULN | ||
| Median | 0.98 | |
| Range | 0.2-23.8 | |
| Serum albumin, g/dL | ||
| Median | 3.8 | |
| Range | 1.0-5.8 | |
| ALP × ULN | ||
| Median | 0.94 | |
| Range | 0.2-13.4 | |
| AST × ULN | ||
| Median | 0.71 | |
| Range | 0.1-12.3 | |
| ALT × ULN | ||
| Median | 0.58 | |
| Range | 0.1-5.2 | |
| Total bilirubin × ULN | ||
| Median | 0.53 | |
| Range | 0.4-4.9 | |
| WBC/μL | ||
| Median | 7,200 | |
| Range | 1,800-46,300 | |
| Lymphocyte count, % | ||
| Median | 18.10 | |
| Range | 1.6-80 | |
| Platelets/μL | ||
| Median | 274,000 | |
| Range | 83,000-917,000 | |
| Hemoglobin, g/dL | ||
| Median | 12.4 | |
| Range | 7-18.4 | |
| No. of previous systemic treatment lines | ||
| Median | 2 | |
| Range | 0-11 | |
| TPTi, weeks per treatment line | ||
| Median | 20 | |
| Range | 1-652 | |
| Drugs and trial type | ||
| Single agent | 1,370 | 62.80 |
| First time in humans, novel cytotoxics | 389 | 17.80 |
| First time in humans, molecularly targeted drugs | 814 | 37.30 |
| Others (ie, pharmacodynamic, pharmacokinetic) | 167 | 7.70 |
| Combinations | 812 | 37.20 |
| Cytotoxics | 95 | 4.40 |
| Molecularly targeted drugs | 170 | 7.80 |
| Cytotoxics and molecularly targeted drugs | 547 | 25.00 |
Abbreviations: ALP, alkaline phosphatase; ECOG, Eastern Cooperative Oncology Group; LDH, lactate dehydrogenase; TPTi, time per treatment index; ULN, upper limit of normal.
Sixty-three percent of patients were treated in trials involving single agents (88% of which were first-in-human) and the remainder in trials involving combinations of novel or established therapies. The majority of patients (78%) took part in a phase I trial that included a molecular targeting agent (Appendix Table A1, online only). A total of 50% of the patients took part in a trial that included a novel cytotoxic (11%) or approved conventional cytotoxic (39%) chemotherapy, half of which involved a combination with a targeted agent.
Data for evaluation of response according to RECIST19 were available for 2,027 patients. Overall objective radiologic response rate was 10%, and disease stabilization rates at 3 and 6 months were 26% and 10%, respectively. The proportion of patients who had progressive disease at first imaging evaluation was 49.5%, and it differed significantly (P < .001) between patients treated in trials that contained conventional cytotoxics (31%) and patients in trials that did not (61%). The median progression-free survival was 10.9 weeks (95% CI, 10.2 to 11.5 weeks).
In this study, 2,158 patients were evaluated for early trial discontinuation. To summarize, 14% of patients (n = 313) discontinued trial participation within the first 21 days of study treatment day 1. Of the patients who stopped early, 69% (n = 215) stopped because of events not attributable to treatment. Instead, the discontinuation causes were: radiologic and/or clinical disease progression (57%), non–drug-related adverse events (7%), or other medical causes (5%). Twenty-six percent of patients (n = 81) stopped early for treatment-related reasons; included in the 26% were six deaths related to toxicity. The remaining 5% of patients stopped early at their own request. The prognostic factors related to 21-day patient-related discontinuation by relative risk included ECOG PS 2 versus 0 (odds ratio [OR], 4.8), ECOG PS 1 versus 0 (OR, 2.4), increased LDH (OR, 1.3), increased alkaline phosphatase (ALP; OR, 1.2), and more than two metastatic sites (OR, 1.2). An a posteriori analysis revealed a significant difference (P < .001) in this early attrition rate between patients enrolled in phase I trials that used conventional cytotoxics (11.0%) and in trials that did not (17%).
Data from 1,929 patients were included in OS analysis and from all patients for early mortality rates. Median OS was 38.6 weeks (95% CI, 36.0 to 41.1 weeks). Early mortality rates within two time periods were determined: 30 days and 90 days from study day 1. The 90-day mortality rate was 16.5% (95% CI, 15.0% to 17.5%), with disease-related deaths in 16% (95% CI, 14.1% to 17.1%) of patients, whereas the 30-day mortality rate was 3% (95% CI, 1.9% to 3.2%), with disease-related deaths in 2% (95% CI, 1.6% to 2.8%) of the patients. Toxicity accounted for 0.4% of all deaths (95% CI, 0.1% to 0.7%), with the majority (0.3%) occurring during the first 30 days. The 90-day mortality rate differed significantly (P < .001) between patients enrolled within trials that used conventional cytotoxics (12%) and trials that did not (19%). No significant differences were found for 30-day death rates and death rates related to toxicity.
Two models were used to derive and validate prognostic models of 90-day survival. In model A, six of 23 variables were validated (P < .05): ECOG PS, albumin, LDH, TPTi, ALP, and number of metastatic sites. Lymphocyte count (%), although it was initially significant in the derivation cohort, was not later validated (Table 2). When continuous variables were transformed into binomial variables, the greater risk (OR, ≥ 4) for 90-day mortality was found in patients with ECOG PS 2 (Table 2). In model B (which did not include ECOG PS), and after excluding all ECOG PS 2 patients, seven prognostic variables were validated: albumin, LDH, TPTi, ALP, number of metastatic sites, lymphocytes, and WBC (all with OR, > 1.5 when transformed into binomial variables). In addition, two other factors were derived but not validated: a diagnosis of upper GI cancer and the presence of liver metastasis (Table 2).
|
| Significant Variables in Univariate Analysis Model A (n = 1,845) | Multivariate Analysis | |||||
|---|---|---|---|---|---|---|
| Derivation Set | Validation Set | |||||
| OR | 95% CI (n = 925) | P | OR | 95% CI (n = 920) | P | |
| ECOG PS | ||||||
| 1 v 0 | 1.9 | 1.10 to 3.28 | .009 | 2.27 | 1.31 to 3.94 | .002 |
| 2 v 0 | 3.96 | 1.54 to 10.2 | .009 | 4.86 | 1.88 to 12.5 | .002 |
| Serum albumin (g/L; as continuous) | 0.93 | 0.90 to 0.97 | < .001 | 0.94 | 0.91 to 0.98 | .001 |
| Low albumin (< 35 g/L) v normal | 3.19 | 2.09 to 4.88 | < .001 | 1.88 | 1.24 to 2.85 | .003 |
| LDH × ULN (as continuous) | 1.39 | 1.18 to 1.63 | < .001 | 1.17 | 1.05 to 1.30 | .005 |
| Increased LDH (> 1 × ULN) v normal | 1.86 | 1.20 to 2.89 | .005 | 2.76 | 1.83 to 4.16 | < .001 |
| No. of metastatic sites (as continuous) | 1.27 | 1.08 to 1.51 | .004 | 1.41 | 1.17 to 1.71 | < .001 |
| ≥ 3 v ≤ 2 | 1.78 | 1.15 to 2.76 | .010 | 2.28 | 1.49 to 3.48 | < .001 |
| ALP (as continuous) | 1.19 | 1.06 to 1.34 | .003 | 1.25 | 1.05 to 1.49 | .013 |
| Increased ALP (> 1 × ULN) v normal | 1.57 | 1.03 to 2.40 | .036 | 2.39 | 1.57 to 3.63 | < .001 |
| CTGR in weeks per treatment (as continuous) | 0.99 | 0.98 to 1.00 | .015 | 0.99 | 0.98 to 1.00 | .032 |
| Short (< 24 weeks) v long | 1.48 | 1.00 to 2.18 | .048 | 1.73 | 1.13 to 2.64 | .012 |
| Lymphocyte/white cell (%) | 0.97 | 0.94 to 0.99 | .011 | 0.98 | 0.95 to 1.01 | .067 |
| Melanoma v other | — | — | .087 | — | — | — |
| Liver metastasis v no | — | — | .094 | — | — | — |
| Time from metastasis/advanced disease to phase I trial entry1 | — | — | .112 | — | — | — |
| Platelet count | — | — | .114 | — | — | — |
| Lymphocyte count | — | — | .344 | — | — | — |
| Hemoglobin level | — | — | .397 | — | — | — |
| WBC | — | — | .545 | — | — | — |
| Model B (n = 1,780) | (n = 892) | (n = 888) | ||||
| Serum albumin (g/L; as continuous) | 0.92 | 0.88 to 0.95 | < .001 | 0.94 | 0.91 to 0.98 | .001 |
| Low albumin (< 35 g/L) v normal | 2.45 | 1.57 to 3.84 | < .001 | 2.60 | 1.66 to 4.06 | < .001 |
| LDH × ULN (as continuous) | 1.39 | 1.07 to 1.60 | < .001 | 1.17 | 1.05 to 1.30 | .005 |
| Increased LDH (> 1 × ULN) v normal | 2.28 | 1.45 to 3.59 | < .001 | 1.74 | 1.10 to 2.74 | .017 |
| No. of metastatic sites (as continuous) | 1.31 | 1.07 to 1.60 | .008 | 1.23 | 1.01 to 1.50 | .042 |
| ≥ 3 v ≤ 2 | 2.84 | 1.84 to 4.40 | < .001 | 1.90 | 1.39 to 2.60 | < .001 |
| ALP (as continuous) | 1.06 | 1.02 to 1.11 | .010 | 1.09 | 1.02 to 1.16 | .010 |
| Increased ALP (> 1 × ULN) v normal | 1.70 | 1.09 to 2.64 | .018 | 1.93 | 1.24 to 3.015 | .004 |
| Lymphocyte/white cell (%; as continuous) | 0.97 | 0.94 to 0.99 | .022 | 0.96 | 0.93 to 0.99 | .016 |
| Low lymphocyte count (≤ 18%) v normal | 1.86 | 1.15 to 3.03 | .012 | 2.04 | 1.26 to 3.31 | .004 |
| WBC/μL (as continuous) | 1.06 | 1.02 to 1.11 | .010 | 1.09 | 1.02 to 1.16 | .010 |
| WBC ≥ 10,500/μL v normal | 2.08 | 1.33 to 3.27 | .001 | 1.52 | 1.05 to 2.22 | .029 |
| CTGR in weeks per treatment (as continuous) | 0.99 | 0.98 to 1.00 | .004 | 0.98 | 0.97 to 0.99 | .009 |
| Short (< 24 weeks) v long | 1.48 | 1.13 to 1.93 | .005 | 1.78 | 1.13 to 2.81 | .013 |
| Upper GI cancer v other | 2.00 | 1.12 to 3.56 | .019 | 1.46 | 0.80 to 2.67 | .222 |
| Liver metastasis v no | 1.77 | 1.11 to 2.82 | .026 | 0.87 | 0.54 to 1.41 | .868 |
| Time from metastasis/advanced disease to phase I trial entry | — | — | .099 | — | — | — |
| Nodal metastasis v no | — | — | .116 | — | — | — |
| AST | — | — | .542 | — | — | — |
| Platelet count | — | — | .644 | — | — | — |
| Hemoglobin level | — | — | .902 | — | — | — |
| Lymphocyte count | — | — | .998 | — | — | — |
Abbreviations: ALP, alkaline phosphatase; CTGR, clinical tumor growth rate; ECOG, Eastern Cooperative Oncology Group; LDH, lactate dehydrogenase; OR, odds ratio; PS, performance score; ULN, upper limit of normal.
We also explored a posteriori whether these prognostic factors differed between patients enrolled in phase I trials that used conventional cytotoxics and those that did not. In a multivariate analysis of all patients treated within trials that used conventional cytotoxic drugs, the main prognostic factors for 90-day mortality were albumin, number of metastatic sites, LDH, and WBC. The significant and independent prognostic factors for patients treated with nonconventional cytotoxic drugs were ECOG PS, albumin, lymphocytes, number of metastatic sites, LDH, ALP, and TPTi.
To identify a simpler predictive model that contained a smaller number of variables than the predictor, based on the sum of the scores for models A and B (Table 3), we estimated the area under the curve (AUC) of the receiving operator characteristic curve of each possible predictor on the basis of the sum of the scores for each combination of variables considered. All combinations were then classified according to AUC (Appendix Table A2, online only). The best combinations using three factors were albumin, LDH, and ECOG PS (AUC, 73.8%) in model A, and albumin, LDH, and number of metastatic sites (AUC, 71.3%) in model B. This latter combination comprised the three prognostic factors in the RMH score3,6 (Table 3). The highest AUC values were found in the model containing all the factors in model A (AUC, 75.5%) and model B (AUC, 73.8%). To identify whether a combination was superior to the RMH score, the receiving operator characteristic curves of the best combinations by number of incorporated prognostic factors (three, four, five, six, or seven factors) were compared with the combination of albumin, LDH, and number of metastatic sites (RMH score), but differences in AUC were not statistically significant in any comparison (Appendix Table A2).
|
| Prognostic Score Models | |||
|---|---|---|---|
| RMH Score (any ECOG PS) | |||
| Low albumin (< 3.5 g/dL) | + 1 | Normal albumin (≥ 3.5 g/dL) | + 0 |
| High No. of metastatic sites (≥ 3) | + 1 | Low No. of metastatic sites (≤ 2) | + 0 |
| Increased LDH (> ULN) | + 1 | Normal LDH (≤ ULN) | + 0 |
| European Model A | |||
| ECOG PS 2 | + 3 | ||
| ECOG PS 1 | + 1 | ECOG PS 0 | + 0 |
| Low albumin (< 3.5 g/dL) | + 1 | Normal albumin (≥ 3.5 g/dL) | + 0 |
| Increased LDH (> ULN) | + 1 | Normal LDH (≤ ULN) | + 0 |
| High No. of metastatic sites (≥ 3) | + 1 | Low No. of metastatic sites (≤ 2) | + 0 |
| Low TPTi (< 24 weeks per treatment) | + 1 | Long TPTi (≥ 24 weeks per treatment) | + 0 |
| Increased ALP (> ULN) | + 1 | Normal ALP (≤ ULN) | + 0 |
| European Model B (ECOG PS 2 excluded) | |||
| Low albumin (< 3.5 g/dL) | + 1 | Normal albumin (≥ 3.5 g/dL) | + 0 |
| Increased LDH (> ULN) | + 1 | Normal LDH (≤ ULN) | + 0 |
| High No. of metastatic sites (≥ 3) | + 1 | Low No. of metastatic sites (≤ 2) | + 0 |
| Low TPTi (< 24 weeks per treatment) | + 1 | Long TPTi (≥ 24 weeks per treatment) | + 0 |
| Increased ALP (> ULN) | + 1 | Normal ALP (≤ ULN) | + 0 |
| Low lymphocyte count (< 18%) | + 1 | Normal lymphocyte count (≥ 18%) | + 0 |
| Increased WBC (> 10,500/μL) | + 1 | Normal WBC (≤ 10,500/μL) | + 0 |
| Overall Predictive Discriminatory Ability | |||
| Prognostic Score | AUC (%) | 95% CI | P (v RMH) |
| Model A | 75.5 | 72.1 to 78.9 | .245 |
| Model B | 73.9 | 70.2 to 77.6 | .330 |
| RMH | 71.3 | 67.5 to 75.1 | 1.000 |
| Predictive Discriminatory Ability With Different Cutoffs | |||
| Prognostic Score | Specificity (%) | Sensitivity (%) | OCCR (%) |
| ECOG PS 0-1 | |||
| Model A score 6 v < 6 | 99.1 | 9.1 | 84.2 |
| Model B score 7 v < 7 | 99.8 | 6.4 | 83.7 |
| RMH score 3 v < 3 | 95.9 | 20.3 | 83.7 |
| Model A score ≥ 5 v < 5 | 90.4 | 44.3 | 82.1 |
| Model B score ≥ 5 v < 5 | 87.9 | 47.5 | 81.3 |
| RMH score ≥ 2 v < 2 | 85.5 | 50.1 | 79.9 |
| ECOG PS 2 | |||
| Model A score ≥ 7 v < 7 | 86.1 | 46.2 | 69.4 |
| RMH score 3 v < 3 | 87.2 | 40.7 | 72.6 |
| Model A score ≥ 6 v < 6 | 58.3 | 73.1 | 64.5 |
| RMH score ≥ 2 v < 2 | 56.4 | 81.5 | 71.0 |
Abbreviations: ALP, alkaline phosphatase; AUC, area under the curve; ECOG, Eastern Cooperative Oncology Group; LDH, lactate dehydrogenase; OCCR, overall correct classification rate; PS, performance score; RMH, Royal Marsden Hospital; TPTi, time per treatment index; ULN, upper limit of normal.
The impact on OS of the eight prognostic variables validated in models A and B for 90-day mortality was further evaluated by using Cox regression. All these variables were statistically significant, with PS 2 versus 0 carrying the largest hazard ratio (2.93). Other factors that were significantly associated with a worse outcome included PS 1 versus 0, albumin less than 3.5 g/dL, increased LDH, increased ALP, more than two metastatic sites, lymphocyte count less than 18%, WBC more than 10,500/μL, and CTGR less than 24 weeks per line of systemic treatment. The hazard ratio for these variables ranged from 1.33 to 1.77 with overlapping 95% CIs in most cases (Appendix Table A3, online only).
Subsequently, we applied the RMH score and model A separately to those patients with PS 2 (Table 3 provides details of calculation of scores). In both instances, patients with higher scores (RMH score of 3 and model A scores of 4 to 5) clearly had worse outcomes, with a median OS of less than 10 weeks (Fig 2A and Appendix Fig A1, online only). When these scoring systems and model B were applied to patients with PS 0 to 1, there was again clear stratification of OS outcomes (Figs 2B to 2D). In this setting, models A (score 6) and B (scores 6 to 7) identified patients with OS of less than 11 weeks, whereas the highest scoring stratum by the RMH model achieved a median OS of 14.6 weeks.
Fig 2. Kaplan-Meier overall survival (OS) curves. (A) OS curves according to the Royal Marsden Hospital (RMH) score for patients with Eastern Cooperative Oncology Group performance score (ECOG PS) 2 only (n = 70). OS curves in patients with PS 0 to 1 only (n = 1,502) according to (B) RMH score, (C) European prognostic score model A, and (D) European prognostic score model B. HR, hazard ratio.
This multi-institutional European study was carried out by using data from individual patients, and we were able to derive a precise, yet broad and representative picture of phase I oncology drug development activity in the modern era of molecularly targeted agents. The study comprised 2,182 patients treated from 2005 to 2007 in 14 independent oncology phase I units and therefore provides more up-to-date and precise estimates of efficacy and survival than those reported in smaller and single-institution series.3,6–16 Previous studies examining risks and benefits in phase I trials were based on summaries of phase I clinical trials rather than individual patient data; moreover, they covered trials during the 1990s and early 2000s when almost all studies involved cytotoxic chemotherapy.20,21
OS and objective response rates in our study are consistent with those in other recent studies.3,10,14,15,20,21 The significance of disease stabilization at 3 and 6 months is unclear in the absence of data from randomized trials. A systematic collection of individual patient data on the rate of disease progression before entry into phase I trials would be important in further analyses of treatment efficacy, and this has been highlighted as a future EDDN priority. The conduct of phase I trials has generally been considered safe,20,21 and the low rate of deaths related to toxicity (0.4%) observed in our study lends further support to this. In addition to providing an update on efficacy and safety, we wanted to identify factors leading to early trial discontinuation and death. Our results indicated that the main cause (approximately 70%) for early trial discontinuation was related to disease progression and/or concomitant medical events but not to treatment.
To pursue our main purpose—the identification and validation of prognostic factors—two robust prognostic models were constructed and validated and a total of eight prognostic variables were identified. Variables such as LDH, serum albumin, and number of metastatic sites were previously described as part of the RMH score.3,6 Other factors, such as WBC and lymphocytes (both included in model B) or PS (model A), were identified in other studies.7–9,11,14 In our study, a strong negative prognostic value of PS 2 was detected, supporting results from older series.8,9,11 In addition, two new prognostic factors were identified: ALP and TPTi. Neither the time from diagnosis to phase I trial entry nor the number of systemic treatment lines have previously been considered as independent prognostic factors in phase I trials, which could be explained by tumor and treatment heterogeneity. However, TPTi comprises both factors and thus may more accurately mirror the biologic behavior before study entry than either factor alone.18
The overall performances of the prognostic models of 90-day mortality derived from this study, according to their AUCs, were consistently over 70%, which indicates fair to good predictive discrimination accuracy. There was a trend toward improvement in overall correct classification rates when a larger number of variables was incorporated, but this improvement was small compared with the greater complexity introduced. Overall, optimal prognostic score cutoffs resulted in overall correct classification rates of more than 80%, high specificities (> 85%), and moderate sensitivities ranging from 44% to 51% (Table 3). We could also increase specificity and reduce sensitivity by increasing the score cutoffs and vice versa. It was interesting to note that overall, the novel prognostic models did not perform significantly differently from the RMH score (Table 3).
To understand the impact of using any of these models in patient selection for phase I trials in everyday practice, we have illustrated their application in a random series of 200 phase I a priori eligible patients (Fig 3) with an overall 90-day mortality rate of 16.5% (37 of 200 patients). Our data indicate that more than 70% of these patients who died within 90 days would have discontinued the trial within 21 days and, presumably, replacement within the trial would be necessary in most cases. Exclusion of these patients with a poor prognosis should therefore be beneficial to the conduct of trials by reducing patient attrition. However, this would also reduce overall recruitment by 20% (40 patients), thus depriving a proportion of eligible patients of the opportunity to participate in phase I trials.
Fig 3. Prognostic score performance. Each pie chart illustrates the total number and the proportion of patients surviving and dying within the first 90 days on trial. The pie chart on the left represents a random series of 200 patients initially eligible for phase I trials who had Eastern Cooperative Oncology Group performance score 0 to 1. The pie charts on the top right and bottom right illustrate the performance of Royal Marsden Hospital (RMH) score and the European prognostic score B, respectively. The application of both scores reduces the 90-day mortality by half, but it also reduces the total number of patients recruited by 20%.
Specific examples are noteworthy in the day-to-day application of these results to selection of patients for phase I oncology trials. For instance, patients with PS 2 are commonly excluded from phase I trials. However, as illustrated by our study, there is a wide spectrum of OS outcomes for patients with PS 2. At one end, patients with an RMH score of 3 clearly had the worst 90-day mortality rate (70%); the exclusion of such patients from phase I trials may result in the least harm to the patient and be advantageous for the conduct of the trial. At the other end, patients with PS 2 with the lowest RMH score had a 90-day mortality rate of 0%. For patients with PS 0 to 1, a high RMH score can differentiate a subgroup with a 90-day mortality of 45%, whereas the use of higher scores in the novel models A (score 6) and B (score 6 to 7) allows the identification of a subgroup of patients with an even higher 90-day mortality rate (60% to 70%).
Conversely, as many as 20% to 35% of patients with poor prognosis scores (ie, model A score 5 to 6, model B score 6 to 7, or RMH score 3) survive beyond 24 weeks. We hasten to add that a longer OS does not necessarily translate into clinical benefit, merely that all such patients should not be considered poor candidates a priori.
Although the results of our study provide additional validation to the RMH score, we also recognize the limitations of scoring systems built exclusively on clinical and routine analytic parameters, which do not take into account the impact of treatment. In addition, our database is entirely derived from a European population and might not be representative of early clinical trials conducted elsewhere. We also argue that a further major improvement in patient selection will be derived by the application of predictive molecular biomarkers that reflect tumor and host biology.22 To facilitate the large-scale analysis of this and other future strategies for optimizing the conduct of phase I clinical trials, collaborations between participating units, such as the EDDN, will be an important step forward.
Written on behalf of the European Drug Development Network.
Supported by the Drug Development Unit of the Royal Marsden National Health Service (NHS) Foundation and the Institute of Cancer Research and the European Drug Development Network (EDDN) centres. The Drug Development Unit of the Royal Marsden NHS Foundation and the Institute of Cancer Research are supported by a program core grant from Cancer Research UK, by Grant No. 110722 from the Experimental Cancer Medicine Centre, and by Grant No. CA51/A7401 from the National Institute of Health Research Biomedical Research Centre. The other UK Experimental Cancer Centres are also supported in part by program grants from Cancer Research UK and the respective Departments of Health. D.O. was a recipient of the Spanish Society of Medical Oncology Fellowship 2009-2011 and 2010 American Society of Clinical Oncology (ASCO) Annual Meeting Merit Award, and a 35th European Society for Medical Oncology (ESMO) Congress Merit Award were provided by the ASCO Cancer Foundation and ESMO Foundation.
Presented in part at the 46th Annual Meeting of the American Society of Clinical Oncology, Chicago, IL, June 4-8, 2010, and the 35th Annual Meeting of the European Society of Medical Oncology, Milan, Italy, October 8-12, 2010.
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
The author(s) indicated no potential conflicts of interest.
Conception and design: David Olmos, Roger P. A'Hern, Jaap Verweij, Patrick Schöffski, Josep Tabernero, Stan B. Kaye
Administrative support: David Olmos, Jaap Verweij, Patrick Schöffski, Stan B. Kaye
Provision of study materials or patients: David Olmos, Silvia Marsoni, Rafael Morales, Carlos Gomez-Roca, Jaap Verweij, Emile E. Voest, Patrick Schöffski, Nicolas Penel, Jan H. Schellens, Gianluca del Conte, Andre T. Brunetto, T.R. Jeffry Evans, Richard Wilson, Elisa Gallerani, Ruth Plummer, Josep Tabernero, Jean-Charles Soria, Stan B. Kaye
Collection and assembly of data: David Olmos, Silvia Marsoni, Rafael Morales, Carlos Gomez-Roca, Jaap Verweij, Emile E. Voest, Patrick Schöffski, Joo Ern Ang, Nicolas Penel, Jan H. Schellens, Gianluca del Conte, Andre T. Brunetto, T.R. Jeffry Evans, Richard Wilson, Elisa Gallerani, Ruth Plummer, Josep Tabernero, Stan B. Kaye
Data analysis and interpretation: David Olmos, Roger P. A'Hern, Jaap Verweij, Patrick Schöffski, Joo Ern Ang, Jean-Charles Soria, Stan B. Kaye
Manuscript writing: All authors
Final approval of manuscript: All authors
| 1. | J Verweij : ‘No risk, no fun': Challenges for the oncology phase I clinical trial time-performance Eur J Cancer 44: 2600– 2607,2008 Crossref, Medline, Google Scholar |
| 2. | EA Eisenhauer, PJ O'Dwyer, M Christian , etal : Phase I clinical trial design in cancer drug development J Clin Oncol 18: 684– 692,2000 Link, Google Scholar |
| 3. | HT Arkenau, D Olmos, JE Ang , etal : Clinical outcome and prognostic factors for patients treated within the context of a phase I study: The Royal Marsden Hospital experience Br J Cancer 98: 1029– 1033,2008 Crossref, Medline, Google Scholar |
| 4. | A Dowlati, S Manda, J Gibbons , etal : Multi-institutional phase I trials of anticancer agents J Clin Oncol 26: 1926– 1931,2008 Link, Google Scholar |
| 5. | ZA Nurgat, W Craig, NC Campbell , etal : Patient motivations surrounding participation in phase I and phase II clinical trials of cancer chemotherapy Br J Cancer 92: 1001– 1005,2005 Crossref, Medline, Google Scholar |
| 6. | HT Arkenau, J Barriuso, D Olmos , etal : Prospective validation of a prognostic score to improve patient selection for oncology phase I trials J Clin Oncol 27: 2692– 2696,2009 Link, Google Scholar |
| 7. | HT Arkenau, D Olmos, JE Ang , etal : 90-Days mortality rate in patients treated within the context of a phase-I trial: How should we identify patients who should not go on trial? Eur J Cancer 44: 1536– 1540,2008 Crossref, Medline, Google Scholar |
| 8. | T Bachelot, I Ray-Coquard, G Catimel , etal : Multivariable analysis of prognostic factors for toxicity and survival for patients enrolled in phase I clinical trials Ann Oncol 11: 151– 156,2000 Crossref, Medline, Google Scholar |
| 9. | C Han, JP Braybrooke, G Deplanque , etal : Comparison of prognostic factors in patients in phase I trials of cytotoxic drugs vs new noncytotoxic agents Br J Cancer 89: 1166– 1171,2003 Crossref, Medline, Google Scholar |
| 10. | A Italiano, C Massard, R Bahleda , etal : Treatment outcome and survival in participants of phase I oncology trials carried out from 2003 to 2006 at Institut Gustave Roussy Ann Oncol 19: 787– 792,2008 Crossref, Medline, Google Scholar |
| 11. | L Janisch, R Mick, RL Schilsky , etal : Prognostic factors for survival in patients treated in phase I clinical trials Cancer 74: 1965– 1973,1994 Crossref, Medline, Google Scholar |
| 12. | D Olmos, RD Baird, TA Yap , etal : Baseline circulating tumor cell counts significantly enhance a prognostic score for patients participating in phase I oncology trials Clin Cancer Res 17: 5188– 5196,2011 Crossref, Medline, Google Scholar |
| 13. | N Penel, JP Delord, ME Bonneterre , etal : Development and validation of a model that predicts early death among cancer patients participating in phase I clinical trials investigating cytotoxics Invest New Drugs 28: 76– 82,2010 Crossref, Medline, Google Scholar |
| 14. | N Penel, M Vanseymortier, ME Bonneterre , etal : Prognostic factors among cancer patients with good performance status screened for phase I trials Invest New Drugs 26: 53– 58,2008 Crossref, Medline, Google Scholar |
| 15. | J Wheler, AM Tsimberidou, D Hong , etal : Survival of patients in a Phase 1 Clinic: The M. D. Anderson Cancer Center experience Cancer 115: 1091– 1099,2009 Crossref, Medline, Google Scholar |
| 16. | N Yamamoto, T Tamura, M Fukuoka , etal : Survival and prognostic factors in lung cancer patients treated in phase I trials: Japanese experience Int J Oncol 15: 737– 741,1999 Medline, Google Scholar |
| 17. | AT Brunetto, JE Ang, D Olmos , etal : A study of the pattern of hospital admissions in a specialist Phase I oncology trials unit: Unplanned admissions as an early indicator of patient attrition Eur J Cancer 46: 2739– 2745,2010 Crossref, Medline, Google Scholar |
| 18. | JE Ang, D Olmos, Garcia V Moreno , etal: Utility of novel dynamic clinical indices in patients (pts) enrolled in a phase I (Ph I) oncology trial as markers of prognosis and treatment benefit J Clin Oncol 29: 178s,2011 suppl abstr 2555 Link, Google Scholar |
| 19. | P Therasse, SG Arbuck, EA Eisenhauer , etal : New guidelines to evaluate the response to treatment in solid tumors: European Organization for Research and Treatment of Cancer, National Cancer Institute of the United States, National Cancer Institute of Canada J Natl Cancer Inst 92: 205– 216,2000 Crossref, Medline, Google Scholar |
| 20. | E Horstmann, MS McCabe, L Grochow , etal : Risks and benefits of phase 1 oncology trials, 1991 through 2002 N Engl J Med 352: 895– 904,2005 Crossref, Medline, Google Scholar |
| 21. | TG Roberts Jr, BH Goulart, L Squitieri , etal : Trends in the risks and benefits to patients with cancer participating in phase 1 clinical trials JAMA 292: 2130– 2140,2004 Crossref, Medline, Google Scholar |
| 22. | DD Von Hoff, JJ Stephenson Jr, P Rosen , etal : Pilot study using molecular profiling of patients' tumors to find potential targets and select treatments for their refractory cancers J Clin Oncol 28: 4877– 4883,2010 Link, Google Scholar |
Hendrik-Tobias Arkenau, Daniel S.W. Tan, and Rebecca Kristeleit, The Royal Marsden National Health Service Foundation Trust, Sutton, United Kingdom; Alessandro Cattaneo, Southern Europe New Drug Organization Foundation, Milan, Italy; Jordi Rodon, Gaetano Aurilio, and Cristina Suarez, Vall d'Hebron University Hospital, Universitat Autònoma de Barcelona, Barcelona, Spain; Patricia de Vos and Merlijn Cranendock, Erasmus University Medical Center, Rotterdam, the Netherlands; Luca Gianni, Fondazione Istituto di Ricovero e Cura a Carattere Scientifico Istituto Nazionale dei Tumori, Milan, Italy; Cristiana Sessa, Istituto Oncologico della Svizzera Italiana, Bellinzona, Switzerland; Carol Hopkins, The Beatson West of Scotland Cancer Centre, Glasgow, United Kingdom; Angela Morrison, Centre for Cancer Research & Cell Biology, Queen's University Belfast, United Kingdom; Jane Margetts, Northern Centre for Cancer Care, Freeman Hospital, Newcastle upon Tyne, United Kingdom; Jacques Bonneterre, Centre Oscar Lambret, Lille, France; and Els Witteveen, University Medical Center Utrecht, Utrecht, the Netherlands.
The standard clinicopathologic parameters collected at study entry were tumor type, age, sex, Eastern Cooperative Oncology Group performance status (ECOG PS), CBC (hemoglobin, total WBC, lymphocyte count, and platelet count), biochemistry (including serum albumin, lactate dehydrogenase, AST, ALT, alkaline phosphatase, and total bilirubin), number of metastases, sites of metastasis, number of prior lines of systemic cancer treatments, time from advanced disease diagnosis to phase I study entry, and disease status at study entry. In addition, we calculated the time per treatment index, a log ratio of the time interval between diagnosis of advanced/metastatic cancer and phase I trial entry divided by the number of lines of systemic treatment [log (Time + 1/treatments + 1)].18 The number of lines of systemic anticancer treatment may be defined as the sum of all monotherapy or combination systemic therapies used to treat advanced/metastatic cancer (but not in the adjuvant or neoadjuvant setting) including any cytotoxic chemotherapy, hormonotherapy, immunotherapies, and molecularly targeted therapies. Other data collected included time to trial discontinuation and causes and tumor responses classified according to Response Evaluation Criteria in Solid Tumors (RECIST).19
The primary end point was survival beyond 90 days from study day 1. No a priori formal statistical plan was formulated. However, by using the Fisher-Irwin test, we estimated that 848 patients were required to validate a prognostic variable (dichotomized by the median), which associates with a minimum 1.5-fold increase in the risk of 90-day mortality if the following parameters are used: an incidence of 15% in 90-day mortality in the phase I population, an alpha error of 5%, and a beta error of 20%. We expected that the recruitment over 3 years at the 14 participating centers would yield sufficient patients for this study. In addition, two different models were defined a priori: model A (Fig 1A) consisted of 23 baseline variables in all eligible patients, whereas model B (Fig 1A) excluded patients with ECOG PS 2 and did not included PS as a variable in the model.
A posteriori we found that 1,845 and 1,780 patients were eligible for 90-day survival analysis in models A and B, respectively. Thus for each model, patients were randomly divided into two groups: one for the derivation of significant prognostic factors and the other for validation of the prognostic model. Each half yielded more than 848 patients who were eligible for analysis.
Before analysis, the tumor type variable was transformed in 10 separate binomial variables (breast cancer v other, upper GI cancer v other, lower GI cancer v other, genitourinary cancer v other, gynecologic cancer v other, sarcoma v other, melanoma v other, lung and mesothelioma v other, head and neck cancer v other, miscellaneous v other). Continues variables were initially analyzed without transformation for the derivation and validation of the prognostic models.
For the derivation of the prognostic model, we first identified factors related to 90-day mortality in univariate analyses by using the χ2 test, the Mantel-Haenszel χ2 test, or the Mann-Whitney U test when appropriate. Following this, all variables with a significance level less than 0.10 in the previous phase were considered in a multivariate analysis by using logistic regression. Subsequently, all the variables with a P value less than .05 (Wald test) in the derivation set were further tested in the validation set by using logistic regression.
To simplify the application of prognostic factors validated as continuous variables, we transformed them into binomial variables. All cutoffs for the conversion of these continuous variables into binary categories were chosen on the basis of consideration of the standard reference values (normal range v low or increased) or according to the median values. The only exception to this was the time per treatment index; despite a median of 20 weeks, the cutoff of 24 weeks (approximately 6 months) was chosen because it is a commonly used time point in clinical practice and trials. After dichotomization of each continuous variable with prognostic value, they were incorporated in a multivariate logistic regression analysis that used first the derivation set and then the validation set to confirm their prognostic value.
Following this, we explored whether more parsimonious models (simplest possible model with fewer variables) could be identified. Thus, all potential combinations of the previously validated prognostic factors, by number of factors (eg, one, two, three, up to seven), were constructed. By using the full data set, we estimated area under the curve (AUC) of the receiver operating characteristic curve for each combination. The AUCs were used to estimate the predictive discriminatory performance of each combination of prognostic factors. AUCs were also compared by using a nonparametric method.18,19
To further characterize the performance of the best prognostic scores, we estimated the sensitivities, specificities, and overall classification rates by using different prognostic score cutoffs (to define a positive or negative result of the test). For each combination, the mean and SE of each of these parameters were estimated from a total of 100 simulations each of 100 patients. Simulations were obtained from the full data set by using a random sampling technique with replacement.
The Kaplan-Meier method was used to estimate progression-free survival and overall survival. The effect of the 90-day mortality prognostic factors on overall survival was also studied. Early patient-related trial discontinuation (within 21 days) was defined as an event being directly or indirectly related to disease but not attributable to treatment. Prognostic factors for this event were explored by using a logistic regression model in all eligible patients for 21-day trial discontinuation analysis (Fig 1A). SPSS version 16.0 (SPSS, Chicago, IL) was used in the analysis.
|
| Therapy | No. | % |
|---|---|---|
| Receptor tyrosine kinases | 772 | 47.5 |
| Downstream-signaling kinases | 211 | 12.4 |
| Angiogenesis | 126 | 7.7 |
| Epigenetic/transcription regulation | 93 | 5.5 |
| Apoptosis | 84 | 4.9 |
| Cell cycle/mitosis regulation | 65 | 4.0 |
| DNA repair | 64 | 3.9 |
| Vascular disrupting agents | 53 | 3.2 |
| Protein degradation | 44 | 2.7 |
| Cell adhesion and migration | 30 | 1.8 |
| Stress/heat shock response | 22 | 1.3 |
| Steroid hormone synthesis | 17 | 1.0 |
| Others | 50 | 3.1 |
*In 67 of 1,698 patients receiving targeted therapies, the target was not detailed.
|
| No./Order | Factors Included | AUC (%) | Cp | −2 LL | χ2 P(v RMH) |
|---|---|---|---|---|---|
| Using prognostic factors derived from model A(includes patients with ECOG PS 2) * | |||||
| 3 factors | Albumin, LDH, ECOG PS | 73.80 | .625 | ||
| 4 factors | Albumin, LDH, ECOG PS, No. of metastatic sites | 74.40 | .458 | ||
| 5 factors | Albumin, LDH, ECOG PS, No. of metastatic sites, ALP | 75.00 | .354 | ||
| 6 factors | Albumin, LDH, ECOG PS, No. of metastatic sites, ALP, TPTi | 75.50 | .245 | ||
| Subsets of 1 variable | |||||
| 1 | Albumin | 68.61 | 98.278 | 1,493.20 | |
| 2 | LDH | 63.73 | 143.171 | 1,549.46 | |
| 3 | ECOG PS | 62.76 | 141.509 | 1,531.34 | |
| 4 | ALP | 61.59 | 184.818 | 1,384.67 | |
| 5 | No. of metastatic sites | 60.66 | 169.890 | 1,564.55 | |
| Subsets of 2 variables | |||||
| 1 | Albumin, ECOG PS | 71.53 | 63.543 | 1,452.31 | |
| 2 | Albumin, LDH | 70.97 | 49.018 | 1,452.66 | |
| 3 | Albumin, ALP | 70.45 | 89.032 | 1,289.01 | |
| 4 | LDH, ECOG PS | 70.18 | 80.134 | 1,480.19 | |
| 5 | Albumin, No. of metastatic sites | 69.93 | 76.967 | 1,469.43 | |
| Subsets of 3 variables | |||||
| 1 | Albumin, LDH ,ECOG PS | 73.70 | 15.386 | 1,412.51 | |
| 2 | Albumin, ECOG PS, ALP | 72.95 | 60.061 | 1,253.70 | |
| 3 | Albumin, ECOG PS, No. of metastatic sites | 72.26 | 49.233 | 1,435.55 | |
| 4 | Albumin, LDH, No. of metastatic sites (RMH score) | 72.22 | 54.938 | 1,418.73 | |
| 5 | Albumin, ECOG PS, TPTi | 72.21 | 32.903 | 1,433.71 | |
| Subsets of 4 variables | |||||
| 1 | Albumin, LDH, ECOG PS, No. of metastatic sites | 74.63 | 26.699 | 1,224.73 | |
| 2 | Albumin, LDH, ECOG PS, ALP | 74.31 | 5.392 | 1,400.24 | |
| 3 | Albumin, LDH, ECOG PS, TPTi | 74.05 | 7.246 | 1,379.82 | |
| 4 | Albumin, ECOG PS, ALP, TPTi | 73.61 | 50.982 | 1,232.18 | |
| 5 | Albumin, ECOG PS, No. of metastatic sites, ALP | 73.48 | 47.392 | 1,238.87 | |
| Subsets of 5 variables | |||||
| 1 | Albumin, LDH, ECOG PS, No. of metastatic sites, ALP | 75.09 | 17.379 | 1,213.18 | |
| 2 | Albumin, LDH, ECOG PS, ALP, TPTi | 75.04 | 18.415 | 1,204.19 | |
| 3 | Albumin, LDH, ECOG PS, No. of metastatic sites, TPTi | 74.79 | -4.488 | 1,365.43 | |
| 4 | Albumin, ECOG PS, No. of metastatic sites, ALP, TPTi | 74.12 | 37.085 | 1,216.04 | |
| 5 | Albumin, LDH, No. of metastatic sites, ALP, TPTi | 73.52 | 30.310 | 1,220.13 | |
| Subset of all 6 variables (best subset, AUC criterion) | |||||
| 1 | Albumin, LDH, ECOG PS, No. of metastatic sites, ALP, TPTi | 75.49 | 8.000 | 1,191.37 | |
| 10 best subsets (AUC criterion) | |||||
| 1 | Albumin, LDH, ECOG PS, No. of metastatic sites, ALP, TPTi | 75.49 | 8.000 | 1,191.37 | |
| 2 | Albumin, LDH, ECOG PS, No. of metastatic sites, ALP | 75.09 | 17.379 | 1,213.18 | |
| 3 | Albumin, LDH, ECOG PS, ALP, TPTi | 75.04 | 18.415 | 1,204.19 | |
| 4 | Albumin, LDH, ECOG PS, No. of metastatic sites, TPTi | 74.79 | -4.488 | 1,365.43 | |
| 5 | Albumin, LDH, ECOG PS, No. of metastatic sites | 74.63 | 26.699 | 1,224.73 | |
| 6 | Albumin, LDH, ECOG PS, ALP | 74.31 | 5.392 | 1,400.24 | |
| 7 | Albumin, ECOG PS, No. of metastatic sites, ALP, TPTi | 74.12 | 37.085 | 1,216.04 | |
| 8 | Albumin, LDH, ECOG PS, TPTi | 74.05 | 7.246 | 1,379.82 | |
| 9 | Albumin, LDH, ECOG PS | 73.70 | 15.386 | 1,412.51 | |
| 10 | Albumin, ECOG PS, ALP, TPTi | 73.61 | 50.982 | 1,232.18 | |
| Using prognostic factors derived from model B(does not include patients with ECOG PS 2and does not consider ECOG PS)† | |||||
| 3 factors | Albumin, LDH, No. of metastatic sites(RMH score) | 71.30 | 1.000 | ||
| 4 factors | Albumin, LDH, No. of metastatic sites, lymphocytes | 72.70 | .598 | ||
| 5 factors | Albumin, LDH, No. of metastatic sites, ALP, lymphocytes | 73.10 | .502 | ||
| 6 factors | Albumin, LDH, No. of metastatic sites, ALP, lymphocytes, WBC | 73.80 | .351 | ||
| 7 factors | Albumin, LDH, No. of metastatic sites, ALP, TPTi, lymphocytes, WBC | 73.90 | .330 | ||
| Subsets of 1 variable | |||||
| 1 | Albumin | 67.09 | 91.642 | 1,407.64 | |
| 2 | LDH | 66.19 | 128.045 | 1,415.72 | |
| 3 | No. of metastatic sites | 64.32 | 107.465 | 1,436.77 | |
| 4 | Lymphocytes | 62.95 | 138.288 | 1,456.21 | |
| 5 | WBC | 61.72 | 155.097 | 1,281.67 | |
| Subsets of 2 variables | |||||
| 1 | Albumin, LDH | 69.82 | 67.895 | 1,369.25 | |
| 2 | Albumin, No. of metastatic sites | 69.79 | 38.309 | 1,365.48 | |
| 3 | LDH, lymphocytes | 69.59 | 69.162 | 1,361.50 | |
| 4 | Albumin, No. of metastatic sites | 69.04 | 72.334 | 1,389.70 | |
| 5 | Albumin, lymphocytes | 68.93 | 82.784 | 1,209.70 | |
| Subsets of 3 variables | |||||
| 1 | Albumin , LDH, No. of metastatic sites (RMH score) | 71.79 | 20.158 | 1,323.44 | |
| 2 | Albumin , LDH, lymphocytes | 71.66 | 23.216 | 1,349.93 | |
| 3 | Albumin, LDH, WBC | 71.14 | 63.252 | 1,166.11 | |
| 4 | Albumin, ALP, lymphocytes | 71.03 | 66.762 | 1,174.83 | |
| 5 | LDH, ALP, lymphocytes | 70.99 | 46.005 | 1,180.22 | |
| Subsets of 4 variables | |||||
| 1 | Albumin, LDH, No. of metastatic sites, lymphocytes | 72.71 | 10.912 | 1,311.96 | |
| 2 | Albumin, LDH, ALP, lymphocytes | 72.69 | 29.572 | 1,139.60 | |
| 3 | Albumin, LDH, ALP, WBC | 72.55 | 34.860 | 1,168.25 | |
| 4 | Albumin, LDH, No. of metastatic sites, WBC | 72.40 | 13.934 | 1,337.96 | |
| 5 | Albumin, LDH, lymphocytes, WBC | 72.19 | 16.072 | 1,319.78 | |
| Subsets of 5 variables | |||||
| 1 | Albumin, LDH, No. of metastatic sites, ALP, lymphocytes | 73.38 | 19.903 | 1,127.79 | |
| 2 | Albumin, LDH, No. of metastatic sites, ALP, WBC | 73.30 | 24.988 | 1,155.56 | |
| 3 | Albumin, LDH, No. of metastatic sites, TPTi, lymphocytes | 73.11 | 0.070 | 1,276.31 | |
| 4 | Albumin, LDH, No. of metastatic sites, lymphocytes, WBC | 73.00 | 6.995 | 1,308.38 | |
| 5 | Albumin, LDH, ALP, lymphocytes, WBC | 72.93 | 27.219 | 1,137.21 | |
| Subsets of 6 variables | |||||
| 1 | Albumin, LDH, No. of metastatic sites, ALP, lymphocytes, WBC | 73.68 | 17.671 | 1,125.43 | |
| 2 | Albumin, LDH, No. of metastatic sites, ALP, TPTi, lymphocytes | 73.61 | 8.659 | 1,104.38 | |
| 3 | Albumin, LDH, No. of metastatic sites, TPTi, lymphocytes, WBC | 73.39 | −2.228 | 1,273.76 | |
| 4 | Albumin, LDH, No. of metastatic sites, ALP, TPTi, WBC | 73.34 | 15.804 | 1,133.66 | |
| 5 | Albumin, LDH, ALP, TPTi, lymphocytes, WBC | 73.02 | 18.728 | 1,115.74 | |
| Subset of all 7 variables (best subset, AUC criterion) | |||||
| 1 | Albumin, LDH, No. of metastatic sites, ALP, TPTi, lymphocytes, WBC | 73.82 | 8.000 | 1,102.99 | |
| 10 best subsets (AUC criterion) | |||||
| 1 | Albumin, LDH, No. of metastatic sites, ALP, TPTi, lymphocytes, WBC | 73.82 | 8.000 | 1,102.99 | |
| 2 | Albumin, LDH, No. of metastatic sites, ALP, lymphocytes, WBC | 73.68 | 17.671 | 1,125.43 | |
| 3 | Albumin, LDH, No. of metastatic sites, ALP, TPTi, lymphocytes | 73.61 | 8.659 | 1,104.38 | |
| 4 | Albumin, LDH, No. of metastatic sites, TPTi, lymphocytes, WBC | 73.39 | -2.228 | 1,273.76 | |
| 5 | Albumin, LDH, No. of metastatic sites, ALP, lymphocytes | 73.38 | 19.903 | 1,127.79 | |
| 6 | Albumin, LDH, No. of metastatic sites, ALP, TPTi, WBC | 73.34 | 15.804 | 1,133.66 | |
| 7 | Albumin, LDH, No. of metastatic sites, ALP, WBC | 73.30 | 24.988 | 1,155.56 | |
| 8 | Albumin, LDH, No. of metastatic sites, TPTi, lymphocytes | 73.11 | 0.070 | 1,276.31 | |
| 9 | Albumin, LDH, ALP, TPTi, lymphocytes, WBC | 73.02 | 18.728 | 1,115.74 | |
| 10 | Albumin, LDH, No. of metastatic sites, lymphocytes, WBC | 73.00 | 6.995 | 1,308.38 |
Abbreviations: −2LL, log likelihood; ALP, alkaline phosphatase; AUC, area under the curve; Cp, Mallows' Cp, fitness coefficient of a regression model; ECOG, Eastern Cooperative Oncology Group; LDH, lactate dehydrogenase; PS, performance score; RMH, Royal Marsden Hospital; TPTi, time per treatment index.
*Includes best performing prognostic models for 90-day mortality by number of parameters (3 to 6) and AUC.
†Includes best performing prognostic models for 90-day mortality by number of parameters (3 to 7) and AUC.
|
| Variable | Median OS (weeks) | 95% CI | HR | 95% CI | Log-rank P |
|---|---|---|---|---|---|
| ECOG PS | .0031−22 | ||||
| 0 | 54 | 49 to 59 | |||
| 1 | 31 | 28 to 34 | 1.60 | 1.44 to 1.78 | |
| 2 | 17 | 13 to 22 | 2.93 | 2.22 to 3.86 | |
| Serum albumin | .0026−21 | ||||
| Normal (≥ 3.5 g/dL) | 47 | 43 to 50 | 1.77 | 1.58 to 1.97 | |
| Low (< 3.5 g/dL) | 25 | 22 to 26 | |||
| LDH | |||||
| Normal (≤ ULN) | 47 | 43 to 52 | 1.60 | 1.43 to 1.79 | .0044−14 |
| Increased (> ULN) | 30 | 26 to 33 | |||
| No. of metastatic sites | |||||
| Low (≤ 2) | 44 | 41 to 48 | 1.46 | 1.31 to 1,63 | .0067−9 |
| High (≥ 2) | 28 | 24 to 31 | |||
| TPTi, weeks | |||||
| Long (≥ 24) | 51 | 46 to 56 | 1.36 | 1.21 to 1.53 | .0045−4 |
| Short (< 24) | 36 | 33 to 38 | |||
| ALP | |||||
| Normal (≤ ULN) | 44 | 40 to 49 | 1.32 | 1.19 to 1.47 | .0033−4 |
| Increased (> ULN) | 32 | 29 to 35 | |||
| Lymphocyte count | |||||
| Normal/high (≥ 18%) | 53 | 48 to 57 | 1.70 | 1.54 to 1.88 | .0068−22 |
| Low (< 18%) | 29 | 27 to 32 | |||
| WBC | |||||
| Normal (≤ 10,500/μL) | 41 | 39 to 44 | 1.59 | 1.38 to 1.82 | .0034−8 |
| Increased (> 10,500/μL) | 24 | 20 to 28 |
Abbreviations: ALP, alkaline phosphatase; ECOG, Eastern Cooperative Oncology Group; HR, hazard ratio; LDH, lactate dehydrogenase; OS, overall survival; PS, performance score; TPTi, time per treatment index; ULN, upper limit of normal.
Fig A1. Kaplan-Meier overall survival (OS) curves according to European prognostic score A in patients with Eastern Cooperative Oncology Group (ECOG) performance score 2 only (n = 70). All patients had a score higher than 3 (scoring included albumin, lactate dehydrogenase, number of metastatic sites, alkaline phosphatase, and clinical tumor growth rate in addition to ECOG 2 [score +3]).
