Download Relationship between objective responses in phase I trials and

Original article
Annals of Oncology 13: 1300–1306, 2002
DOI: 10.1093/annonc/mdf202
Relationship between objective responses in phase I trials
and potential efficacy of non-specific cytotoxic investigational
new drugs
I. Sekine*, N. Yamamoto, H. Kunitoh, Y. Ohe, T. Tamura, T. Kodama & N. Saijo
Department of Internal Medicine, National Cancer Center Hospital, Tokyo, Japan
Received 24 September 2001; revised 7 December 2001; accepted 9 January 2002
Background: Although the evaluation of new investigational drugs in phase I, II and III trials requires
considerable time and patient resources, only a few of these drugs are ultimately established as anticancer drugs.
Materials and methods: We collected papers of phase I trials by a Medline search using the key
words ‘Neoplasms/Drug Therapy in MeSH’ and ‘Phase I’ for the period from 1976 to 1993. A drug was
defined as ‘effective’ if a regimen including the drug produced positive results in at least one phase III
trial. We analyzed the relationship between objective (complete and partial) responses in phase I trials
and the effectiveness of agents in phase III trials.
Results: A total of 399 single-agent phase I trials of cytotoxic agents in adult patients with solid tumors
were obtained. Further clinical investigation was not recommended in 36 trials (9%) because of severe
toxicity. In the remaining 363 trials, 174 drugs were evaluated and the median number of trials for each
drug was two (range one to nine). Objective responses were observed in 495 (4.1%) of 12 076 patients,
178 (49%) of 363 trials, and 115 (66%) of 174 drugs. Of the 174 drugs, 48 (28%) were considered to be
effective. Percentages of effective drugs rose as the number of responders in phase I trials increased.
Logistic regression analyses showed the number of responders to be significantly associated with drug
effectiveness [odds ratio = 1.16 (1.06–1.27), P = 0.001 for 174 drugs; odds ratio = 1.16 (1.05–1.28),
P = 0.0038 for 363 trials]. Although 10 active drugs failed to produce an objective response in phase I
trials, seven of them produced a tumor regression of <50%, and three reportedly produced objective
responses in phase I trials conducted before 1975. The numbers of responders among patients with lung,
ovarian, breast or colorectal cancer, but not those among patients with lymphoma, melanoma, sarcoma
or renal-cell carcinoma, were associated significantly with drug effectiveness against the respective
tumors.
Conclusions: Objective responses observed in phase I trials are important for determining the future
development of an anticancer drug.
Key words: investigational new drugs, objective response, phase I trial, phase III trial
Introduction
Investigational new drugs (INDs) are evaluated in phase I, II
and III trials. The objectives of phase I trials are to evaluate
toxicity and determine a recommended dose for phase II trials.
The phase II trials are conducted to determine the response
rate (RR) in patients with a specific tumor, and therefore as
many phase II trials as the number of tumor types are theoretically required. In addition, combinations of an IND with other
*Correspondence to: Dr I. Sekine, Department of Internal Medicine,
National Cancer Center Hospital, Tsukiji 5-1-1, Chuo-ku,
Tokyo 104-0045, Japan. Tel: +81-3-3542-2511; Fax: +81-3-3545-3567;
E-mail: [email protected]
© 2002 European Society for Medical Oncology
anticancer agents are also evaluated before the phase III trials,
because combination chemotherapy including two to four
agents rather than a single agent is commonly used in these
trials as well as in clinical practice. The end point of the phase
III trials is usually patient survival, and in the limited situation,
quality of life. To establish the effectiveness of an IND or a
new regimen containing an IND, these end points are assessed
in comparison with a reference arm, the standard treatment for
the disease [1].
These processes are costly, time-consuming and require
considerable patient resources, but only a few drugs are finally
established as anticancer drugs. Because any INDs for which
intolerable toxicity is not observed in a phase I trial have
theoretically been considered suitable for phase II trials [2, 3],
1301
the INDs will often undergo several phase II trials, even if they
show no activity, before development of the agent is discontinued. Furthermore, several INDs entering phase III trials
were ultimately determined to be ineffective. Thus, it is
important to decide whether development of an IND should be
continued or terminated in the early phase of clinical trials. In
order to predict potential drug efficacy from the results of
phase I trials, we analyzed the relationship between objective
(complete and partial) responses in phase I trials and the final
effectiveness of the agents in phase III trials.
Table 1. Classes of agents evaluated in 174 phase I trials
Agent class
No. of agents (%)
Antimetabolites
41 (24)
Alkylating agents
29 (17)
Anthracyclines and their related compounds
22 (13)
Platinum compounds
13 (7)
Microtubule inhibitors
10 (6)
Topoisomerase II inhibitors
6 (3)
Topoisomerase I inhibitors
Materials and methods
A Medline search using the keywords ‘Neoplasms/Drug Therapy in
MeSH’ and ‘Phase I’ during the period from 1976 to 1993 yielded 1927
papers. Of these, 547 reported single-agent phase I trials in adult patients
with malignancies. A total of 399 trials were selected from the 547,
according to the following criteria: (i) trials of cytotoxic agents; (ii) trials
of patients with solid tumors; and (iii) trials in which tumor responses
were reported. Of the 148 excluded trials, 94 were trials of non-cytotoxic
agents, 30 were trials for patients with leukemia, and in 24 antitumor
effects were not reported.
Objective tumor response was defined as the disappearance of all
known disease or a decrease of at least 50% in total tumor size for at least
4 weeks without the appearance of new lesions. Response rate was determined by the following equation: RR = [(No. responders)/(No. eligible
and treated patients)] × 100 (%).
Phase III trials of a regimen including an IND versus a conventional
treatment, and phase III trials comparing different IND doses were collected by Medline searches using the keywords ‘phase III, randomized, or
compare’ and a drug ‘name’. Of these, a positive study was defined as one
in which an IND regimen showed a better RR, progression-free survival,
or overall survival than a conventional treatment, or in which an investigational arm showed the same RR, progression-free survival and overall
survival, but less toxicity or improved quality of life when compared with
a reference arm. A drug was defined as ‘effective’ if at least one positive
study was identified.
For a drug evaluated in two or more phase I trials, the total numbers of
patients and responders were compiled. The numbers of responders and
RRs were counted for each drug and for each trial. Associations between
these response parameters in phase I trials and drug effectiveness in phase
III trials were evaluated using logistic regression analyses. The actual
calculations were carried out using the STATISTICA (Stat-Soft, Tulsa,
OK, USA) software system.
Results
Of the 399 single-agent phase I trials of cytotoxic agents in
adult patients with solid tumors, further clinical investigation
was not recommended in 36 trials (9.02%) because of severe
toxicity. Objective responses were analyzed for the current
study in the remaining 363 trials. The median (range) number
of patients in a trial was 29 (6–139). The total number of
agents evaluated in the 363 trials was 174. Median (range)
numbers of trials and patients for each agent were two (1–9)
and 52.5 (7–334), respectively. The most common class of
agents was antimetabolites (24%), followed by alkylating
2 (1)
Others
51 (29)
Table 2. Numbers of responders for 174 agents and 363 trials
Tumor response
No. of agents (%)
No. of trials (%)
0
59 (33.7)
185 (51.0)
1
33 (19.4)
69 (19.0)
2
27 (15.4)
52 (14.3)
3–4
21 (12.1)
28 (7.7)
5–9
21 (12.1)
23 (6.3)
10–19
9 (5.1)
5 (1.4)
20≤
4 (2.3)
1 (0.3)
0
59 (33.7)
185 (51.0)
0.1–5.0
64 (37.1)
73 (20.1)
5.1–10
32 (17.7)
59 (16.3)
10.1≤
19 (11.4)
46 (12.7)
No. of responders
Response rates
agents (17%) and anthracyclines and their related compounds
(13%) (Table 1). The total number of patients enrolled in the
363 trials was 12 686. Of these, 291 ineligible (2.3%) or
untreated patients, and 319 patients (2.5%) with leukemia
were excluded from the current analysis. Thus, 12 076 patients
were the subjects of this study. Tumor types in these patients
were lung cancer in 2458 (20.3%), colorectal cancer in 2080
(17.2%), breast cancer in 925 (7.6%), melanoma in 738 (6.1%),
sarcoma in 639 (5.3%), head and neck cancer in 566 (4.7%),
renal cell cancer in 524 (4.3%), ovarian cancer in 522 (4.3%),
gastric cancer in 389 (3.2%), lymphoma in 309 (2.5%), pancreatic cancer in 192 (1.6%), cervical cancer in 165 (1.3%),
prostate cancer in 133 (1.1%), bladder cancer in 128 (1.0%),
esophageal cancer in 89 (0.7%), brain tumor in 73 (0.6%), others
in 889 (7.3%) and not reported in 1260 (10.4%).
Objective responses were observed in 495 (4.1%) of the
12 076 patients, 178 (49%) of the 363 trials and 115 (66%) of
the 174 drugs. The number of responders and RRs are summarized in Table 2. The RRs by tumor type were evaluated in
10 816 patients whose tumor type was reported. The remain-
1302
ing 1260 (10.1%) patients whose tumor type was not specified
were excluded from this analysis. None of these 1260 patients
responded to chemotherapy. In the 10816 patients, the RRs
were <10% regardless of tumor type, except for lymphoma
and brain tumor, both of which had an RR of 18% (Table 3).
Of the 174 agents evaluated in phase I trials, 61 (35.0%)
went into phase III trials, and 48 (27.6%) had at least one
positive phase III study and were determined to be effective
(Table 4). Median (range) RRs in phase I studies for the
48 agents that had at least one positive phase III study, the
13 agents that had negative phase III studies, and the remaining 113 agents that did not go into phase III studies were 5.3%
(0–22%), 2.7% (0–13%) and 1.9% (0–29%), respectively.
Only differences in RRs between the first and the third were
statistically significant (P <0.001, Mann–Whitney test). As to
agents that produced two or more responders in a phase I trial,
or a total of five or more responders in phase I trials for the
agent, approximately half of such agents were effective in
phase III trials (Table 5). There was a significant correlation
between the number of responses in phase I trials and the
number of effective agents. The odds ratios of the number of
responses for efficacy were 1.16 [95% confidence interval
(CI) 1.06–1.27] for the 174 agents and 1.16 (95% CI 1.05–1.28)
for the 363 trials. A similar correlation was also observed
between RRs and the number of effective agents (Table 6).
Approximately half of the agents producing an RR of >5% in
a phase I trial or an RR of >10% in phase I trials for the agent
were determined to be effective. The odds ratios of the RRs for
Table 3. Numbers of objective responses and response rates by tumor type
Tumor type
Responders (n = 495)
Number
%
107
19.2
Lymphoma
54
Melanoma
45
Ovarian cancer
Breast cancer
Lung cancer
Total number of patients
(n = 10 816)
Response rate (%)
2458
4.4
10.9
309
17.5
9.1
738
6.1
44
8.9
522
8.4
44
8.9
625
4.8
Colorectal cancer
33
6.7
2080
1.6
Sarcoma
32
6.5
639
5.0
Renal-cell carcinoma
21
4.2
524
4.0
Head and neck cancer
15
3.0
566
2.7
Brain tumor
13
2.6
73
17.8
Prostate cancer
10
2.0
133
7.5
Gastric cancer
9
1.8
389
2.3
Bladder cancer
8
1.6
128
6.3
Cervical cancer
7
1.4
165
4.2
Pancreatic cancer
5
1.0
192
2.6
Esophageal cancer
Others
3
0.6
89
3.4
45
9.1
889
9.3
Table 4. Number of positive studies for the 174 agents
No. of
positive studies
No. of
agents
Agents
1
9
Hexamethylmelamine, aclarubicin, 6-thioguanine, triazinate, 6-mercaptopurine, suramin, mitolactol,
nedaplatin, peptichemio
2
9
Temozolomide, PCNU, floxuridine, pepleomycin, aziridinylbenzoquinone, actinomycin D, oxaliplatin,
topotecan, teniposide
3–5
8
Streptozotocin, cytarabine, doxifluridine, tegafur, amsacrine, chlorambucil, pirarubicin, irinotecan
6–10
9
Thiotepa, paclitaxel, docetaxel, lonidamine, idarubicin, carmofur, hydroxyurea, ifosfamide, gemcitabine
11≤
13
Vindesine, mitoxantrone, cyclophosphamide, semustine, daunorubicin, epirubicin, 5-fluorouracil, methotrexate,
vinblastine, vincristine, carboplatin, cisplatin, etoposide
There were 126 agents for which no positive studies were reported. PCNU, [1-(2-chloroethyl)-3-(2,6-dioxo-3-piperidly)-1-nitosourea].
1303
Table 5. Responses in phase I trials and efficacies of the agents
No. of responders
No. of drugs
Total
No. of trials
Effective drugs
%
Total
Trials of
effective drugs
%
10
17
185
49
26
0
59
1
33
2
6
69
18
26
2
27
10
37
52
30
58
3–4
21
7
33
28
15
54
5–9
21
10
48
23
14
61
10≤
13
9
69
6
2
33
Total
174
48
28
363
128
35
OR (95% CI)
1.16 (1.06–1.27),
P = 0.001
Trials of
effective drugs
%
1.16 (1.05–1.28),
P = 0.0038
OR, odds ratio for the number of responders in phase I trials; CI, confidence interval.
Table 6. Response rates in phase I trials and efficacies of the agents
Response rate (%)
No. of drugs
Total
No. of trials
Effective drugs
%
Total
0
59
10
17
185
49
26
0.1–5.0
64
14
22
73
20
27
5.1–10.0
32
12
38
59
27
46
10.1≤
19
12
63
46
32
70
Total
174
48
28
363
128
35
OR (95% CI)
1.14 (1.06–1.22),
P <0.001
1.11 (1.07–1.16),
P <0.001
OR, odds ratio for the response rate in phase I trials; CI, confidence interval.
effectiveness were 1.14 (95% CI 1.06–1.22) for the 174 agents
and 1.11 (95% CI 1.07–1.16) for the 363 trials.
Responses in phase I trials and efficacies of the agents
according to tumor type were evaluated in patients with eight
tumor types for which >20 responders were identified. The
numbers of responders among patients with lung, ovarian and
colorectal cancer were associated with the numbers of effective agents for the respective tumors, but not with the numbers
of effective agents for the other tumors. The number of
responders among breast cancer patients was associated with
the numbers of effective agents for breast cancer as well as the
other cancers. Among patients with lymphoma, melanoma,
sarcoma or renal-cell carcinoma, however, the numbers of
responders were not associated with the numbers of effective
agents for any of the tumors (Table 7).
Ten agents that failed to produce an objective response in
phase I trials were found to be effective in phase III trials. Of
these, seven produced a tumor regression of <50%, and three
were reported to produce objective responses in phase I trials
conducted before 1975 (Table 8).
Discussion
Although phase I trials are not designed primarily to evaluate
the therapeutic effects of an agent, a few authors have
discussed the relationship between the responses observed
during phase I trials and clinical drug development. Estey et
al. [4] reviewed 187 phase I trials and found that the median
phase I RR for active agents in phase II trials was 4.3%, and
2.7% for those inactive in phase II trials, with no significant
difference between the two RRs. In contrast, Von Hoff and
Turner [5] reviewed 228 phase I trials and found that all of the
agents that made it to market had a median of six responders in
their phase I trials. They concluded that if no responses were
noted in a phase I trial, the agent had a low probability of being
marketed. In the present study, we have demonstrated the
number of responders and RRs in phase I trials to be important
information for predicting potential drug effectiveness in
phase III trials. If two or more responders and/or an RR of
>5% were noted in a phase I trial, half of such agents will
possibly be efficacious against cancer. Of 10 agents without
objective responses in phase I trials, nine showed minor
1304
Table 7. Response in phase I trials and efficacies of the agents by tumor type
No. of responders in
phase I trials
No. of agents
Agents effective against the tumor
No.
Agents effective against other tumors
%
P
No.
%
P
Lung cancer
0
135
6
4
–
18
13
–
1–2
30
6
20
–
11
37
–
3≤
9
6
67
<0.01
5
56
NS
Ovarian cancer
0
162
6
4
–
30
19
–
1–2
8
2
25
–
5
63
–
3≤
4
1
25
<0.05
2
50
NS
Breast cancer
0
149
7
5
–
22
15
–
1–2
19
4
21
–
8
42
–
3≤
6
3
50
<0.001
5
83
<0.001
Colorectal cancer
0
152
6
4
–
25
16
–
1–2
19
4
21
–
4
21
–
3≤
3
1
33
<0.01
1
33
NS
Lymphoma
0
144
6
4
–
26
18
–
1–2
28
3
11
–
10
36
–
3≤
2
0
0
0
0
NS
NS
Melanoma
0
148
2
14
–
28
19
–
1–2
22
2
9
–
7
32
–
3≤
4
0
0
NS
1
25
NS
Sarcoma
0
161
4
2
–
31
19
–
1–2
11
0
0
–
6
55
–
3≤
2
0
0
NS
0
0
NS
Renal-cell carcinoma
0
156
1
0.6
–
33
21
–
1–2
18
0
0
–
5
28
–
3≤
0
–
–
NS
–
NS
–
NS, not statistically significant.
responses in this study or objective responses in phase I trials
conducted before 1975. Thus, if no tumor shrinkage effect is
obtained in a phase I trial, further development of the drug
would not be recommended.
Selection of tumor types for phase II trials is another issue
to be discussed. The National Cancer Institute clinical drug
development program in the 1970s recommended that each
drug be evaluated in phase II trials involving at least six different tumor types, including breast, colon and lung cancers, as
well as melanoma, acute leukemia and lymphoma [6]. Recent
development of a pre-clinical screening system for INDs,
using well-characterized tumor cell lines or focusing on welldefined molecular targets, may provide useful and reliable
information on the activity of an agent against a particular type
of tumor [7]. This study suggested that an IND producing
objective responses in patients with lung, breast, ovarian or
colorectal cancer may be among the ‘very promising’ agents
for treatment of the respective tumors. Thus, it is reasonable to
include patients with various common cancers in phase I trials
for exploratory evaluation of the activity of an agent. In
1305
Table 8. Active agents that failed to produce an objective response in phase I trials
Agent
Responses in phase I trials conducted before 1975
Responses <50% in the current study (no. of patients)
Carmofur
–
None
Hexamethylmelamine
Objective responsea
None
Methotrexate
–
Minor response (1)
Teniposide
Objective responseb
None
Chlorambucil
–
Minor response (1)
Nedaplatin
–
Twenty percent tumor regression (1)
Lonidamine
–
Minor response (2)
Mitolactol
Objective responsec
Minor response (4)
Aziridinylbenzoquinone
–
Minor response (4)
Hydroxyurea
–
Minor response (1)
a
Cancer Chemother Rep 1965; 48: 49–52.
Cancer Chemother Pharmacol 1982; 7: 87–91.
c
Cancer Chemother Rep 1971; 55: 61–65.
b
contrast, no such association was observed in lymphoma,
melanoma, sarcoma or renal-cell carcinoma. Relatively large
numbers of patients with melanoma, sarcoma or renal-cell
carcinoma have been enrolled in phase I trials, because few
standard chemotherapeutic regimens have been established
for these tumors. If an objective response is observed in
patients with these tumors, however, this would not support
going into phase II trials for any type of tumors.
The main limitation of the current study is related to the
design of phase I trials, which are not directed at evaluating
responses: (i) many patients in phase I trials have no measurable lesion; (ii) the tumor type was not reported in 10.4%
(1260 of 12076) of patients; (iii) the patient population was
highly heterogeneous with regard to previous treatment, performance status and disease extent, all of which are thought to
influence the response to an agent. For the first reason, we calculated RRs using the number of eligible and treated patients
as the denominator in the equation in place of the number of
patients evaluable for responses. This substitution is thought
to have little influence on the results of our study, because the
association between RRs in phase I trials and drug effectiveness in phase III trials did not differ from that between the
number of responders and drug effectiveness.
Biases inherent in patient accrual and publication of phase I
studies are to be pointed out. More phase I studies and more
patient resources are often invested in compounds that demonstrate some responsiveness in phase I studies. These studies
are also more likely to be reported or published. Thus, it is a
self-fulfilling prophecy that higher RRs will be seen with
active agents in phase I studies, especially in published phase I
studies. Similarly, once an agent has demonstrated activity in
a specific tumor type in phase I studies, there is a skewing of
accrued patients with that specific tumor type, which would
skew activity away from other tumor types. The definition of
‘effective agents’ in this study is another issue. It is arbitrary
and may not be optimal. We consider ‘at least one positive
phase III study’ to be clear, objective, and easy to confirm,
although there remains room for bias in the criteria for a
‘positive study’. We evaluated all phase III trials identified by
Medline searches regardless of their quality, because the
quality of phase III trials is greatly affected by the publication
year. Old trials are often of lower quality than more recent
trials, but the former can be very important for assessing longsustained antineoplastic agents. Von Hoff and Turner [5] used
‘marketing of the agent’ as the definition of an efficacious
agent. We did not adopt their definition in this study, because
government approval requirements for INDs vary among
countries. Therefore, ‘marketing of the agent’ cannot serve as
a standardized criterion. Finally, this retrospective analysis is
largely focused on non-specific cytotoxic agents developed in
a period that might be very different from the present time. To
evaluate recent cytotoxic agents whose molecular target is
established, target inhibition as well as tumor responses in
phase I studies may be important factors for further development of the agent.
In conclusion, the numbers of responders and RRs in phase
I trials of INDs successfully predicted their potential efficacy.
The number of responders among lung, ovarian, breast and
colorectal cancer cases, but not among lymphoma, melanoma,
sarcoma and renal-cell carcinoma cases, was associated
significantly with drug efficacy against the respective tumors.
Objective responses observed in phase I trials are important
for determining the direction of future IND trials.
Acknowledgements
We thank Mihoko Kanazawa for her collection and arrangement of numerous reports on phase I trials.
1306
References
1. Simon R. Design and analysis of clinical trials. In DeVita VT Jr,
Hellman S, Rosenberg SA (eds): Cancer: Principles and Practice of
Oncology, 6th edition. Philadelphia, PA: Lippincott Williams &
Wilkins 2001; 521–538.
2. Wittes RE, Marsoni S, Simon R, Leyland-Jones B. The phase II trial.
Cancer Treat Rep 1985; 69: 1235–1239.
3. Marsoni S, Hoth D, Simon R et al. Clinical drug development: an
analysis of phase II trials, 1970–1985. Cancer Treat Rep 1987; 71:
71–80.
4. Estey E, Hoth D, Simon R et al. Therapeutic response in phase I trials
of antineoplastic agents. Cancer Treat Rep 1986; 70: 1105–1115.
5. Von Hoff DD, Turner J. Response rates, duration of response, and
dose response effects in phase I studies of antineoplastics. Invest New
Drug 1991; 9: 115–122.
6. DeVita VT, Oliveriio VT, Muggia FM et al. The drug development
and clinical trials programs of the division of cancer treatment,
National Cancer Institute. Cancer Clin Trial 1979; 2: 195–216.
7. Chu E, Grever MR, Chabner BA. Cancer drug development. In
DeVita VT Jr, Hellman S, Rosenberg SA (eds): Cancer: Principles
and Practice of Oncology, 6th edition. Philadelphia, PA: Lippincott
Williams & Wilkins 2001; 345–356.