Recruitment, outcomes, and toxicity trends in phase I oncology trials: Six‐year experience in a large institution

Abstract Background With the rapid influx of novel anti‐cancer agents, phase I clinical trials in oncology are evolving. Historically, response rates on early phase trials have been modest with the clinical benefit and ethics of enrolment debated. However, there is a paucity of real‐world data in this setting. Aim To better understand the changing landscape of phase I oncology trials, we performed a retrospective review at our institution to examine patient and trial characteristics, screening outcomes, and treatment outcomes. Methods and results We analyzed all consecutive adult patients with advanced solid organ malignancies who were screened across phase I trials from January 2013 to December 2018 at a single institution. During this period, 242 patients were assessed for 28 different trials. Median age was 64 years (range 30–89) with an equal sex distribution. Among 257 screening visits, the overall screen failure rate was 18%, resulting in 212 patients being enrolled onto a study. Twenty‐six trials (93%) involved immunotherapeutic agents or molecular targeted agents either alone or in combination, with only two trials of cytotoxic agents (7%). Twenty‐two (13.4%) of the 209 treated patients experienced a total of 33 grade 3 or higher treatment‐related adverse events. There was one treatment‐related death (0.5%). Of 190 response‐evaluable patients, 7 (4%) had a complete response, 34 (18%) a partial response, and 59 (31%) experienced stable disease for a disease control rate of 53%. The median overall survival for our cohort was 8.0 (95% CI: 6.8–9.2) months. Conclusion The profile of phase I trials at our institution are consistent with the changing early drug development landscape. Response rates and overall survival in our cohort are superior to historically reported rates and comparable to contemporaneous studies. Severe treatment‐related toxicity was relatively uncommon, and treatment‐related mortality was rare.


| INTRODUCTION
Phase I trials represent a crucial step wherein a novel therapeutic agent makes the transition from the pre-clinical to clinical stage, thus providing a foundation for a potentially successful drug development program. 1 These studies involve the early exploration of treatments or treatment combinations in humans. Determination of safety and tolerability is the primary objective, as well as establishing the maximum tolerated dose and/or the recommended phase II dose. 2 However, early phase trials in oncology historically have had low success rates, with the chance of eventual approval for a tested drug being 7%-the lowest among all medical specialties as reported in a 2014 survey. 1,3 Additionally, previously reported clinical outcomes including low response rates (4%-10%), poor overall survival (OS, 5-6 months), and modest disease control rates (DCR, 20%-25%) have brought into question the therapeutic appeal and ethical justification of phase I trial enrolment. [4][5][6] Nevertheless, the landscape of early phase oncology trials is changing. A meta-analysis of phase I trials conducted between 2014 and 2015 demonstrated encouraging response rates of 20%. 7 Trials that used an enrichment design (specific tumor type or biomarker driven), explored drug combinations, or had an expansion cohort were associated with even higher response rates. 2,7 More recently, owing to improvements in genomics and growing emphasis on precisionbased medicine, master protocols in the form of basket and umbrella trials have been increasingly employed to study targeted agents in cancer research. Basket trials are clinical studies investigating agent(s) targeting a common predictive risk factor (commonly a biomarker) across various tumor types, whereas umbrella trials test multiple targeted interventions in a single disease, which has been stratified into various subgroups based on different biomarkers or molecular signatures. 8 The American Society of Clinical Oncology recently released a position statement on phase I trials, reiterating that, while remaining an integral part of clinical cancer research, these trials do indeed have therapeutic intent. 9 Further reinforcing the importance of early phase trials, the US Food and Drug Administration in 2012 announced the "Breakthrough therapy designation for experimental drugs" to expedite the development of promising drugs based on preliminary clinical evidence. 10,11 Notable examples of drugs to benefit from this pathway are the programmed death receptor (PD-1) targeting antibody pembrolizumab in melanoma, and the small molecule tyrosine kinase inhibitor ceritinib in non-small lung cancer possessing the anaplastic lymphoma kinase gene rearrangement. 12,13 Both drugs went on to be granted accelerated approvals for their respective indications in 2014, less than 5 years after the first patient was enrolled in the corresponding phase I trial. [14][15][16] While the expedited approval pathways do not apply to the majority of agents investigated in phase I trials, these examples illustrate that well-designed phase I trials have the potential to streamline drug development and ultimately allow for earlier patient access to effective therapies.
Much of the published literature reporting on the trends and outcomes of phase I trials have taken place in the era of cytotoxic agents. Few reviews have included molecular targeted agents (MTAs) and immuno-oncology (IO) agents, with even fewer addressing combination trials, thus failing to shed light on the most recent trends. Additionally, large systematic reviews of early phase trials rely on published results of trials and are therefore inherently prone to publication bias. The rate of unpublished trials is reported to be as high as 30% 17 and this gap in the results could skew the overall interpretation of phase I trial outcomes. To better understand the evolving landscape of early phase drug development, we undertook a retrospective review of all phase I oncology trials enrolling patients over a 6-year period between 2013 and 2018 at a single tertiary Australian center.
We report on patient demographics, trial characteristics, safety, and treatment outcomes. were recorded. Toxicity data were determined for the population of patients that received at least one dose of study drug. Chi-square (χ 2 ) testing was performed to detect any differences between ORR based on trial type and trial category. OS was defined as the time from consent to death from any cause. Kaplan-Meier estimates of survival were calculated separately for patients grouped by trial type and referral type (early vs late). Ninety-day mortality (90DM) rates were calculated from the date of trial enrolment for the entire cohort. Statistical analysis was performed using IBM SPSS Statistics for Windows, Version 25.0. Armonk, New York.

| RESULTS
Twenty-eight phase I trials in solid tumors were conducted at our center over the 6-year study period and 242 patients were screened ( Figure 1). Thirteen patients were screened for more than one trial (including two patients who each screened for three different trials), yielding a total of 257 screening visits ( Figure 1). Of these visits, there were 45 incidents of screen failure (18%). The most common reasons for screen failure were abnormal laboratory values out of the required range for eligibility (n = 14%, 31%) and deterioration in performance status prior to dosing despite fulfilling performance status criteria at screening (n = 8%, 18%). Other causes of ineligibility were secondary to protocol-defined exclusions including comorbid illness (n = 3%, 7%), concurrent second malignancy (n = 3%, 7%), brain metastases (n = 2%, 4%), absence of measurable disease (n = 2%, 4%), absence of requisite biomarker(s) (n = 1%, 2%), prohibited concomitant medications (n = 1%, 2%), and prolonged corrected QT interval on baseline electrocardiogram (n = 1%, 2%). Three patients were enrolled but did not commence treatment. Therefore, the toxicity-evaluable safety cohort of subjects who received a minimum of one dose of study drug consisted of 209 patients (86%). The response-evaluable cohort consisted of 190 (79%) patients who had at least one response assessment.

| Trial characteristics and recruitment
Of the 21 trials, eight (29%) were first-in-human (FIH). Most studies were histology-agnostic while four were specific to tumor type (mesothelioma, small cell lung cancer, and two prostate cancer trials). Of the 28 trials, only 1 (4%) was investigator-initiated, with the remaining 27 being industry-sponsored. Four (14%) trials required the presence of a tissue-based biomarker for study eligibility, which was confirmed by central laboratory assessment during a "prescreening" process; these included a BRAF V600E mutation (1) Table 2.

| Responses and survival
Of all patients (n = 209) who received at least one dose of trialspecified treatment, 190 (91%) had a disease response assessment.
Nineteen (9%) patients came off trial prior to the first scheduled response assessment scan due to the following reasons-early clinical progression (n = 10), cancer-related death (n = 4), toxicity (n = 4) and unrelated medical illness (n = 1). Forty-one (22%) patients had a confirmed response as defined by RECIST v1.

| Toxicity
Clinically significant grade 2 and all ≥ grade 3 TRAEs and the corresponding trials by drug class are detailed in Table 3. Nineteen  and OS are often secondary endpoints due to relatively small numbers of patients recruited to early phase trials. 19 Rates of response have historically been modest, which in turn has fueled the major criticism of phase I oncology trials-a debatable risk-benefit ratio for patients enrolled. 18 In our study, ORR was 22% and DCR was 53%, independent of drug class, comparable to that of recently published data. 7 Also consistent with the trend in recent reviews, 7,16,33 we observed only a small number of cytotoxic drug trials and a predominance of IO and combination trials, most notably in P2. Potential reasons proposed for the improving anti-tumor activity seen in phase I trials have included the presence of expansion cohorts, biomarker-driven trials, growing numbers of combination studies as well as more effective therapies. 2,7 One or more of these factors are applicable to most (80%) of the trials, we have conducted during this six-year period and therefore could explain some of our findings. The median OS of our entire cohort was 8.0 months, comparable to previously reported survival on phase I trials of 8-10 months. 19,34,35 It is interesting to note that although the median OS between IO and non-IO trials was the same, there was a late separation of the curves, which may be driven by the durability of responses that are commonly associated with IO therapies.
The growing success of phase I trials has encouraged referral for earlier participation as a therapeutic option as opposed to a last resort; the early referral rate at our institution may reflect this trend, where almost half of all patients (47%) were referred either untreated for advanced disease or after only one line of systemic therapy. We can speculate that trials investigating IO and MTAs were attractive to referrers, and such studies were already starting to feature by 2013, when our study period commenced. Additionally, phase I trials in our unit provided an opportunity for patients to access anti-PD-1/PD-L1 drugs in the absence of drug approval and government reimbursement, likely contributing to earlier referral patterns. We found that patients referred early also had an improved OS compared with those referred later. Although OS is typically longer in earlier lines of therapy for approved agents or combinations in many tumor types, the longer OS seen in the phase I setting from our cohort is potentially a reflection that agents from drug classes with proven activity were being employed.
A screen failure rate of 18% compared favorably to the previously reported rate of 25% in phase I trials. 36 The leading causes of screen failure at our center were similar to those in the published literature, namely, out-of-range laboratory values and the deterioration of health prior to dosing. Although screen failures are inevitable, the relatively low rates we observed may have been in part due to the proportion of early referrals when patients are typically more robust and retain a better performance status, as well as appropriate patient selection prior to the screening process.
The issue of risk and potential harm associated with phase I trials in oncology has long been debated. 18,22,25 Our study revealed relatively low rates of high grade TRAEs and only one treatment-related death. These findings demonstrate the relative safety of phase I trial enrolment. The incidence of irAEs in the IO trials was low with no study. These master protocols have lately emerged as critical tools in investigating targeted therapies and data pertaining to their influence on early phase clinical research would ideally feature in a study of this kind. Nevertheless, the major strength of our study is its real-world representation of individual patient data. There is certainly a recognized need to share and access patient-level phase I trial data in order to optimize trial design, identify important safety issues and ultimately improve patient care. 37 Previous systematic reviews of trends in phase I oncology trials have been criticized due to inherent publication bias as they drew results from PubMed searches. Consequently, the response rates reported could possibly be an overestimate of the true result. A future registry-based database would be of great value to monitor trends and outcomes in the dynamic field of early drug development.
In conclusion, our study adds to the growing body of evidence supporting phase I oncology trials as valid treatment options. It highlights the complexities surrounding design, endpoints, biomarker use, and clinical outcome reporting. Notably, there is a paucity of such data in an Australian context and hence the findings of this study are unique and valuable when considering the evolving phase I trial landscape in oncology. The 90DM rate of 20% in a good performance status group highlights the poor prognosis for most patients with advanced solid organ cancer and hence it is incumbent on clinicians to exercise caution while conducting early phase trials by carefully consenting patients and offering reasonable expectations based on preclinical and clinical evidence. Finally, as next generation sequencing and other forms of biomarker identification become more prevalent, the role of optimal patient selection when conducting early phase oncology trials will become increasingly relevant.