Patients
A total of 145 patients were included in the study. The eligibility criteria were as follows: adult patients with an incurable (i.e., surgically unresectable or metastatic disease) and lethal (i.e., ≥ 50% 2-year cancer-associated mortality) cancer; patients with cancer of unknown primary or a rare tumor with no approved therapies; patients with at least one of the following: unresectable disease, metastatic disease, medically unfit for surgical resection but with an expected survival of > 3 months, disease where no conventional therapy leads to a survival benefit > 6 months, actionable alterations determined by FoundationOne; no prior systemic cancer treatment; no prior anti-tumor agents; ability to understand and the willingness to sign a written informed consent; Eastern Cooperative Oncology Group Performance Status of 0 to 1; measurable disease; New York Heart Association Functional Classification I-II; adequate organ function; able to swallow and retain oral medication; must have evaluable tissue/blood for testing; and negative serum pregnancy test and use of one form of pregnancy prevention. Exclusion criteria were as follows: two oncologists disagree on prognosis or resectability; medical disorder that would confound study analyses; and pregnant, breast-feeding, or not using pregnancy prevention. All participants provided written informed consent to the Investigation of Profile-Related Evidence Determining Individualized Cancer Therapy (I-PREDICT) study (groups 1 and 2), as well as separate written informed consent for any investigational drug trials to which they were navigated, per Internal Review Board approval guidelines. The trial opened to enrollment on February 13, 2015. The data cutoff date was November 1, 2019. All data for each patient are included in Additional file 1: Table S1.
Study design and treatment
I-PREDICT (NCT02534675, https://clinicaltrials.gov/ct2/show/NCT02534675) is a cross-institutional, prospective navigation trial. The study design and outcomes for patients with previously treated, unresectable or metastatic cancers (group 3) have been reported [5]. Herein, we utilized the same study protocol (Additional file 2) to investigate the feasibility, efficacy, and safety of administering customized, molecularly matched combination therapies to patients with treatment-naïve unresectable (group 1) or metastatic (group 2) lethal cancers with an expected 2-year survival of less than 50%.
Genomic profiling of tumor tissues (236–405 genes) or blood-derived circulating tumor DNA (ctDNA) (62 genes) was conducted by hybrid capture-based next generation sequencing (NGS). In a subset of patients had PD-L1 immunohistochemistry (IHC) [antibody SP142 (Ventana) or 22C3 (Dako)], tumor mutational burden (TMB) and microsatellite status were also assessed, using previously described methods (Foundation Medicine, Inc.; CLIA-licensed and CAP-accredited laboratory; Cambridge, MA. https://www.foundationmedicine.com) [6,7,8,9,10,11,12,13]. TMB results were reported as follows: TMB-High corresponds to ≥ 20 Muts/Mb, TMB-Intermediate corresponds to 6-19 Muts/Mb, and TMB-Low corresponds to ≤ 5 Muts/Mb.
Treatment recommendations and potential overlapping drug toxicities were discussed by a molecular tumor board (MTB; either ad hoc just-in-time electronic exchange or weekly face-to-face meetings). The ad hoc just-in-time electronic MTB always included the co-authors (namely, Jason Sicklick, Shumei Kato, Ryosuke Okamura, Pradip De, Casey Williams, Brian Leyland-Jones, Razelle Kurzrock) and the treating physician. The in-person weekly MTB also included other oncologists (medical, surgical, gynecologic and radiation), pharmacologists, cancer biologists, geneticists, radiologists, pathologists, basic scientists, and bioinformaticians, as well study coordinators/navigators and medication acquisition specialists as we previously described [14,15,16,17]. These individuals varied from week to week. Unique to I-PREDICT, there was not a predesignated set of drugs and drug combinations determined by the MTB. Instead, all possible drug combinations were used, unless the MTB felt that there was reason to believe or reported/known evidence that a combination would be toxic and therefore contraindicated. The administration of customized, molecularly matched combinations (including targeted, chemotherapeutic, hormonal agents, biologic agents, and immunotherapies) was emphasized. The treating oncologist rendered the final decision regarding therapy choice. Thus, this study was uniquely patient centered since the combination of drugs was determined by the molecular alterations present in the patient’s tumor, rather than by having a limited set of preconceived options.
This study was cross-institutional [two centers: University of California San Diego (UCSD) Moores Cancer Center and Avera Cancer Institute in South Dakota] with all physicians at each site able to consent and enroll patients. Investigators updated the study information by teleconference (twice monthly) and via face-to-face meetings for UCSD study members. Data review retreats were held every 1–2 months.
Matching score
We compared differences in outcomes according to a previously reported molecular Matching Score (MS) [5, 18,19,20]. In short, Matching Score is defined as the number of alterations (excluding variants of unknown significance, VUS) targeted by administered drugs divided by the total number of characterized alterations (excluding VUS) identified. We did not distinguish between potential driver versus passenger mutations. The higher the score (range, 0–100%), the “better” the match.
For instance, if a patient’s tumor harbored six characterized genomic alterations and they were given two agents that targeted three of these alterations, the score would be calculated as 50% (3 of 6). Investigators that calculated the scores (JKS, SK, RK) were blinded to patient outcomes at the time of calculations. If a patient had ≥ 2 genomic reports, the abnormalities in each test report were counted, because there can be heterogeneity between two tissue biopsies or between blood and tissue samples.
Other considerations were also relevant: (i) if a participant had two or more genomic aberrations that were in the same gene and potentially had the same signal/pathway impact, these aberrations were counted as one; (ii) two aberrations in the same gene that potentially had different oncogenic impacts or were structurally distinct (e.g., amplification and mutation) were counted as two since they have different or additive effects; (iii) if two drugs simultaneously targeted the same aberration in a well-established synergistic manner (e.g., the FDA-approved combinations of dabrafenib and trametinib for BRAF aberrations, or pertuzumab and trastuzumab for ERBB2 aberrations), the impact was counted twice in both the numerator and denominator; and (iv) estrogen receptor-positive or androgen receptor-positive expression by IHC targeted by a hormone modulator (such as, letrozole) was also counted as one in both the numerator and denominator.
For small molecule inhibitors, matching was based on preclinical low inhibitory concentration 50% (IC50) of the drug for the target (generally less than 100 nM) or for signal transduction effectors immediately downstream of the aberrant gene product. Antibodies were considered matched if their primary target was the product of the molecular alterations. Patients whose cancers harbored a BRCA-related gene anomaly were considered matched if they received platinum agents or PARP inhibitors. Also, if a participant was given checkpoint blockade immunotherapy, the score was assigned as 100% for results of microsatellite instability-high (MSI-High) or high tumor mutational burden (TMB) or high positive programmed death ligand 1 (PD-L1) expression (≥ 30%) on IHC; the score was 50% for results of low positive PD-L1 on IHC or TMB-intermediate. Individuals in the immunotherapy-treated group who had, as an example, TMB-intermediate and were scored at 50%, and received matched targeted agents, had the total score calculated as 50% + (X% ÷2) [note: X = (number of alterations targeted by agents administered)/(total number of alterations)]. For example, if a malignancy had intermediate TMB and the participant was given checkpoint blockade, but also had an PIK3CA and a FGFR alteration and received an FGFR inhibitor in addition to checkpoint blockade, the score was 50% + 25% = 75%. We also considered TP53 abnormalities as matched to drugs with anti-VEGF/VEGFR activity because several reports have demonstrated that TP53 alterations are correlated with VEGFA up-regulation and that VEGFA inhibitory therapies associated with better treatment outcomes in patients with TP53-mutant cancers [21,22,23,24,25,26]. No match was scored at over 100%. More details on Matching Score calculations are in our previous report [5].
We stratified patients according to Matching Scores ≥ 60% versus 1–59% versus unmatched (0%). The matching score cut-off point of 60% was determined by using ROC-AUC of 0.733 for the Disease Control Rate (DCR = SD ≥ 6 months + PR + CR rate). We determined that the optimal DCR cutoff (for maximal rate in high matching group) was 17 of 25 pts (68%) versus 13 of 43 pts (30%) (P = 0.005). A second stratification of Matching Score > 50% versus ≤ 50% was also analyzed to recapitulate the assessment method used in our previously published study in patients with pretreated metastatic disease (5). Further gradations in Matching Score (0%, 1–39%, 40–59%, 60–99%, and 100%) were also used for some analyses. Median PFS data for the correlation analysis with these Matching Score grades were calculated by drawing the Kaplan-Meier curve.
Primary endpoint and study objective
The primary study objective was to determine the feasibility of using molecular testing to determine therapy for patients with previously treated cancers with incurable biology (≥ 50% 2- year cancer-associated mortality). The primary endpoint was the proportion of patients who receive molecularly targeted matched treatment after recommendations based on genomic analysis.
Secondary and exploratory endpoints
Secondary and exploratory endpoints included the following: proportion of patients with actionable genomic alterations, incidence of high-grade adverse events, disease control rate [DCR; stable disease (SD) ≥ 6 months or partial/complete response (PR/CR)], progression-free survival (PFS), and overall survival (OS). Treatment response was evaluated using Response Evaluation Criteria in Solid Tumors (RECIST), version 1.1 [27]. Tumor assessments were performed by means of computed tomography and magnetic resonance imaging, at baseline and about every eight weeks thereafter. All RECIST measurements were performed by independent central radiology review. DCRs between groups were compared using Fisher’s exact test. With the sample size of 75 evaluable patients and the accrual ratio of 1:2 (high Matching Score group [≥ 60%]: low Matching Score group [< 60%]), we would have > 80% power for detecting the estimated DCR of 65% in the high Matching Score group versus 30% in the low Matching Score with 0.05 type I error. This was calculated a priori by J.J.L. (biostatistician). The PFS and OS were calculated from the date of treatment initiation to disease progression, or any cause of death, respectively. Patients who were progression-free (for PFS) or alive (for OS) at the date of last analysis were censored at that date. The investigators calculated Matching Scores blinded to patients’ outcomes. Serious adverse events (SAEs) were graded according to the Common Terminology Criteria for Adverse Events, version 4.03 [28]. Given that toxicity and drug dosages are major concerns when administering de novo combinations, we elected to re-visit the data after the initial data cutoff of 11/01/2019. Thus, for the SAEs and dose adjustments, we re-analyzed the data thru 03/24/2021.
Dosing drug combinations
The UCSD Moores Cancer Center and Avera Cancer Institute Data Safety Monitoring Committees monitored the safety of this study at the respective institutions. To minimize toxicities due to customized de novo combinations, dosing of each combination was discussed by the MTB as we previously reported [5]. Recommended dosing was based on the safety data gleaned from previous studies and PharmD recommendations [29,30,31,32]. For the most part, two-drug de novo combinations were started at ~ 50% of the usual dose of each drug. Three-drug de novo combinations were started at ~ 33% of the usual dose of each drug. Patients were fully informed of risks when treated with regimens lacking phase 1 data. Based upon tolerability, patients received escalating doses of drugs with regular monitoring by their treating physician. Additional file 3: Fig. S1 demonstrates the intra-patient dose adjustments, last drug dose (as a percentage of the standard drug dose for each agent), and the median percent of standard dose for each drug. No two patients had the same drug combinations and therefore, there is no toxicity correlation with any specific combination.
Statistical analysis
All data was compiled in a Microsoft Access 2013 (version 15.0) database. In the nature of a hypothesis-driven trial, we performed sample size calculations and originally planned to enroll 75 evaluable treated patients and estimated that 40% (N = 30) of the patients would be assigned to the matched group (Arm A) versus 60% (N = 45) to the unmatched group (Arm B). However, as precision medicine has become more established, the rate of matching has increased. We ultimately enrolled 76 evaluable treated patients (of 145 treatment-naïve patients consented); 54 (71% of evaluable patients; 37% of enrolled patients) were administered > 1 drug matched to their molecular profile (Additional file 3: Fig. S2; all molecular, treatment, Matching Score, and outcomes data for each patient are included in Additional file 1: Table S1). Logistic regression was performed for binary endpoints to estimate the odds ratio (OR). Variables with P < 0.15 in the univariate analyses were entered into the multivariate analyses. The Kaplan–Meier method was used for PFS and OS analyses. Survival comparisons were made by the log-rank test. Cox regression models were used to estimate the hazard ratio (HR) in multivariable analysis. Statistical analyses were performed using SPSS version 24.0.