Purpose The aim of this analysis was to investigate the potential for ulixertinib (BVD-523) to prolong cardiac repolariza- tion. The mean prolongation of the corrected QT (QTc) interval was predicted at the mean maximum drug concentrations of the recommended phase 2 dose (RP2D; 600 mg BID) and of higher concentrations. In addition, the effect of ulixertinib on other quantitative ECG parameters was assessed. Methods In a two-part, phase 1, open-label study in adults with advanced solid tumors, 105 patients [24 in Part 1 (dose escalation) and 81 in Part 2 (cohort expansion)] were included in a QT prolongation analysis. Electrocardiograms (ECGs) extracted from 12-lead Holter monitors, along with time-matched pharmacokinetic blood samples, were collected over 12 h on cycle 1 day 1 and cycle 1 day 15 and analyzed by a core ECG laboratory. Results A small increase in heart rate was observed on both study days (up to 5.6 bpm on day 1 and up to 7 bpm on day 15). The estimated mean changes from baseline in the study-specific QTc interval (QTcSS), at the ulixertinib C , were max − 0.529 ms (90% CI − 6.621, 5.562) on day 1 and − 9.202 ms (90% CI − 22.505, 4.101) on day 15. The concentration: QTc regression slopes were mildly positive but not statistically significant [0.53 (90% CI − 1.343, 2.412) and 1.16 (90% CI − 1.732, 4.042) ms per µg/mL for days 1 and 15, respectively]. Ulixertinib had no meaningful effect on PR or QRS intervals. Conclusions Ulixertinib administered to patients with solid tumors at clinically relevant doses has a low risk for QT/QTc prolongation or any other effects on ECG parameters. Registration The study is registered at Clinicaltrials.gov (NCT01781429) and was sponsored by BioMed Valley Discoveries. Keywords Cardiac safety · ECG · Exposure:response modeling · Holter · Oncology · QT · QTc Introduction * Boaz Mendzelevski Ulixertinib is a potent and selective small molecule inhibi- Boaz.Mendzelevski@CardiacSafetyConsultants.com tor of the extracellular signal-regulated kinases ERK1 and Cardiac Safety Consultants Ltd, 4 Hallswelle Road, ERK2. Ulixertinib inhibits growth and survival of cancer London NW11 0DJ, UK cells in cultured cell lines, including melanoma, colorec- Statistik Georg Ferber GmbH, Riehen, Switzerland tal, and pancreatic cell lines harboring BRAF or RAS muta- The University of Texas MD Anderson Cancer Center, tions, as well as in animal models. Tumor response was Houston, TX, USA assessed in 101 patients treated with ≥ 600 mg twice daily Memorial Sloan Kettering Cancer Center, New York, NY, (BID) ulixertinib, of whom 14 had a partial response per USA Response Evaluation Criteria in Solid Tumors (RECIST Massachusetts General Hospital Cancer Center, Harvard v1.1) criteria . While ulixertinib modestly inhibited Medical School, Boston, MA, USA (IC , 3.4 µM) the human ether-á-go-go-related gene, it BioMed Valley Discoveries Inc., Kansas City, MO, USA did not significantly prolong the cardiac action potentials recorded from dog Purkinje fibers at concentrations of Shanghai Hengrui Pharmaceutical Co., Ltd, Shanghai, China up to 10 µg/mL. In animal studies, no significant cardio- Bioclinica Inc, Princeton, NJ, USA vascular findings were observed with acute (single) oral Stanford University, Stanford, CA, USA Vol.:(0123456789) 1 3 1130 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 dosing of ulixertinib of up to 50 mg/kg in dogs [maximum interval and patients treated with multiple medications that observed concentration (C ) = 17.3 µM]. In addition, can introduce challenges for QT assessments and poten- max ulixertinib is highly protein bound in multiple species, tially lead to false positive outcomes. including human (99.9–100%). Thus, pre-clinical data suggested that ulixertinib would have a favorable cardiac safety profile and low potential for inducing QT/QTc pro- longation in patients. Materials and methods Overall improved efficacy of novel cancer therapies is leading to higher survival rates and a larger population of Study design and objectives cancer patient survivorship , emphasizing the need for improving drug safety and reducing systemic and organ- The clinical study (Clinicaltrials.gov identifier, specific toxicities of new agents. Commensurate with this NCT01781429) was a first-in-human, two-part, open- progress, recognizing and managing cardiovascular toxicity label, multicenter phase 1 study designed to assess the of cancer therapies are a clinical and regulatory focus. A key safety, PK, and PD of escalating doses of ulixertinib in safety concern is the potential of drugs, particularly small patients with advanced malignancies. The study was molecule new chemical entities, to prolong the electrocardio- composed of two parts: a dose-escalation phase (Part 1) graphic (ECG) heart rate corrected QT interval (QTc) and to and a cohort-expansion phase (Part 2). Part 1 established potentially cause the lethal cardiac arrhythmia Torsades de dose-limiting toxicity (DLT), the maximum tolerated dose Pointes . After a series of high profile drug withdrawals (MTD), and the preliminary recommended phase 2 dose and non-approvals in the 1990s and early 2000s, the Inter- (RP2D). Part 1 used an accelerated single-patient cohort national Conference on Harmonisation (ICH) adopted the design, followed by a standard 3 + 3 design, informed by ICH-E14 guidance in May 2005 , calling for a methodi- the accrued safety experience throughout the study. Intra- cal assessment of the potential pro-arrhythmia risk of new patient dose escalation for patients entering the study at drugs early in clinical development, typically by conducting dose levels lower than the RP2D was allowed under spe- a dedicated thorough QT (TQT) study . cific circumstances. The MTD was defined as the high- A key limitation of the ICH-E14 mandated TQT study est dose cohort at which ≤ 33% of patients experienced is that a TQT study is scientifically more robust in healthy ulixertinib-related DLTs in the first 21 days of treatment. volunteers than in patients with serious illnesses, such as The RP2D was defined as the MTD and was additionally cancer patients receiving toxic anticancer drugs . For this informed by observations related to PK, PD, and cumula- reason, and concurrent with recent advancements in the car- tive toxicity observed after multiple cycles. The RP2D diac safety assessment paradigm, which favor QT assess- was determined to be 600-mg BID during Part 1. In Part 2, ments in routine, early phase drug development clinical tri- patients were initially treated at the preliminary RP2D. als , oncology cardiac safety (QT) assessments are often Patients received oral doses of ulixertinib BID in 21-day performed in early phase oncology studies with relatively treatment cycles until disease progression, unacceptable small patient cohorts. This practice is further supported by toxicity, or another withdrawal criterion was met. Treat- the recent update to the ICH E14 guideline, which supports ment cycles were intended to be administered consecu- the use of exposure–response (ER) analysis as an alternative tively without interruption; however, dosing interruptions to the by-time-point analysis as the primary basis for cardiac and/or dose reductions were allowed when necessary to safety regulatory decisions . manage toxicities. Here, we report on the potential for QT/QTc prolon- ECG data were collected continuously using high-fidelity gation effects from a phase 1 oncology study designed 12-lead Holter recorders at cycle 1 day 1 and cycle 1 day 15 to assess the safety, pharmacokinetic (PK), pharmacody- for 12 ± 2 h during PK sampling. Standard 12-lead ECGs namics (PD), and efficacy of ulixertinib in patients with were extracted in triplicates at timepoints corresponding advanced solid tumors. The present analysis is one of the with PK sampling. ECG extraction timepoints were sched- first to use an advanced Exposure:Response Modeling uled for cycle 1 day 1 and cycle 1 day 15 at the following (ERM) approach, in keeping with recent clinical and regu- timepoints: 0 h (pre-dose) and 0.5, 1 (± 5 min), 2, 4, 6, 8 latory progress in the field. The current ERM strategy is (± 10 min), and 12 h (± 2 h) post-dose. In Part 1 (dose esca- geared toward early phase cardiac safety assessments in lation), ECG data collection was performed for all patients; small clinical trials, because it uses the full ECG data set while in Part 2 (cohort expansion), ECG monitoring was in a comprehensive manner. This is especially relevant stopped after a suc ffi ient number of patients with valid ECG for oncology cardiac safety studies, which typically lack data were enrolled, based on power calculations using data placebo control, and which often involve small patient from Part 1. The current analyses were based on all patients cohorts with a large variability in heart rate and QT/QTc included in Parts 1 and 2 of the study with valid ECG data. 1 3 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 1131 their preceding R–R interval were annotated and measured. Patient population Measurements were performed from lead V3 (primary lead), with the aim of obtaining data from the same lead for each Enrolled patients were men and women ≥ 18 years of age with histologically confirmed metastatic or advanced- patient wherever possible. ECG parameters for any patient and timepoint were considered valid if they were based on stage malignant solid tumors for which no curative therapy existed. Patients had to have an Eastern Cooperative Oncol- at least two readable ECGs of the triplicate. For QTc determination, two common QT correction ogy Group (ECOG) performance status of 0 or 1 and a life expectancy of ≥ 3 months. Other eligibility requirements (QTc) methods were used. QTcF was calculated using 1/3 the Fridericia formula (QTcF = QT × RR ). Correction included adequate cardiac, renal, hepatic, and bone marrow function. Adequate cardiac function was defined as a left was performed for each replicate, and the median across replicates was taken for each timepoint. QTcSS, a study- ventricular ejection fraction (LVEF) of > 50% [assessed by multi-gated acquisition (MUGA) or echocardiography] and specific QTc, was derived from individual ECGs obtained at the baseline (drug-free) period by fitting a linear mixed- QTc < 470 ms. For Part 2 only, patients with BRAF, NRAS, or MEK mutations who had measurable disease by RECIST effects model with logQT as the dependent variable, logRR as the covariate, and a random intercept per patient were enrolled. −β Standard exclusion criteria were applied, including gas- (QTcSS = QT × RR , where β was the regression coef- ficient of logRR ). The estimates of the model parameters trointestinal conditions that could impair absorption of the study drug, and uncontrolled or severe intercurrent or were tabulated with two-sided 95% confidence intervals (CIs). In addition, for the pre-dose values, a data set with chronic medical conditions. Patients could not take any can- cer-directed therapy (e.g., chemotherapy, hormonal therapy, individual QT, QTcF, QTcSS, and R–R values for each rep- licate was used to assess the appropriateness of the correc- biologic therapy, or immunotherapy) within 28 days or 5 half-lives (whichever was shorter) before the first dose of tion method. QTcF was to be used as the primary correction method if the two-sided 95% CI for β included 1/3, the expo- ulixertinib. A minimum of 10 days was required between termination of any investigational drug and administration nent for the Fridericia correction; otherwise, QTcSS was to be considered primary. The appropriateness of the correction of ulixertinib; and any drug-related toxicity, except alo- pecia, had to have recovered to grade 1 or less. Concur- methods was investigated as outlined by Tornøe et al. . rent therapy with any other investigational agent or drugs known to be strong inhibitors of cytochrome P450 (CYP) ECG analysis sets and validity criteria enzymes CYP1A2, CYP2D6, or CYP3A4 or strong inducers of CYP3A4 was prohibited. ECG parameters for any patient and timepoint were consid- ered valid if they were based on at least two valid replicate All patients gave written informed consent before the start of the pre-study examination. The study was conducted ECGs. In addition, two ECG analysis sets were defined as follows: (1) an ECG set included all patients in the safety according to the protocol and in compliance with ICH Good Clinical Practice guidelines. population who had a valid baseline QTc value and at least one valid post-baseline QTc value. Patients for whom no ECG assessments baseline ECG could be extracted before the PK blood draw for time zero were included if three replicate ECGs could be Digital 12-lead ECGs were recorded continuously using a extracted in the time window between the first PK draw (i.e., baseline) and the first drug administration. Visit 4 (cycle 12-lead digital Holter recorder (M12R, Global Instrumenta- tion LLC, Manlius, NY, USA) equipped with a removable 1 day 15) data were included only if this visit actually took place within 21 days after day 1 (i.e., the day of first drug standard secure digital memory card. The recorder transmit- ted ECG data continuously, via a Bluetooth connection, to a administration). (2) An extended ECG set consisted of all patients in the ECG set and allowed for patients without a nearby laptop computer that transmitted the data to the ECG core laboratory (Bioclinica Inc, Princeton, NJ, USA) using valid pre-treatment baseline ECG to be included if three replicate ECGs could be extracted in the first 15 min after a high-speed internet connection. Standard 12-lead ECGs were extracted automatically first drug administration. For day 15, this analysis set also included patients irrespective of the actual day the ECG from the continuous 12-lead Holter recordings in triplicates at predefined timepoints and analyzed programmatically assessments were made, provided the patient received the study drug on that day. In addition, an Exposure:Response using an automated algorithm (M12A Enterprise Holter Sys- tem, Global Instrumentation LLC). All ECGs were manually (ER) analysis was the intersection of the ECG set and the PK population, as defined in the overall study statistical analy - adjudicated by a board certified cardiologist, and all ECGs of a given patient were read by the same cardiologist. For sis plan. The ER set was the set of primary interest in this analysis and was used except where specified. The extended each 12-lead ECG, three consecutive PQRST complexes and 1 3 1132 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 ER set was the intersection of the extended ECG set and the a random intercept and slope per patient . In addition, PK population. This analysis set was used for robustness a binary factor indicating what dosing regimen the patient analyses. was on was included. For the purpose of this analysis, it was defined prospectively that patients who received an average Definition of baseline dose of 300-mg BID or below of ulixertinib were assigned to the low-dose group, and all other patients were included Baseline was defined as the timepoint of pre-dose assess- in the high-dose group. In the absence of a placebo group, ments on cycle 1 day 1 for each study part. The baseline this factor was introduced as a surrogate for placebo. The ECG value was the mean of the triplicate pre-dose ECG val- degrees of freedom for the model estimates were determined ues on cycle 1 day 1. For analyses based on the extended ER by the Kenward–Roger method . From the model, the set, the ECG parameters extracted in the first 15 min after slope (i.e., the regression parameter for the concentration) dosing were used as replacement for missing values. Post- was estimated together with the two-sided 90% CIs and other baseline ECGs were all ECGs obtained after the defined model parameters. Since time is included as a factor in this baseline timepoint. model, it allows for the separation of variability attributed to the concentration of the drug from spontaneous diurnal QTc analysis endpoints variability; in other words, the estimated time effect can be interpreted as the estimate of diurnal changes independent The analyses were defined in a prospective statistical analy - of drug concentrations and vice versa. sis plan. The primary QTc variable was designated as QTcp. The geometric mean (across patients) C on days 1 and max The primary endpoint was the change in QTcp from pre-dose 15 (gmC 1 and gmC 15, respectively) of the 600-mg max max baseline (ΔQTcp). Both QTcF and QTcSS were assessed. dose was used to estimate the effect of ulixertinib, along Secondary endpoints included the baseline-adjusted effect with the two-sided 90% CI. The effect estimate was based on heart rate and PR and QRS intervals and the frequency on the respective primary model as the contrast between of ECG parameters exceeding a set of given limits. Analy- patients with the concentration of interest in the high-dose ses were performed on the ECG set. Patients for whom no group minus a patient with concentration zero in the low- baseline could be extracted before the PK draw at time zero dose group. For the computation of the CIs, the random were included if three replicate ECGs were extracted in the nature of C was ignored. Supplementary predictions were max time window between the first PK draw (i.e., baseline) and performed at 1.5 times the concentration of gmC 1 and max the first drug administration. Day 15 data were only included gmC 15. max if the day 15 visit actually took place no later than 21 days Since a joint graphical display of the original ΔQTc after start of treatment. and concentration data and the estimated regression line would be difficult, because the influence of time cannot be PK parameters depicted in a two-dimensional display, we decided to present the regression line for the predicted effect together with the For each patient and post-dose QTc value, a correspond- raw ΔQTc data over the concentration. This meant that the ing plasma level was calculated by interpolation from the regression lines would not necessarily follow the trend seen available values. Linear interpolation was used before T in the scatter plots. The discrepancy between the two is the max (the time to reach C ), and log-linear interpolation was effect attributed to time and not to drug concentration. max used after T . For times between 0.5- and 6-h post-dose, max individual values were excluded if the time between ECG and blood sample collection exceeded 10 min. For later Assessment of appropriateness of the primary timepoints, this window was increased to 2 h, and extrapo- model lation beyond the last PK measurement was allowed for up to 20 min. For the above model to be valid, the key assumptions were the absence of hysteresis and the linearity of the concen- ER analysis tration-QTc relationship. The former was investigated by a comparison of the time courses of double differences of QTc The primary analysis was based on the primary QTc vari- (ΔΔQTc) and concentration (ΔΔC) (i.e., the difference in able and was performed independently for days 1 and 15 mean QTc and mean concentration between the high-dose with patients from Parts 1 and 2 pooled. It was based on a group and the low-dose group). Since low-dose ulixertinib linear mixed-effects model with ΔQTcp as the dependent was administered only in Part 1 of the study, this investiga- variable, interpolated drug plasma concentration as a con- tion was done in Part 1 only. Briefly, hysteresis was consid- tinuous covariate, time and part as categorical factors, and ered present only if (1) there was a prolongation of more 1 3 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 1133 than 5 ms in ΔΔQTc and (2) a delay in the occurrence of the nonsignificant; abnormal ECG, possibly significant; or largest prolongation compared with T was > 1 h. unreadable ECG. Any new treatment-emergent repolariza- max Linearity was tested independently in Parts 1 and 2 by tion abnormalities (i.e., ST segment abnormalities), T-wave adding a quadratic term to the primary linear model and test- morphology, and AV conduction abnormalities were ing this term. If there was an indication that a linear model summarized. was inappropriate in one of the analyses of Parts 1 or 2, alternative non-linear models were to be investigated, and Combined analysis the primary analysis was then to be repeated for the model found to best accommodate the non-linearity detected based Analyses were performed separately for Parts 1 and 2 at on the Akaike information criterion. the end of each study segment. In this report, we present a joint analysis of the combined Parts 1 and 2 data set, using Robustness analyses the methods described above. In addition, although the data reported here include all dose levels, the focus of this report To assess robustness of the primary analysis, variants of is on the ulixertinib RP2D of 600-mg BID. This is not only the analysis were performed. A joint model was fitted on because this dose will be the recommended therapeutic dose, data from both day 1 and day 15. This model was simi- but also because this dose group has the largest amount of lar to those used in the primary analysis, but had day and data since all patients in Part 2 (and the combined data set) interaction day by concentration as additional terms. The initially received this dose (Table 1). term time was defined as time since first dose (i.e., the same times on days 1 and 15 were considered different factor lev - els). If the interaction term day by concentration was not Results significant at the two-sided 5% level, a model without this term was also fitted. The primary model for day 1 was also Patient enrollment, demographics, and baseline fitted to the extended ER set and the day 15 model was fit- characteristics ted to the extended day 15 ER set. These extended analy- sis sets included baseline data obtained in the first 15 min A total of 105 patients, a subset of the total number of after drug administration for cases where no pre-dose value patients enrolled in this phase 1 study , were included in could be obtained; and for day 15, values from those patients the analysis (Table 1). In Part 1, 24 patients who qualified where the day 15 visit was actually performed more than 21 for the ECG assessment were enrolled; six patients were in days after start of treatment. The primary analysis was also the low-dose group. On day 1, all 24 patients were included repeated for the correction method (QTcF or QTcSS) that in the ECG set; while on day 15, only 18 patients (five in was not used in the primary analysis. the low-dose group) were eligible for inclusion. In Part 2, a total of 85 patients were evaluated for ECG assessment, Categorical analysis of quantitative ECG parameters and 81 were eligible for inclusion in the ECG set. All 81 patients contributed data on day 1, but only 55 had valid Incidences and percentages of patients with the follow- data on day 15. ing values were summarized: QTcF values > 450, > 480, Demographics and baseline characteristics of the patients and > 500 ms (treatment-emergent); ΔQTcF of > 30 and are provided in Table 2. The most common cancer types > 60 ms; PR values > 200 ms, which represent an increase were melanoma (37%), colorectal cancer (29%), non-small from baseline of at least 25%; QRS values > 110 ms, which cell lung cancer (9%), and lung cancer not otherwise speci- represent an increase from baseline of at least 25%; and heart fied (6%). All patients had received prior cancer therapy. rate decreases > 25% from baseline to a rate of < 50 bpm and increases > 25% from baseline to a rate of > 100 bpm. Table 1 Patient disposition of the ECG set by dose received, study part, and day Morphological analysis Dose received (BID) Part 1 Part 2 Incidence counts for new treatment-emergent morphologi- Day 1 Day 15 Day 1 Day 15 cal abnormalities were also summarized using the following Low dose (10–300 mg) 6 5 categories for clinical interpretation of the ECGs: normal High dose (450–900 mg) 18 13 ECG [sinus rhythm between 50 and 100 bpm with normal 450 mg 0 1 P waves, atrioventricular (AV) conduction, QRS complex 600 mg 81 54 and ST segment, and T-wave morphologies]; new (i.e., Total 24 18 81 55 not present at baseline) abnormal ECG findings, probably 1 3 1134 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 Table 2 Demographic and Parameter Total (n = 105) baseline characteristics Age (years), mean (SD) 59.1 (12.43) Sex, n (%) Female 42 (40) Male 63 (60) Race, n (%) White 90 (86) Black African Heritage or African American 6 (6) Asian 5 (5) Other 4 (4) BMI (kg/m ), mean (SD) 27.32 (5.93) Baseline ECOG performance status, n (%) 0 37 (35) 1 67 (64) 2 1 (< 1) Cancer type, n (%) Melanoma 39 (37) Colorectal 20 (19) NSCLC 9 (9) Lung (NOS) 6 (6) Other cancers (e.g., glioblastoma, thyroid, prostate, gastrointestinal, pancreatic, salivary 31 (29) gland, squamous cell carcinoma, etc) LVEF (%) n = 104 (> 99) Assessed by echocardiogram , n (%) 99 (94) Assessed by MUGA scan , n (%) 5 (5) Mean LVEF (SD) 62.15 (6.243) Median LVEF 62.8 Minimum, maximum LVEF 43.0, 82.0 ECOG Eastern Cooperative Oncology Group, LVEF left ventricular ejection fraction, MUGA multigated acquisition, NOS not otherwise specified, NSCLC non-small cell lung cancer, SD standard deviation Echocardiogram and MUGA were not performed on one patient at the baseline evaluation Most patients (94%) had their tumor molecularly genotyped, the Fridericia correction, the QTcSS was selected as the pri- and abnormalities (e.g., mutations, fusions, and rearrange- mary correction method for this analysis. ments) in BRAF, KRAS, and MEK were documented. Medi- A linear regression of QTcF and QTcSS on the R–R inter- cal histories included a cardiac disorder in 15% (16 of 105) val (in seconds) further supported the superiority of QTcSS of patients; these disorders included coronary artery disease, by showing a regression coefficient closer to zero (10.2 vs atrial and ventricular fibrillation, tachycardia and bradycar - 37.2 ms/s for QTcF). Likewise, the RMSS for QTcSS was dia, aortic valve disorder, pericardial effusion, and pericar - slightly smaller than that for QTcF (18.6 vs 19.1 ms/s). Con- ditis. In addition, 50% of patients were reported to have a sequently, QTcSS was confirmed as the primary endpoint history of hypertension. variable for this analysis. Selection and validation of primary endpoint variable Exclusion of individual measurements and interpolation of PK values Consistent with the statistical analysis plan, the primary QTc endpoint parameter was tested and validated before com- PK values obtained at the time of ECG samplings were inter- mencing analysis. The estimate of the QTcSS model param- polated from the available PK data as explained above (see eter (based on the individual replicates) yielded a slope of “PK parameters”). Overall, 84 measurements, i.e., 6% of all 0.39 ms/s with a two-sided 95% CI of 0.348–0.422 ms/s. measurements, had to be excluded because of violation of Since these CIs do not include 0.333, which characterizes time constraints. 1 3 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 1135 Appropriateness of model Predictions of QTc effect based on the primary models In Part 1, ΔΔQTcSS and ΔΔQTcF, the mean time-matched differences in QTc between the high- and low-dose groups, The predicted QTcSS change from baseline values at indicated a QTc shortening rather than a prolongation. the days 1 and 15 C were both negative (− 0.529 and max Consequently, hysteresis was considered inconsequential. − 9.202 ms, respectively), with both two-sided 90% CIs well Likewise, there was no indication that the linear model below the 10 ms regulatory threshold (5.562 and 4.101 ms was inappropriate. for days 1 and 15, respectively). The predictions for higher exposures at 1.5 times the observed C were also below max 5 ms for both days, and the 90% CIs were well below 10 ms Primary analysis for days 1 and 15 (6.401 and 6.977 ms, respectively). The wider CIs on day 15 were attributable to the large variability The primary models showed positive, although non-signif- of this prediction on that study day (Table 4). icant regression slopes, for the relationship between ulix- Consequently, the predictions based on the two models ertinib concentration and QTcSS (0.53 and 1.16 ms per µg/ (for days 1 and 15) excluded a QTc effect of concern on both mL for days 1 and 15, respectively) (Table 3; Fig. 1 for day days for concentrations associated with ulixertinib doses of 1 and Fig. 2 for day 15). The standard error for the slope 600 mg and above. The width of the CI on day 15 indicated estimates was comparable in both models, but the standard that, in this group of cancer patients, a prediction of the drug errors for all other parameters, including the estimates for effect after 15 days of treatment is quite uncertain; however, the time effect not attributable to the drug (Table 3) were given a terminal phase elimination rate of approximately substantially larger for day 15. In both models, the param- 9–11 h, ulixertinib is most likely at steady state by day 15. eter for Part was not significant; for this reason, pooling of Thus, a further increase in the QTc under the same condi- the two parts of the study seemed appropriate. The param- tions is quite unlikely. Furthermore, PK analyses of samples eter for dose group was small and not significant for day 1. collected after day 15 confirm that steady state was achieved However, for day 15, it was substantially larger and with by day 15. Even under these conditions, a relevant QTc pro- p = 0.068, and it was significant at the two-sided 10% level longation of concern can be excluded. but not significantly different from zero at the two-sided 5% level because of the larger standard error. Since the Sensitivity and robustness analyses low-dose group was used as a surrogate for placebo in this analysis, the parameter for dose group was an indicator for To further corroborate the model and investigate the robust- the appropriateness of the linear model used. For day 1, ness of its outcomes, we performed a number of sensitivity this parameter did not give any indication that the model analyses, including fitting a joint model for the day 1 and was inappropriate. For day 15, the result was less clear, but day 15 data, refitting the joint model for day 1 and day 15 if a two-sided p value of 5% was taken as the criterion, a without the interaction term, and fitting the primary model linear model without hysteresis appeared acceptable. to QTcF. Separate analyses of the two parts of the study were also performed (data not shown). Table 3 Key parameters of the Day Parameter Estimate SE DF t value p value 90% CI primary models 1 C:E slope 0.53 1.14 506 0.469 0.639 (− 1.343, 2.412) Dose group − 1.12 2.88 171 − 0.388 0.698 (− 5.880, 3.643) Part − 0.56 1.68 211 − 0.331 0.741 (− 3.327, 2.215) 15 C:E slope 1.16 1.74 109 0.664 0.508 (− 1.732, 4.042) Dose group − 11.55 6.24 90.9 − 1.849 0.068 (− 21.925, − 1.171) Part 6.68 4.07 105 1.642 0.104 (− 0.072, 13.430) Model used dQTcSS ~ C + dg + time + part + (1 + C|subjP) Kenward–Roger approximation was used C:E slope concentration:effect (QTc) regression slope, CI confidence interval, DF degrees of freedom, SE standard error The parameter for dose group replaces the intercept in a model with time as factor. It is an indicator for the appropriateness of the model, with values significantly different from zero indicating misfit 1 3 1136 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 y= -1.1183 +0.5343 *x -10 -20 -30 -40 -50 -55 0 1 2 3 4 5 6 Concentration(ug/ml) *Note that the regression line ignores the effect of time and therefore does not pass through the center of the data cloud Fig. 1 Primary model: raw ΔQTCSS values vs concentration on day 1 A joint model combining days 1 and 15 was performed Eec ff ts on other ECG parameters as a sensitivity analysis and produced an estimated con- centration-QTc slope that was nonsignificant and negative For the purpose of summarizing the ECG parameters, data [− 4.9 ms (90% CI − 11.2, 1.4) for day 1 and − 3.6 ms from the high-dose group (600 and 900 mg), which was (− 10.529, 3.328) for day 15]. As a result, the predictions primarily the 600-mg group (Table 1), are presented. The and CIs were also smaller. However, as pointed out above, mean heart rate at baseline was 75.8 bpm for the high-dose the smaller standard error of the slope estimates, which group. Change from baseline data for this group showed a resulted in substantially lower upper limits of the CI for small increase in heart rate during the 12-h monitoring, up the predictions, should be interpreted with caution because to 5.6 bpm on day 1 (at 4-h post-dose) and up to 7 bpm on of the observed differences in variability between the two day 15 (at 6- and 12-h post-dose) (Table 5). The mean PR study days. Furthermore, the results using QTcF did not interval at baseline was 164.3 ms for the high-dose group. differ substantially from those obtained with QTcSS. It Change from baseline data showed a small decrease in the should be noted that the slope on day 15 was somewhat PR interval on day 1 of up to 1.8 ms (at 12-h post-dose) larger when using QTcF, but still not statistically signifi- and a small increase of up to 3.7 ms on day 15 (at 1-h post- cant. The predicted QTcF at the day 15 C (− 15.012 ms) dose), followed by a small decrease thereafter. Mean QRS max was even more negative than that seen with QTcSS and data were quite stable on both days with small increases in reached statistical significance. Likewise, the upper limit QRS of up to 0.8 and 3.1 ms on days 1 and 15, respectively. of the CI for this prediction was − 1.928 ms compared with Uncorrected QT interval data largely showed reduc- 4.101 ms for QTcSS. tions of up to 10.3 ms on day 1 and 13.0 ms on day 15 1 3 ΔQTcSS (ms) Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 1137 y= -11.5480 +1.1551* x -10 -20 -30 -40 -50 -55 0 2 4 6 Concentration(ug/ml) *Note that the regression line ignores the effect of time and therefore does not pass through the center of the data cloud Fig. 2 Primary model: raw ΔQTCSS values vs concentration on day 15 Categorical analyses Table 4 Predicted QTcSS effect based on the primary models The number of patients exceeding the predefined QTcF Day Dose/condition Concentra- Prediction 90% CI thresholds (450, 480, and 500 ms) at any time was generally tion (µg/ (ms) small and mostly confined to the lower threshold of > 450 ms mL) [seven of 88 patients (8%) had a QTcF value > 450 ms, one 1 C 1.102 − 0.529 (− 6.621, 5.562) had a value > 480 ms, and one had a value > 500 ms]. The max 1.5C 1.652 − 0.236 (− 6.873, 6.401) changes from baseline in QTcF (> 30 or > 60 ms) were also max 15 C 2.031 − 9.202 (− 22.505, small, primarily at the lower level (> 30 ms), and more fre- max 4.101) quently on day 15 (7 vs 3%). No patient had a change from 1.5C 3.046 − 8.030 (− 23.036, max baseline of > 60 ms on day 1, and only one patient reached 6.977) this value at day 15. Model used dQTcSS ~ C + dg + time + part + (1 + C|subjP) Heart rate changes were primarily positive [i.e., increases above the predefined thresholds (> 100 bpm and Δ% > 25%)], quite frequent (47% at any time), and occurred equally on both study days. Reductions in heart (both at 12-h post-dose). Mean QTcSS values showed rate (< 50 bpm and Δ% < − 25%) were observed in 8% of trivial changes of up to 4.6 and 6.3 ms on days 1 and 15, patients, primarily on day 1. Overall, these data support respectively (both at 2-h post-dose). QTcF mean values the findings of the summary statistics, suggesting a small followed a similar trend with maximum values of 4.2 and increase in heart rate. 5.2 ms on days 1 and 15, respectively, (again, at 2-h post- Changes in PR values above the threshold of 200 ms dose for both days). and ΔPR > 25% were observed in only one patient on day 1 3 ΔQTcSS (ms) 1138 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 Table 5 Mean (SD) change from baseline in ECG parameters (high-dose group) Study day ECG timepoint HR (bpm) PR (ms) QRS (ms) QT (ms) QTcSS (ms) QTcF (ms) Day 1 0.5 h post-dose − 2.1 (4.8) 2.6 (13.3) 0.4 (4.9) 5.2 (10.9) 1.3 (8.9) 1.9 (8.5) 1 h post-dose − 0.9 (6.0) 1.1 (8.7) 0.8 (5.0) 3.7 (12.5) 2.4 (9.4) 2.6 (8.8) 2 h post-dose 1.5 (7.3) − 0.4 (8.5) 0.6 (4.7) 1.2 (16.5) 4.6 (9.5) 4.2 (9.5) 4 h post-dose 5.6 (8.7) − 1.2 (9.5) 0.8 (5.2) − 7.5 (20.6) 3.4 (12.2) 1.8 (12.5) 6 h post-dose 4.4 (9.4) − 0.9 (18.0) 0.0 (5.4) − 7.1 (22.9) 1.2 (12.6) − 0.0 (13.3) 8 h post-dose 4.8 (8.6) − 0.8 (16.1) 0.5 (5.7) − 8.5 (20.5) 1.2 (11.0) − 0.2 (11.4) 12 h post-dose 4.8 (10.0) − 1.8 (13.6) 0.1 (5.6) − 10.3 (25.0) − 1.1 (12.6) − 2.5 (13.4) Day 15 prior to dosing 3.2 (13.5) 2.1 (11.5) 0.6 (6.7) − 8.0 (34.0) − 2.3 (17.0) − 3.1 (18.1) 0.5 h post-dose 1.2 (11.5) 2.9 (13.7) 3.1 (5.9) − 0.6 (31.5) 2.0 (16.0) 1.7 (17.2) 1 h post-dose − 0.9 (6.0) 3.7 (14.3) 1.9 (6.0) 1.4 (30.5) 4.9 (16.6) 4.5 (17.3) 2 h post-dose 3.8 (11.7) 2.0 (15.1) 1.6 (5.9) − 2.1 (29.7) 6.3 (15.1) 5.2 (15.7) 4 h post-dose 6.1 (11.2) 2.9 (12.3) 1.7 (6.6) − 9.2 (29.1) 3.1 (16.9) 1.4 (17.5) 6 h post-dose 7.0 (12.9) − 2.1 (14.1) 1.3 (5.9) − 11.6 (31.7) 1.8 (14.5) − 0.1 (16.0) 8 h post-dose 6.5 (12.9) − 0.9 (12.8) 1.5 (6.0) − 10.6 (31.5) 2.2 (14.4) 0.4 (15.8) 12 h post-dose 7.0 (12.2) − 1.7 (13.0) 2.1 (7.0) − 13.0 (31.8) 0.4 (15.3) − 1.5 (16.7) N = up to 99 High–dose = 600–900 mg Bpm beats per minute, HR heart rate, ms millisecond, SD standard deviation 1. Considered in conjunction with the central tendency Discussion analysis, these changes are not clinically meaningful. No QRS prolongation meeting the predefined criteria Analysis of cardiac repolarization (QT/QTc) data from (> 110 ms and ΔQRS > 25%) were observed. oncology drug development studies is particularly chal- Summaries of the overall ECG diagnostic classification lenging because of study design limitations, patients’ showed no relevant change from the baseline distribution conditions, and a range of confounding factors. Oncology (53, 11, and 34% for normal, insignificantly abnormal, and studies usually do not include a placebo control group, significantly abnormal, respectively) on day 1. However, may have a limited PK exposure due to drug toxicity, on day 15 an increase in the proportion of significantly may be underpowered for QT/QTc assessment, and are abnormal ECGs was observed (42, 2, and 56%, respec- often conducted in busy hospital oncology departments tively) 12-h post-dose. These observations need to be con- or outpatient clinics less experienced with the collec- sidered in the context of the advanced illness of the patient tion of high-quality ECG data. Moreover, cancer patients population; thus, ECG changes are to be expected. often have serious medical conditions and are susceptible to clinical deterioration due to disease progression and treatment-related complications, including multi-organ Morphological analyses drug toxicities, intercurrent infections, and fluid and elec - trolyte disturbances . In addition, they may have had Repolarization abnormalities, comprised of ST segment prior exposure to drugs associated with cardiotoxic effects and T wave and U wave changes, including events of QT and are likely to be treated with multi-drug combinations prolongation (> 450 ms) or QT shortening (< 340 ms), with known or potential cardiac adverse effects, includ- were reported in similar proportions of patients at base- ing QT/QTc interval prolongation and possibly other ECG line (5%) and post-baseline (up to 7%). Likewise, T wave and cardiovascular effects. All of the above may lead to morphology changes, including biphasic, flat, inverted, higher variability in heart rate and the QT/QTc interval, notched, and peaked T waves, were similarly reported at further undermining the ability to rule-in or rule-out a rel- baseline (15%) and post-baseline timepoints (up to 18%). evant QT/QTc effect. For all of these reasons, the current Conduction abnormalities, consisting of all the types of ICH-E14 paradigm is a challenge when applied to onco- AV blocks (although no Mobitz 2 or complete heart block logic agents, where healthy volunteer studies are not pos- events were observed), were reported in 10% of patients sible. As a consequence, recent changes to the ICH-E14 at baseline and up to 13% post-baseline. 1 3 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 1139 guidance promoting the use of ERM analysis in early The primary analysis, designed to separate the effect of phase studies, present a unique opportunity for improved drug concentration from diurnal and ultra-diurnal effects not QT analysis in oncology drug development. attributable to the drug concentration, yielded a nonsignifi- In the current study, we used advanced ERM in an early cant positive slope of about 0.53 ms per µg/mL. The pre- phase oncology study. Although this method lacks placebo dicted diurnal variation of the mean QTcSS interval (inde- data , it is nevertheless expected to be more powerful pendent of drug concentration) showed an estimated peak than the traditional by-timepoint analysis, and promising effect of 4.88 ms at 2-h post-dose on day 1 and 10.87 ms at results have also been obtained without the availability of 2-h post-dose on day 15, with corresponding upper CIs of placebo data . Moreover, the use of the low-dose group 9.196 and 18.426 ms, respectively. Notably, the estimated as a surrogate for placebo should further enhance the value QTc values related to diurnal variability are much larger than of this method. the predicted drug effects, emphasizing the relatively small Selection of the most appropriate QTc endpoint variable magnitude of drug effect, if any, on the QTc interval (noting for a given study population may have an impact on estimat- that both slopes were not significantly different from zero). ing the QT/QTc effect of a drug, especially if there is a con- Overall, these findings are consistent with a favorable current effect on heart rate . For this purpose, we investi- repolarization profile of ulixertinib. No relevant changes in gated the relationship between the raw (uncorrected) QT and cardiac conduction (PR and QRS intervals) or repolarization R–R intervals during drug-free study periods. The resulting morphology (ST segment and T wave) were observed in this drug-free QT:RR regression slope, estimated at 0.39 ms/s study. A small negative mean QTc effect that, at the expected (90% CI 0.348, 0.422 ms/s), was then used as the power therapeutic exposures, produced mean changes below 5 ms exponent for the QTcSS. This population-specific QT:RR and upper 90% CIs that were well below the 10-ms regula- slope was somewhat higher than the Fridericia (QTcF) tory threshold for non-oncologic drugs and well below the exponent (0.33) and was shown to be a superior correction commonly used 20-ms threshold for oncologic medications. method for this data set. However, since the observed effect The QTc data analyzed with ulixertinib in this single-agent on heart rate was relatively small, no meaningful differences study support its further development, and its metabolism between the correction methods were observed. by multiple metabolic pathways suggests a small risk for higher drug exposures due to drug interactions or metabolic Eec ff t on ECG parameters inhibition. Analysis of the heart rate data, using descriptive statistics, suggests a small increase in heart rate of up to 6 and 7 bpm Study limitations and data interpretation on days 1 and 15 at 4 and 6-h post-dose, respectively. The appropriateness of the ER model developed for this Similar to other oncology studies, the main limitations of study to assess the relationship between ulixertinib exposure this phase 1 oncology study was the absence of a placebo (concentrations) and QTc ee ff ct was thoroughly investigated control group and thus the ability to adjust for disease pro- and confirmed following the predefined statistical analysis gression and have a direct correction for diurnal variability. plan. Tests for linearity and hysteresis (delayed effect) con- In addition, the large variability of the QT/QTc intervals, firmed that a linear model was appropriate and that a cor - particularly on day 15, is a challenge commonly observed rection for hysteresis was not required. Sensitivity analyses in oncologic studies. In the present analysis, the absence of to examine the robustness of the model and the selected a placebo group was compensated for by using the low-dose primary QTc variable were also performed and confirmed groups as surrogate for placebo. This allowed an estima- the fitness of the model parameters and predictions. tion of a dose group effect that was shown to be a sensitive On the basis of the primary model, the predicted QTcSS indicator for the appropriateness of the model. However, effect at the day 1 estimated C (1.102 µg/mL) showed a it must be kept in mind that the low-dose group consisted max small mean negative change of − 0.529 ms (i.e., QTc short- of only six patients. In addition, a possible change in the ening) with an upper 90% CI of 5.562 ms (i.e., worst-case patients’ conditions during the first 2 weeks of the study, scenario QTc prolongation). The predicted mean QTcSS resulting in the observed higher variability in the QTc data change for day 15 at the estimated C (2.031 µg/mL) on day 15, cannot be excluded. This is also reflected in the max showed a larger mean shortening of − 9.202 ms with an larger standard error for all parameter estimates, except for upper CI of 4.101 ms. The wide CI on day 15 (− 22.505, the slope parameter for this day. 4.101) reflects, in part, the expected wide variability in this Compared with the analysis based on day 1, results based severely ill population of oncology patients , although it on day 15 seem to be more prone to errors, as reflected by may also indicate a progressive change in patients’ condition the large width of the CIs for the predictions. Consider- during the study or, possibly, a deterioration in data quality. ing this caveat, reassuringly the predictions for day 15 1 3 1140 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 Open Access This article is distributed under the terms of the Crea- consistently excluded a QTc prolongation of concern (i.e., tive Commons Attribution 4.0 International License (http://creat iveco 10 ms and above). mmons.or g/licenses/b y/4.0/), which permits unrestricted use, distribu- While this dose group effect was small and non-signif- tion, and reproduction in any medium, provided you give appropriate icant for day 1, it was larger for day 15. Although with a credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. p value of 0.068, it misses the threshold where the model would be considered inadequate, the moderate fit of the day 15 model somewhat limits the value of the predictions for References this day. As pointed out earlier, such a limitation is not unex- pected in a study in oncologic patients. 1. Sullivan RJ, Infante JR, Janku F, Wong DJL, Sosman JA, Keedy The totality of the data, including those collected from V, Patel MR, Shapiro GI, Mier JW, Tolcher AW, Wang-Gillam A, patients in the dose-escalation and early cohort-expansion Sznol M, Flaherty K, Buchbinder E, Carvajal RD, Varghese AM, phases, along with statistical power considerations, led to a Lacouture ME, Ribas A, Patel SP, DeCrescenzo GA, Emery CM, Groover AL, Saha S, Varterasian M, Welsch DJ, Hyman DM, Li decision by the ulixertinib team to discontinue collection of BT (2017) First-in-class ERK1/2 inhibitor ulixertinib (BVD-523) additional Holter data from patients enrolled at the end of in patients with MAPK mutant advanced solid tumors: results of a cohort expansion. phase I dose-escalation and expansion study. Cancer Discov. https ://doi.org/10.1158/2159-8290.CD-17-1119 2. American Cancer Society (2016–2017) Cancer treatment & sur- vivorship facts & figures. https ://www.cance r.org/conte nt/dam/ Conclusion cance r-org/resea rch/cance r-facts -and-stati stics /cance r-treat ment- and-sur vi vorsh ip-facts -and-figur es/cance r-tr eat ment-and-sur vi On the basis of the analysis of the ECG data from this phase vorsh ip-facts -and-figur es-2016-2017.pdf. Accessed 29 Jan 2018 3. Roden DM (2004) Drug-induced prolongation of the QT interval. 1 oncology study in patients with solid tumors treated with N Engl J Med 350:1013–1022 ulixertinib administered at clinically relevant doses, primar- 4. International Council for Harmonisation (2005) The clinical ily at the RP2D of 600-mg BID, ulixertinib has a low risk evaluation of QT/QTc interval prolongation and proarrhythmic for QT/QTc prolongation. potential for non-antiarrhythmic drugs E14. http://www.ich.org/ filea dmin/Publi c_Web_Site/ICH_Produ cts/Guide lines /Effic acy/ E14/E14_Guide line.pdf. Accessed 29 Jan 2018 Acknowledgements The authors thank the following investigators who 5. Stockbridge N, Zhang J, Garnett C, Malik M (2012) Practice and participated in the phase 1 study (NCT01781429): Infante JR, Wong challenges of thorough QT studies. J Electrocardiol 45(6):582–587 DJL, Sosman JA, Keedy V, Patel MR, Shapiro GI, Mier JW, Tolcher 6. Sarapa N, Britto MR (2008) Challenges of characterizing proar- AW, Wang-Gillam A, Flaherty K, Buchbinder E, Carvajal RD, Var- rhythmic risk due to QTc prolongation induced by nonadjuvant ghese AM, Lacouture ME, Ribas A, Patel SP,and Hyman DM. Authors anticancer agents. Expert Opin Drug Saf 7(3):305–318 also wish to thank Mary Varterasian, MD for scientific review of the 7. Darpo B, Benson C, Dota C, Ferber G, Garnett C, Green CL, Jaru- manuscript and Deborah Sommerville, MWC, ELS for providing edi- gula V, Johannesen L, Keirns J, Krudys K, Liu J, Ortemann-Renon torial assistance. C, Riley S, Sarapa N, Smith B, Stoltz RR, Zhou M, Stockbridge N (2015) Results from the IQ-CSRC prospective study support Compliance with ethical standards replacement of the thorough QT study by QT assessment in the early clinical phase. Clin Pharmacol Ther 97(4):326–335 Conflict of interest BM was formerly employed by Bioclinica and is 8. International Council for Harmonisation (2015) E14 Implementa- currently an independent consultant at Cardiac Safety Consultants Ltd. tion working group. ICH E14 guideline: questions and answers: GF is independent consultant at Statistik Georg Ferber GmbH and is the clinical evaluation of QT/QTc interval prolongation and proar- consulting for Bioclinica. FJ, BTL and RJS were investigators on the rhythmic potential for non-antiarrhythmic drugs (R3). http://www. clinical study where the data reported here were collected. DW is em- ich.org/filea dmin/Publi c_W eb_Site/ICH_Pr odu cts/Guide lines / ployed by BioMed Valley Discoveries Inc. WC was formerly employed Effic acy/E14/E14_Q_As_R3__Step4 .pdf. Accessed 29 Jan 2018 by Bioclinica and is currently employed by Shanghai Hengrui Phar- 9. Tornøe CW, Garnett CE, Wang Y, Florian J, Li M, Gobburu JV maceutical Co., Ltd. JJ and OW are employed by Bioclinica. PS is an (2011) Creation of a knowledge management system for QT anal- independent consultant and faculty member at Stanford University and yses. J Clin Pharmacol 51(7):1035–1042 is consulting for BioMed Valley Discoveries. All authors had full con- 10. Garnett C, Needleman K, Liu J, Brundage R, Wang Y (2016) trol of all primary data, which are available for review upon request. Operational characteristics of linear concentration-QT models for assessing QTc interval in the thorough QT and phase I clinical Ethical approval All procedures performed in studies involving human studies. Clin Pharmacol Ther 100(2):170–178 participants were in accordance with the ethical standards of the insti- 11. Kenward M, Roger J (1997) Small sample inference for fixed tutional and/or national research committee and with the 1964 Helsinki effects from restricted maximum likelihood. Biometrics declaration and its later amendments or comparable ethical standards. 53:983–997 12. Ferber G, Lorch U, Täubel J (2015) The power of phase I studies Welfare of animals This article does not contain any studies with ani- to detect clinical relevant QTc prolongation: a resampling simula- mals performed by the authors. tion study. Biomed Res Int. https://doi.or g/10.1155/2015/293564 13. Ferber G, Wang D, Täubel J (2014) Concentration-ee ff ct modeling Informed consent Informed consent was obtained from all individual based on change from baseline to assess the prolonging effect participants included in the study. of drugs on QTc together with an estimate of the circadian time course. J Clin Pharmacol 54(12):1400–1406 1 3 Cancer Chemotherapy and Pharmacology (2018) 81:1129–1141 1141 14. Garnett CE, Zhu H, Malik M, Fossa AA, Zhang J, Badilini F, drug-induced heart rate changes or other autonomic effects. Am Li J, Darpö B, Sager P, Rodriguez I (2012) Methodologies to Heart J 163(6):912–930 characterize the QT/corrected QT interval in the presence of 1 3
Cancer Chemotherapy and Pharmacology – Springer Journals
Published: Mar 30, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera