Adaptive designs from a Data Safety Monitoring Board perspective: Some controversies and some case studies.

This article describes vignettes concerning interactions with Data Safety Monitoring Boards during the design and monitoring of some clinical trials with an adaptive design. Most reflect personal experiences by the author.

Adaptive sample size modification in clinical trials: start small then ask for more?

We consider sample size re-estimation in a clinical trial, in particular when there is a significant delay before the measurement of patient response. Mehta and Pocock have proposed methods in which sample size is increased when interim results fall in a 'promising zone' where it is deemed worthwhile to increase conditional power by adding more subjects. Our analysis reveals potential pitfalls in applying this approach. Mehta and Pocock use results of Chen, DeMets and Lan to identify when increasing sample size, but applying a conventional level α significance test at the end of the trial does not inflate the type I error rate: we have found the greatest gains in power per additional observation are liable to lie outside the region defined by this method. Mehta and Pocock increase sample size to achieve a particular conditional power, calculated under the current estimate of treatment effect: this leads to high increases in sample size for a small range of interim outcomes, whereas we have found it more efficient to make moderate increases in sample size over a wider range of cases. If the aforementioned pitfalls are avoided, we believe the broad framework proposed by Mehta and Pocock is valuable for clinical trial design. Working in this framework, we propose sample size rules that apply explicitly the principle of adding observations when they are most beneficial. The resulting trial designs are closely related to efficient group sequential tests for a delayed response proposed by Hampson and Jennison.

Group sequential enrichment design incorporating subgroup selection.

An important component of clinical trials in drug development is the analysis of treatment efficacy in patient subgroups (subpopulations). Because of concerns of multiplicity and of the small sample sizes often involved, such analyses can present substantial statistical challenges and may lead to misleading conclusions. As a confirmatory seamless phase II/III design, we propose an adaptive enrichment group sequential procedure whereby resources can be concentrated on subgroups most likely to respond to treatment. Stopping boundaries are defined through upper and lower spending functions. The procedure is presented in terms of the efficient score, enabling the analysis of normal, binary, or time-to-event data. It addresses the dilution effect by eliminating populations at the first stage that appear likely to derive no therapeutic benefit. It subsequently proceeds with the definitive assessment of treatment efficacy among the remaining pooled populations using a group sequential design. The procedure provides strong protection of familywise type I error rate, and we employ a bootstrap algorithm to obtain point and interval estimates that are adjusted for the selection bias. We give examples to demonstrate how the design is used. We make comparisons with adaptive two-stage combination test procedures and with a group sequential test that does not account for the presence of subgroups. Numerical results show that the procedure has high power to detect subgroup-specific effects and the use of multiple interim analysis points can lead to substantial sample size savings.

Incident autoimmune conditions among males receiving quadrivalent human papillomavirus vaccine in the United States.

BACKGROUND: The potential for vaccines to induce autoimmunity has been the subject of considerable investigation and autoimmune induction remains a common focus for vaccine safety research. This study assessed the risk of new onset autoimmune conditions among males receiving the 4-valent human papillomavirus (HPV) vaccine (4vHPV). METHODS: Within a US health insurance claims database, we formed a cohort of male 4vHPV vaccine recipients between 2009 and 2016, along with a propensity score matched cohort of males who did not receive the 4vHPV vaccine. The study outcome was new onset autoimmune conditions (20 separate conditions) within four categories (rheumatologic/hematologic, gastroenterologic, endocrinologic and neurologic/ophthalmalogic). Outcomes identified using diagnosis codes were adjudicated through medical record review. Incidence rates (per 1,000 person-years) were estimated for the vaccinated and unvaccinated groups along with rate ratios (RRs). RESULTS: There were 65,606 males receiving at least one dose of 4vHPV vaccine, and 55,670 were matched to a comparator. The matched 4vHPV vaccine cohort provided 35 confirmed cases among 39,735 person-years, for an incidence rate of 0.88 (95% CI: 0.61-1.23), while the comparator cohort provided 47 confirmed cases among 58,215 person-years, an incidence rate of 0.81 (0.59-1.07), a RR of 1.09 (0.70-1.69). The RR within categories was 0.49 (0.10-2.42) for rheumatologic/hematologic, 1.26 (0.58-2.71) for gastroenterologic, 1.11 (0.61-2.02) for endocrinologic and 1.46 (0.21-10.40) for neurologic. CONCLUSIONS: The incidence of autoimmune conditions among males receiving the 4vHPV vaccine was similar to that among unvaccinated males. These results are consistent with other studies that have assessed autoimmunity with the 4vHPV vaccine.

Treatment with duloxetine in adults and the incidence of cardiovascular events.

BACKGROUND: Cardiovascular events are inconclusively associated with duloxetine use in clinical trials and spontaneous reports. This analysis of cardiovascular events in relation to duloxetine use within a large health insurance database provides further data on the association. METHODS: This cohort study was conducted within a population with commercial health insurance. Adults with depression who initiated duloxetine were matched to separate cohorts of initiators of venlafaxine, selective serotonin reuptake inhibitors (SSRIs), and tricyclic antidepressants (TCAs), along with untreated patients with depression, and enrollees without depression. The cohorts were followed for cardiovascular events (acute myocardial infarction, sudden death, hypertensive crisis, arrhythmia, and coronary revascularization), which were identified through health insurance claims and confirmed upon review of underlying medical records. Proportional hazards and Poisson regression models were used for comparisons. RESULTS: There were approximately 64,000 person-years of follow-up among all cohorts (including 17,386 person-years among 21,457 duloxetine initiators), yielding 279 cardiovascular events. Relative to duloxetine initiators, those without depression had lower rates of combined events (incidence rate ratio [IRR], 0.51; 95% confidence interval [CI], 0.32-0.81) and coronary revascularizations (IRR, 0.51; 95% CI 0.29-0.89). The IR of each of the cardiovascular outcomes did not differ across the other cohorts, even accounting for time since last duloxetine dispensing. CONCLUSION: The incidence of cardiovascular events did not differ among duloxetine initiators relative to other antidepressant comparators or those with untreated depression but was higher than those without depression, suggesting that depression itself (or associated morbidities) may affect the risk of cardiovascular events.

Vaccine discontinuation and switching following regulatory interventions in response to rotavirus vaccine contamination with porcine circovirus DNA fragments.

PURPOSE: The Food and Drug Administration temporarily suspended monovalent rotavirus vaccine (RV1) use following discovery of contamination with porcine circovirus fragments and subsequently announced similar contamination of the pentavalent rotavirus vaccine (RV5) but recommended continued use of the product. We assessed the utilization of these vaccines in relation to the announcements. METHODS: Using claims submitted to a commercial health insurer for administration of RV1 and RV5, we estimated the number of administrations of the vaccines and the extent of switching between RV1 and RV5. Procedure codes on submitted claims identified vaccine administrations among infants ≤ 1 year old through 16 June 2010. Among infants who received a first dose of vaccine before the corresponding announcement, and whose second dose was anticipated following the announcement, we estimated the number who received no second dose of rotavirus vaccine. RESULTS: There were 31 178 RV1 initiators and 514 357 RV5 initiators. We observed a 93% reduction in RV1 doses in the month following the recommended suspension of use, coupled with extensive switching to RV5 (90% of subsequent doses) and a reduction in second RV1 doses (from 35.5% incomplete to 40.9%). There was a 15% increase in number of RV5 administrations following announcement of its contamination, with little switching to RV1 but with a possible decrease in completion. CONCLUSIONS: Recommended suspension of RV1 use led to a substantial decrease in use and extensive switching to RV5. The announcement that RV5 was similarly contaminated, but without a corresponding recommendation to suspend use, had little effect on use.

Vaccine initiation and 3-dose series completion of 4vHPV vaccine among US insured males 2012-2016.

OBJECTIVE: The quadrivalent human papillomavirus vaccine (4vHPV, GARDASIL®), was approved in the US in 2009 for use in males aged 9 to 26 for the prevention of HPV-related genital warts, and in 2010 for the prevention of certain HPV-related anogenital diseases. A regimen was approved in 2016 for those who initiate the vaccine series between the ages of 9 to 14 years. We describe patterns of 4vHPV administration among US males before this modification. METHODS: The study used a US health insurance claims database, and included males, age 9 to 26 years, who initiated 4vHPV between 2012 and 2016. Time from first dose to subsequent doses was estimated. Logistic regression identified factors associated with regimen completion. RESULTS: Among 100,786 males who initiated 4vHPV (corresponding to âˆ¼ 13% of male birth cohorts), 50,573 (50.2%) and 25,763 (25.6%) received a second and third dose, respectively. Annual administration was common, with 47% of males receiving 3 doses over 3 years (1 dose per year) as compared to 12% receiving the 3-dose series in the recommended 6-month timeframe. Receipt of 4vHPV was 2.2 (range 1.5 to 2.9) times as likely to occur in summer months compared to other times of the year. Individuals aged 18 to 21 years and those living in Western states or rural regions were less likely to complete the 3-dose regimen. CONCLUSIONS: The real-world patterns of 4vHPV vaccination observed, particularly the low uptake and regimen completion, suggest that better strategies are needed for males to improve 4vHPV vaccine use in males.

Safety of 4-valent human papillomavirus vaccine in males: a large observational post-marketing study.

The 4-valent human papillomavirus (HPV) vaccine (4vHPV vaccine), Gardasil®, is indicated for the prevention of several HPV-related diseases. The objective was to assess the safety of 4vHPV vaccine administered to males as part of routine care. The study used a US health insurance claims database, and included males, age 9 to 26 years, who initiated 4vHPV between October 2009 and December 2016. General safety outcomes were identified using ICD diagnosis codes associated with emergency room visits and hospitalizations in the claims database in risk periods (Days 1-60 and Days 1-14 following vaccine administration) and self-comparison periods (Days 91-150 and 91-104 for the Days 1-60 and Days 1-14 analysis, respectively). Incidence rates (IRs) and relative rates (RRs) with 95% confidence intervals (CIs) were calculated comparing the risk and self-comparison periods. In this study, 114,035 males initiated 4vHPV vaccine and received 202,737 doses. Using the 60-day time window, 5 outcomes had significantly elevated RRs after accounting for multiple comparisons: ear conditions (RR 1.28, 95% CI 1.03-1.59); otitis media and related conditions (RR 1.65, 95% CI 1.09-2.54); cellulitis and abscess of arm (RR 2.17, 95% CI 1.06-4.72); intracranial injury (RR 1.23, 95% CI 1.01-1.50); and concussion (RR 1.29, 95% CI 1.05-1.59). A higher rate of allergic reactions was noted on the day of 4vHPV vaccine receipt compared to other vaccines (21.07 events per 10,000 doses, 95% CI 18.89-23.44 versus 11.44 per 10,000 doses, 95% CI 9.84-13.22). A higher incidence rate of VTE was observed following vaccination but this association was not significant (RR 2.17, 95% CI 0.35-22.74). The 4vHPV vaccine was associated with same-day allergic reactions as well as ear infections, intracranial injury, cellulitis, and concussion within 2 months after vaccination. While allergic reaction and cellulitis are consistent with the known safety profile of 4vHPV vaccine, the association of the other outcomes were determined by an independent Safety Review Committee to be most likely a result of activities common in adolescent males that coincide with the timing of vaccination and not directly related to vaccination itself.Implications and Contributions: The study results support the general safety of routine immunization with 4vHPV vaccine among males to prevent HPV-related diseases and cancers.

Duloxetine for depression and the incidence of hepatic events in adults.

Elevated hepatic enzyme levels and hepatic injuries have been associated with duloxetine use in clinical trials and spontaneous reports, but the association of duloxetine with a broad spectrum of hepatic outcomes has not been assessed observationally. This cohort study of adult duloxetine initiators between 2004 and 2006 based on the Ingenix Research Data Mart involved 6 matched comparator cohorts, including 4 antidepressant initiator groups (venlafaxine, nefazodone, selective serotonin reuptake inhibitors, and tricyclic antidepressants), depressed but untreated patients, and individuals without depression. The cohorts were followed up for hepatic events, and proportional hazards regression compared duloxetine initiators with comparator cohorts, whereas Poisson regression compared duloxetine usage categories to account for changed therapy during follow-up. Approximately 64,000 person-years among 21,457 duloxetine initiators and comparator cohorts yielded 51 hepatic outcome events. Venlafaxine initiators (incidence rate ratio [IRR] = 0.34; 95% confidence interval [CI], 0.12-0.95) and the cohort without depression (IRR = 0.30; 95% CI, 0.10-0.93) had lower incidences of combined hepatic events than duloxetine initiators, whereas no other differences in hepatic events were observed for duloxetine initiators relative to selective serotonin reuptake inhibitors, tricyclic antidepressants, and untreated depressed patients. In as-treated analyses, relative to nonuse, current (IRR = 4.30; 95% CI, 1.45-12.81) and recent (IRR = 5.93; 95% CI, 1.63-21.55) duloxetine use was associated with greater incidence of less severe hepatic outcomes but not hepatic-related death and potential acute hepatic failure. Although duloxetine does not seem to increase the risk of hepatic-related death or acute hepatic failure, it may be associated with an increased risk of certain less severe hepatic events.

Atomoxetine and cerebrovascular outcomes in adults.

OBJECTIVE: The aim of this study was to estimate the association between atomoxetine and cerebrovascular accident (CVA) and transient ischemic attack (TIA) in adults. METHODS: This cohort study conducted within a health insurance database included 21,606 atomoxetine initiators matched to 21,606 stimulant attention-deficit/hyperactivity disorder (ADHD) medication initiators on the basis of propensity scores and a sample from the source population (N = 42,993). Outcomes were confirmed through a medical record review or a National Death Index search. Poisson regression was used to estimate the rate ratio and 95% confidence interval (CI) of CVA or TIA according to the treatment. Cox regression was used to estimate the hazards ratio (HR) and 95% CI for comparisons across cohorts. RESULTS: Forty-four CVAs and 21 TIAs occurred during a mean follow-up of 1.5 years. The rate ratio of the current atomoxetine compared with the current stimulant ADHD medication was 1.38 for CVA (95% CI, 0.42-4.54) and 0.31 for TIA (95% CI, 0.04-2.63). Results for atomoxetine compared with the stimulant ADHD medication according to initial cohort assignment were consistent, with no increased risk for CVA or TIA. An increased risk of TIA was observed between initiation of an ADHD medication compared with the general population (HR, 3.44; 95% CI, 1.13-10.60); however, a similar pattern was not observed for CVA (HR, 0.71; 95% CI, 0.34-1.47). CONCLUSIONS: These results do not support an increased risk of CV events with atomoxetine compared with the stimulant ADHD medication. Users of ADHD medications may be at an increased risk of TIA compared with the general population.

Evaluation of hospital and patient factors that influence the effective administration of surgical antimicrobial prophylaxis.

OBJECTIVE: To analyze and model the patient and healthcare system factors that may interfere with the appropriate administration of surgical antimicrobial prophylaxis. DESIGN: Between 1994 and 1998, surgical-site surveillance data were collected prospectively for a cohort of eligible surgical patients. For all cases, and each individual procedure (cardiothoracic, colonic, gynecologic, orthopedic, or vascular), forward stepwise multiple logistic regression was applied to relate key hospital and patient factors to an effective first prophylactic dose (ie, appropriate administration time, dose, route, and drug). SETTING: A 450-bed, tertiary-care teaching hospital in Canada. PATIENTS: A total of 4,835 patients admitted for surgical procedures who required antimicrobial prophylaxis. RESULTS: Factors positive for an effective first prophylactic dose for all cases were when an order was written (OR, 19.7; CI95, 9.1-42.7; P < .001) and given in the operating room (OR, 13.9; CI95, 7.5-25.6; P < .001). Factors negative for an effective first prophylactic dose were beta-lactam allergy (OR, 0.49; CI95, 0.4-0.61; P < .001) and same-day surgery (OR, 0.57; CI95, 0.4-0.82; P < .001). CONCLUSIONS: With few exceptions, the four factors included in the procedure models showed that when a preoperative order was written or the antibiotic was given in the operating room, a patient was more likely to receive an effective first prophylactic dose. Conversely, when a patient had a beta-lactam allergy or the surgery was performed on the day the patient was admitted, the administration of an effective first prophylactic dose was less likely.

Baseline characteristics and the effect of selenium supplementation on cancer incidence in a randomized clinical trial: a summary report of the Nutritional Prevention of Cancer Trial.

The Nutritional Prevention of Cancer Trial was a randomized, clinical trial designed to evaluate the efficacy of selenium as selenized yeast (200 microg daily) in preventing the recurrence of nonmelanoma skin cancer among 1312 residents of the Eastern United States. Original secondary analyses through December 31, 1993 showed striking inverse associations between treatment and the incidence of total [hazard ratio (HR) = 0.61, 95% confidence interval (CI) = 0.46-0.82], lung, prostate, and colorectal cancer and total cancer mortality. This report presents results through February 1, 1996, the end of blinded treatment. Effect modification by baseline characteristics is also evaluated. The effects of treatment overall and within subgroups of baseline age, gender, smoking status, and plasma selenium were examined using incidence rate ratios and Cox proportional hazards models. Selenium supplementation reduced total (HR = 0.75, 95% CI = 0.58-0.97) and prostate (HR = 0.48, 95% CI = 0.28-0.80) cancer incidence but was not significantly associated with lung (HR = 0.74, 95% CI = 0.44-1.24) and colorectal (HR = 0.46, 95% CI = 0.21-1.02) cancer incidence. The effects of treatment on other site-specific cancers are also described. The protective effect of selenium was confined to males (HR = 0.67, 95% CI = 0.50-0.89) and was most pronounced in former smokers. Participants with baseline plasma selenium concentrations in the lowest two tertiles (<121.6 ng/ml) experienced reductions in total cancer incidence, whereas those in the highest tertile showed an elevated incidence (HR = 1.20, 95% CI = 0.77-1.86). The Nutritional Prevention of Cancer trial continues to show a protective effect of selenium on cancer incidence, although not all site-specific cancers exhibited a reduction in incidence. This treatment effect was restricted to males and to those with lower baseline plasma selenium concentrations.

Selenium supplementation and lung cancer incidence: an update of the nutritional prevention of cancer trial.

Interest in the chemopreventive effects of the trace element selenium has spanned the past three decades. Of >100 studies that have investigated the effects of selenium in carcinogen-exposed animals, two-thirds have observed a reduction in tumor incidence and/or preneoplastic endpoints (G. F. Combs and S. B. Combs, The Role of Selenium in Nutrition Chapter 10, pp. 413-462. San Diego, CA: Academic Press, 1986, and B. H. Patterson and O. A. Levander, Cancer Epidemiol. Biomark. Prev., 6: 63-69, 1997). The Nutritional Prevention of Cancer Trial, a randomized clinical trial reported by Clark et al. (L. C. Clark et al., JAMA, 276: 1957-1963, 1996), showed as a secondary end point, a statistically significant decrease in lung cancer incidence with selenium supplementation. The adjusted hazard ratio (HR) was 0.56 [95% confidence interval (CI), 0.31-1.01; P = 0.05]. These results were based on active follow-up of 1312 participants. This reanalysis used an extended Nutritional Prevention of Cancer Trial participant follow-up through the end of the blinded clinical trial on February 1, 1996. The additional 3 years added 8 cases to the selenium-treated group and 4 cases to the placebo group, and increased follow-up to 7.9 years. The relative risk of 0.70 (95% CI, 0.40-1.21; P = 0.18) is not statistically significant. Whereas the overall adjusted HR is not significant (HR = 0.74; 95% CI, 0.44-1.24; P = 0.26), and the HR for current and former smokers was not significant, the trend is toward a reduction in risk of incident lung cancer with selenium supplementation. In a subgroup analysis there was a nominally significant HR among subjects with baseline plasma selenium in the lowest tertile (HR = 0.42; 95% CI, 0.18-0.96; P = 0.04). The analysis for the middle and highest tertiles of baseline showed HRs of 0.91 and 1.25. The current reanalysis indicates that selenium supplementation did not significantly decrease lung cancer incidence in the full population, but a significant decrease among individuals with low baseline selenium concentrations was observed.

Evaluating the safety of influenza vaccine using a claims-based health system.

INTRODUCTION: As part of the Centers for Disease Control and Prevention's monitoring and evaluation activities for influenza vaccines, we examined relationships between influenza vaccination and selected outcomes in the 2009-2010 and 2010-2011 influenza seasons in a claims-based data environment. METHODS: We included patients with claims for trivalent influenza vaccine (TIV) and/or 2009 pandemic influenza A H1N1 vaccine (H1N1) during the 2009-2010 and 2010-2011 influenza seasons. Patients were followed for several pre-specified outcomes identified in claims. Seizures and Guillain-Barré Syndrome were selected a priori for medical record confirmation. We estimated incidence rate ratios (IRR) using a self-controlled risk interval (SCRI) or a historical comparison design. Outcomes with elevated IRRs, not selected a priori for medical record review, were further investigated with review of claims histories surrounding the outcome date to determine whether the potential event could be ruled-out or attributed to other causes based on the pattern of medical care. RESULTS: In the 2009-2010 season, no significant increased risks for outcomes following H1N1 vaccination were observed. Following TIV administration, the IRR for peripheral nervous system disorders and neuropathy was slightly elevated (1.07, 95% CI: 1.01-1.13). The IRR for anaphylaxis following TIV was 28.55 (95% CI: 3.57-228.44). After further investigation of claims histories, the majority of potential anaphylaxis cases had additional claims around the time of the event indicating alternate explanatory factors or diagnoses. In the 2010-2011 season following TIV administration, a non-significant elevated IRR for anaphylaxis was observed with no other significant outcome findings. CONCLUSION: After claims history review, we ultimately found no increased outcome risk following administration of 998,881 TIV and 538,257 H1N1 vaccine doses in the 2009-2010 season, and 1,158,932 TIV doses in the 2010-2011 season.

The nutritional prevention of cancer: 400 mcg per day selenium treatment.

Nonexperimental studies suggest that individuals with higher selenium (Se) status are at decreased risk of cancer. The Nutritional Prevention of Cancer (NPC) study randomized 1,312 high-risk dermatology patients to 200-mcg/day of Se in selenized yeast or a matched placebo; selenium supplementation decreased the risk of lung, colon, prostate, and total cancers but increased the risk of nonmelanoma skin cancer. In this article, we report on a small substudy in Macon, GA, which began in 1989 and randomized 424 patients to 400-mcg/day of Se or to matched placebo. The subjects from both arms had similar baseline Se levels to those treated by 200 mcg, and those treated with 400-mcg attained plasma Se levels much higher than subjects treated with 200 mcg. The 200-mcg/day Se treatment decreased total cancer incidence by a statistically significant 25%; however, 400-mcg/day of Se had no effect on total cancer incidence.

Adaptive seamless designs: selection and prospective testing of hypotheses.

There is a current trend towards clinical protocols which involve an initial "selection" phase followed by a hypothesis testing phase. The selection phase may involve a choice between competing treatments or different dose levels of a drug, between different target populations, between different endpoints, or between a superiority and a non-inferiority hypothesis. Clearly there can be benefits in elapsed time and economy in organizational effort if both phases can be designed up front as one experiment, with little downtime between phases. Adaptive designs have been proposed as a way to handle these selection/testing problems. They offer flexibility and allow final inferences to depend on data from both phases, while maintaining control of overall false positive rates. We review and critique the methods, give worked examples and discuss the efficiency of adaptive designs relative to more conventional procedures. Where gains are possible using the adaptive approach, a variety of logistical, operational, data handling and other practical difficulties remain to be overcome if adaptive, seamless designs are to be effectively implemented.

Efficient group sequential designs when there are several effect sizes under consideration.

We consider the construction of efficient group sequential designs where the goal is a low expected sample size not only at the null hypothesis and the alternative (taken to be the minimal clinically meaningful effect size), but also at more optimistic anticipated effect sizes. Pre-specified Type I error rate and power requirements can be achieved both by standard group sequential tests and by more recently proposed adaptive procedures. We investigate four nested classes of designs: (A) group sequential tests with equal group sizes and stopping boundaries determined by a monomial error spending function (the 'rho-family'); (B) as A but the initial group size is allowed to be different from the others; (C) group sequential tests with arbitrary group sizes and arbitrary boundaries, fixed in advance; (D) adaptive tests-as C but at each analysis, future group sizes and critical values are updated depending on the current value of the test statistic. By examining the performance of optimal procedures within each class, we conclude that class B provides simple and efficient designs with efficiency close to that of the more complex designs of classes C and D. We provide tables and figures illustrating the performances of optimal designs within each class and defining the optimal procedures of classes A and B.

Confirmatory seamless phase II/III clinical trials with hypotheses selection at interim: opportunities and limitations.

This is a discussion of the following two papers in this special issue on adaptive designs: 'Confirmatory seamless phase II/III clinical trials with hypotheses selection at interim: General concepts' by Frank Bretz, Heinz Schmidli, Franz König, Amy Racine and Willi Maurer, and 'Confirmatory seamless phase II/III clinical trials with hypotheses selection at interim: Applications and practical considerations' by Heinz Schmidli, Frank Bretz, Amy Racine and Willi Maurer.