Table 1.
Types of Biomarker guided non-adaptive designs proposed within the last ten years.
| Types of Biomarker-Guided Non-Adaptive Trial Designs | Utility | Advantages | Limitations |
|---|---|---|---|
| Single arm designs (7 papers) [30,36,37,38,39,40,41] (see Figure 2) | Useful for initial identification and/or validation of a biomarker. | (A1) Considered as a simple statistical design as there is no need for randomization of patients. | (L1) There is no distinction between prognostic and predictive biomarker as patients are not randomized to experimental and control treatment arms. |
| Also called: Nonrandomized clinical trial design, Uncontrolled Cohort Pharmacogenetic Study design | (A2) Simple logistics. | ||
| Examples of actual trials: None identified a | (A3) Not complex statistical design | ||
| (A4) In some cases, these designs may be viewed as ethical as all patients are given the opportunity to experience the experimental treatment. However, they may be viewed as unethical if the novel treatment does not benefit a subgroup of patients or causes adverse events. | |||
| Enrichment designs (71 papers) [1,4,7,8,9,11,13,15,16,18,19,21,23,25,26,27,28,29,30,31,32,33,36,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86] (see Figure 3) | Useful when we aim to test the treatment effect only in biomarker-positive subset for which there is prior evidence that the novel treatment is beneficial, but the candidate biomarker requires prospective validation. | (A5) Evaluates the effect of the experimental treatment in the biomarker-positive subgroup in a simple and efficient way. | (L2) Do not assess whether the experimental treatment benefits the biomarker-negative patients, thus we cannot obtain information about this subgroup. Also unable to demonstrate whether the targeted treatment is beneficial in the entire study population. |
| Also called: Targeted design, Selection design, Efficient Targeted design, Biomarker-Enrichment design, Marker-enrichment design, Gene enrichment design, Enriched design, Clinically enriched Phase III study design, Clinically Enriched Trial design, Biomarker-Enriched design, Biomarker Enriched design, Biomarker Selected trial design, Screening enrichment design, Randomized Controlled Trial (RCT) of test positive design, Population enrichment design | Useful when it is not ethical to assign biomarker-negative patients to the novel treatment for which there is prior evidence that it will not be beneficial for this subpopulation, or that it will harm them. | (A6) Provides clear information about whether the novel treatment is effective for the biomarker-positive subgroup, thus these designs can identify the best treatment for these patients and confirm the usefulness of the biomarker. | (L3) Do not inform us directly about whether the biomarker is itself predictive because the relative treatment efficacy may be the same in the unevaluated biomarker-negative patients. Since these designs only enrol a subgroup of patients, they do not allow for full validation of the marker’s predictive ability. For full validation, a trial would need to randomize all patients in order to test for a treatment–biomarker interaction. |
| Examples of actual trials: CRYSTAL [49], BRIM 3 [49,50,51], EURTAC [49], CLEOPATRA [49], PROFILE 1007 [49,50], LUX-Lung [49], NSABP B-31 and NCCTG N9831 [4,15,16,18,19,28,29,30,31,36,44,46,52,53,54,55,56,57,58,59,60], CALGB-10603 [61], CATNON [62], CODEL [62], Evaluation of epidermal growth factor receptor variant III (EGFRvIII) peptide vaccination [62], N0923 [7,21] , Flex study [64], TOGA trial [47], IPASS [33,43], N0147 [29], PetaCC-8 [29,47], C80405 [29], ECOG E5202 [29] | Recommended when both the cut-off point for determination of biomarker-status of patients and the analytical validity of a biomarker are well established. | (A7) Reduced sample size as the assessment of treatment effect is restricted only to biomarker-positive subgroup. Therefore, if the selected biomarker is “biologically correct” and reliably measured, the used enrichment strategy could result in a large saving of randomized patients. | (L4) Researchers should carefully decide whether or not to follow this strategy as it may be of limited value due to the exclusion of biomarker-negative patients. It may be that the entire population could benefit from the experimental treatment equally irrespective of biomarker status, in which case enrolling only the biomarker-positive patients will result in slow trial accrual, increase of expenses and unnecessary limitation of the size of the indicated patient population. |
| (A8) Enables rapid accumulation of efficacy data. | (L5) Concern over an ethical problem as we cannot include individuals in a clinical trial if it is believed that the treatment is not effective for them, as raised by the US Food and Drug Administration (FDA) [50]. It was based on the facts that the experimental treatment can only be approved for a particular biomarker-defined subpopulation (i.e., biomarker-positive patients) if a companion diagnostic test is also approved, and how the test can be approved if the Phase III trial does not show that the novel treatment does not benefit the biomarker-negative patients. | ||
| (A9) Allow us to avoid potential dilution of the results due to the absence of biomarker-negative patients. For example, if the design had included the biomarker-negative population and the biomarker positivity rate was low as compared to the biomarker negative rate, then the estimation of the overall treatment effectiveness could be diluted as it would be driven by the biomarker-negative subset. | (L6) The accuracy of diagnostic devices used to identify the biomarkers, e.g., biomarker assays, is not always correct [45]. This can result in incorrect selection of biomarker-positive patients and therefore these patients will erroneously be enrolled in a trial yielding biased treatment effect estimates. For example, even when the experimental treatment works well for a specific subgroup, if the biomarker assay is not able to identify this subgroup robustly then a promising treatment may be abandoned. | ||
| (A10) Can be attractive in terms of speed and cost, meaning that patients are provided with tailored treatment sooner. | |||
| Marker Stratified designs (45 papers) [4,10,12,13,15,16,17,18,19,21,25,26,27,30,31,33,44,45,46,49,50,51,53,58,61,62,66,68,71,72,73,74,79,80,81,84,85,86,87,88,89,90,91,92,93] (see Figure 4) | Useful when there is evidence that the novel treatment is more effective in the positive biomarker-defined subgroup than in the negative biomarker-defined subgroup but there is insufficient compelling data indicating that the experimental treatment does not benefit the biomarker-negative patients. | (A11) Ability to assess the treatment effect not only in the entire population but also in each biomarker-defined subgroup. Thus, this design can find the optimal treatment in the entire population and in each biomarker-defined subgroup. | (L7) In situations where there are several biomarkers and treatments this design may not be feasible as it involves randomization of patients between all possible treatment options and may require a large sample size. |
| Also called: Marker-stratified design, Biomarker-stratified design, Stratified-Randomized design, Stratification design, Stratified design, Stratified Analysis design, Marker by treatment – interaction design, Marker-by-treatment interaction design, Treatment by marker interaction design, Treatment-by-marker interaction design, Marker × treatment interaction design, Treatment-marker interaction design, Biomarker-by-treatment interaction design, Non-targeted RCT (stratified by marker) design, Genomic Signature stratified designs, Signature-Stratified design, Randomization or analysis stratified by biomarker status design, marker-interaction design. | (A12) An ethical design even in situations where the biomarker is not useful as no treatment decisions are made based on biomarker status; all decisions are made randomly. Consequently, if the biomarker’s value is in doubt, this design may be preferred. | (L8) May not be feasible when the prevalence of the biomarker is low. | |
| Examples of actual trials: MARVEL (N023) [4,16,30,31,33,44,61,89], GALGB-30506 [15,61], RTOG0825 [45], EORTC 10994 p53 [12,66], IBCSG trial IX [18], MINDACT [18] | (L9) Might be expensive to test the entire population for its biomarker status. | ||
| (L10) Measuring the biomarker up front may be logistically difficult. | |||
| (L11) There is no guarantee of balanced groups for analysis. | |||
| Sequential Subgroup-Specific design (11 papers) [13,14,19,22,53,57,58,60,69,91,94] (see Figure 5) | Recommended when prior evidence indicates that the biomarker-positive subpopulation benefits more from the novel treatment as compared to the biomarker-negative subpopulation. | (A13) Allows for the estimation of treatment effect in biomarker-positive and biomarker-negative subgroups. | (L12) Has less power when there is homogeneity of treatment across the different biomarker defined subgroups as compared to the overall/biomarker-positive designs. |
| Also called: sequential design, Fixed-sequence 2 design, hierarchical fixed sequence testing procedure | (A14) Preserves the overall type I error rates and allows for a smaller sample size than the parallel version mentioned below. | (L13) Need a much larger sample size than the overall/biomarker positive designs if we assume that the treatment effect is relatively homogeneous across the biomarker-defined subsets. | |
| Examples of actual trials: PRIME [49], MARVEL [49] | (A15) Considered as the best direct evidence for clinical decision making as it tests the treatment effectiveness in both the biomarker-positive and biomarker-negative subset in a sequential way. | ||
| (A16) Do not require larger sample size than the overall/biomarker-positive designs when the prevalence of the biomarker-positive patients is small. | |||
| Parallel Subgroup-Specific design (3 papers) [14,49,69] (see Figure 6) | Appropriate when the aim of the study is to give treatment recommendations for each biomarker-defined subgroup separately at the same time. | (A17) Same as (A13), (A16) | (L14) Same as (L12) |
| Also called: Phase III Biomarker-Stratified design | (L15) Allocates the overall level between the two biomarker-defined subgroup tests which means that it will be more difficult to achieve statistical significance in the biomarker-positive subgroup. | ||
| Examples of actual trials: None identified a | |||
| Biomarker-positive and overall strategies with parallel assessment (8 papers) [1,14,36,47,49,69,95,96] (see Figure 7) | Recommended when the aim of the study is to assess the treatment effect in both the entire population and in the biomarker-positive subset but not in the biomarker-negative population. | (A18) Can control the overall type I error . | (L16) Can be overly conservative as in the SATURN trial because of the correlation between the test of treatment effect in the overall study population and in the biomarker subgroups. |
| Also called: Overall/biomarker-positive design with parallel assessment, prospective subset design, hybrid design | (A19) Can require smaller sample size as compared to the subgroup-specific designs, especially when we assume that the novel treatment equally benefits both biomarker-defined subgroups. | (L17) Cannot control the probability of rejecting the null hypothesis of no treatment effect in the biomarker-negative subset when the treatment benefit is restricted to biomarker-positive patients. Consequently, there is a high risk of inappropriately recommending the novel treatment for biomarker-negative patients due to the large treatment effect in biomarker-positive subset. | |
| Examples of actual trials: S0819 [14,49], SATURN [14,36,47,49,95,96], MONET1 [14,49], ARCHER [14,49], ZODIAC [49], MERiDiAN [49] | |||
| Biomarker-positive and overall strategies with sequential assessment (11 papers) [13,14,30,44,49,69,80,84,85,88,94] (see Figure 8) | Might be useful in cases where the experimental treatment is expected to be effective in the overall population. | (A20) Same as (A18), (A19) | (L18) Can be problematic for determining whether the treatment is beneficial in the biomarker-negative subgroup. |
| Also called: Overall/biomarker-positive design with sequential assessment, sequential design, Fixed-sequence 2 design, hierarchical fixed sequence testing procedure | (L19) Same as (L17) | ||
| Examples of actual trials: Trial of letrozole plus lapatinib versus letrozole plus placebo in breast cancer, with the biomarker defined by human epidermal growth factor receptor 2 (HER2) [14], N0147 [30,49] | |||
| Biomarker-positive and overall strategies with fall-back analysis (15 papers) [10,30,36,44,47,49,53,57,60,69,84,88,94,96,97] (see Figure 9) | Recommended when there is insufficient confidence in the predictive value of the biomarker and the novel treatment is assumed to probably benefit all patients. | (A21) Can assess the treatment effect in the biomarker-positive patients, if no benefit is detected in the overall population. | (L20) Same as (L17), (L18) |
| Also called: Biomarker-stratified design with fall-back analysis, fall-back design, prospective subset design, sequential design, other analysis plan design, Fallback design | (A22) Same as (A18), (A19) | ||
| Examples of actual trials: None identified a | |||
| Marker Sequential test design (4 papers) [14,49,69,94] (see Figure 10) | Recommended when biomarkers with strong credentials are available and we have convincing evidence that the novel treatment is more effective in biomarker-positive than in biomarker-negative patients. | (A23) Can provide clear evidence of treatment benefit in the biomarker-positive subgroup and in the biomarker-negative subgroup. | (L21) In situations where biomarker status is not available for some of the patients included in the study, this design can either exclude these patients or include them in the global test, however, further statistical adjustments might be required in that case. |
| Also called: MaST design, hybrid design | Appropriate when we can assume that the treatment will not be beneficial in the biomarker-negative subpopulation unless it is effective for the biomarker-positive subpopulation. | (A24) Enables sequential testing of the treatment effect in the entire study population and in the biomarker-defined subgroups to restrict testing of the treatment effect in the entire population when there is no significant result in the biomarker-positive subset, while controlling the appropriate type I error rates. | (L22) Does not decrease the sample size of the study as it was developed in order to increase the power compared to the sequential subgroup-specific design in situations where the novel treatment benefits equally both biomarker-negative and biomarker-positive patients. |
| Examples of actual trials: ECOG E1910 [14,49] | (A25) Results in higher power as compared to the sequential subgroup-specific design in cases where the treatment effect is homogeneous across the biomarker-defined subgroups. | ||
| (A26) Preserves the power in situations where the treatment effect is restricted only to the biomarker-positive patients and at the same time it controls the relevant type I error rates. | |||
| (A27) Control the type I error rate for the biomarker-negative subgroup over all possible prevalence values. | |||
| (A28) The probability of erroneously concluding that the novel treatment is beneficial for the entire population when the global effect is driven by the biomarker-positive patients is minimized since the design only tests the treatment effect in the entire population when no significant effect is detected in the biomarker-positive subgroup. | |||
| Hybrid designs (14 papers) [1,13,15,29,30,31,36,46,48,55,66,84,88,98] (see Figure 11) | Can be used when there is prior evidence indicating that only a particular treatment is beneficial to a biomarker-defined subgroup which makes it unethical to randomize patients with that specific biomarker status to other treatment options. | (A29) The feasibility of a prognostic biomarker can be tested. | None found. |
| Also called: Mixture design, Combination of trial designs, hybrid biomarker design | (A30) Allows for better risk assessment and improved individualized treatment since it assigns patients to treatments based on risk assessment scores instead of their biomarker status (biomarker-positive and biomarker-negative patients). | ||
| Examples of actual trials: TAILORx [15,48,55,58,63,66], EORTC MINDACT [15,48,55,66], ECOG 5202 study [30,46] | |||
| Biomarker-strategy designs with biomarker assessment in the control arm (21 papers) [15,25,26,32,33,36,45,61,62,64,79,82,85,86,92,93,99,100,101,102,103] (see Figure 12) | Useful when we want to test the hypothesis that the treatment effect based on the personalized approach is superior to that of the standard of care. | (A31) Biomarker can be validated without including all possible biomarker–treatment combinations [26] as in the non-biomarker-based arm all patients receive only the control treatment. | (L23) Unable to inform us whether the biomarker is predictive as these designs are able to answer the question about whether the biomarker-based strategy is more effective than standard treatment, irrespective of the biomarker status of the study population. |
| Also called: Marker strategy design, Biomarker-strategy design, Strategy design, Marker-based strategy design, Marker-based design, Random disclosure design, Customized strategy design, Parallel controlled pharmacogenetic study design, Marker-based strategy design I, Biomarker-guided design, Biomarker-based assignment of specific drug therapy design, Marker-based strategy I design, Biomarker-strategy design with a standard control, Marker strategy design for prognostic biomarkers | (A32) Have the option of testing the biomarker status of patients in the non-biomarker-strategy arm which can aid secondary analyses [26]. | (L24) The evaluation of the true biomarker by treatment effect is not possible as the biomarker-positive patients receive only the experimental treatment and not the alternative treatment (control treatment). Consequently, this design cannot detect the case in which the control treatment might be more beneficial for the entire population. | |
| Examples of actual trials: GILT docetaxel [15], Randomized phase III trial conducted in Spain, dedicated to patients with advanced Non-Small Cell Lung Cancer (NSCLC) candidates for first-line chemotherapy [32,64,100], Study the effect of Magnetic Resonance Imaging (MRI) in patients with low back pain on patient outcome and to evaluate Doppler US of the umbilical artery in the management of women with intrauterine growth retardation (IUGR), Randomized controlled trial in recurrent platinum-resistant ovarian carcinoma [101] | (A33) Able to inform us whether the biomarker is prognostic. | (L25) In case that the number of biomarker-positive patients is very small, then the treatment received will be similar in biomarker-strategy arm and non-biomarker strategy arm. Consequently, the trial might give little information regarding the efficacy of the experimental treatment or it might not be able to detect it. As a result, this type of design should be used when there is an adequate number of biomarker-positive and biomarker-negative patients. | |
| (A34) Can be expanded to investigate several biomarkers and treatments [103]. Additionally, these designs can be attractive when evaluating multiple biomarkers or the predictive value of molecular profiling between several treatment options is to be assessed [45]. | (L26) Unable to compare directly experimental treatment to control treatment as the aim is to compare not the treatments but the biomarker-strategies. | ||
| (A35) Might be used more frequently in the future due to the wide variety of molecular biomarkers, complexity of gene expression arrays, and several treatments directed at similar targets [103]. | (L27) Less efficient designs than biomarker-stratified designs [4,73] and a poor substitute for clinical trials which aim to compare the experimental treatment to control treatment, since it is possible for some patients in both the biomarker-based strategy arm and non-biomarker-based strategy arm to be assigned to the same treatment (due to the existence of biomarker-negative patients in both strategy arms the treatment effect can be diluted) [51]. Consequently, as a large overlap of patients receiving the same treatment might have occurred, the comparison of the two biomarker-strategy arms results in a hazard ratio which is forced towards unity, i.e., no treatment effect exists as the effect of experimental versus control treatment is diluted by the biomarker-based treatment selection. For this reason, a large sample size is needed to detect at least a small overall difference in outcomes between the two biomarker-strategy arms. | ||
| (L28) Should be used only if you want to evaluate a complex biomarker-guided strategy with a variety of treatment options or biomarker categories [73]. | |||
| Biomarker-strategy design without biomarker assessment in the control arm (14 papers) [9,13,17,18,20,25,36,38,61,74,101,104,105,106] (see Figure 13) | In situations where it is not feasible or unethical to test the biomarker in the entire population. | (A36) Galanis et al., 2011 [45] stated that these designs can be attractive when evaluating multiple biomarkers or the predictive value of molecular profiling between several treatment options is to be assessed. Also, Freidlin and Korn, 2010 [73] claimed that these biomarker-strategy designs should be used only if researchers want to evaluate a complex biomarker-guided strategy with a variety of treatment options or biomarker categories. | (L29) Criticized for their potential cost increase due to the fact that patients without predicted responsive biomarker are double enrolled in the trial (biomarker-negative patients receive control treatment in both strategy arms). |
| Also called: Biomarker-strategy design with standard control, Direct-predictive biomarker-based, RCT of testing, Test-treatment, Parallel controlled pharmacogenetic diagnostic study, Marker strategy, Marker-based with no randomization in the non-marker-based arm, Classical, Marker-based strategy, Marker strategy design for prognostic biomarkers | (A37) Same as (A31), (A32), (A33) | (L30) Biomarker-positive and biomarker-negative subpopulations might be more imbalanced as compared with the first type of biomarker-strategy design due to the fact that the randomization to different treatment strategies is performed before the evaluation of the biomarker status (balancing the randomization is useful to ensure that all randomized patients have tissue available). This can happen especially when the number of patients is very small. | |
| Examples of actual trials: A study, which evaluated the use of immediate computed tomography in patients with acute mild head injury [101,104]. | (L31) Same as (L23), (L24), (L25), (L26), (L27) | ||
| Biomarker-strategy design with treatment randomization in the control arm (17 papers) [15,17,26,27,32,36,45,62,64,66,74,86,92,93,106,107,108] (see Figure 14) | In cases where we want to know whether the biomarker is not only prognostic but also predictive, these designs are preferable as compared to the two previously mentioned biomarker-strategy designs. | (A38) These designs have the ability to inform researchers about the potential superiority of the control treatment in the whole population or among a particular biomarker-defined subpopulation. | (L32) Generally require a larger sample size as compared to the marker-stratified designs. |
| Also called: Biomarker-strategy design with a randomized control, Modified marker-based strategy design (for predictive biomarkers), Biomarker-strategy design with randomized control, Marker-based design with randomization in the non-marker-based arm, Marker-based strategy design II, Marker-strategy design, Augmented strategy design, Trial design allowing the evaluation of both the treatment and the marker effect | (A39) Able to inform us whether the biomarker is prognostic or predictive. | (L33) Same as (L27) | |
| Examples of actual trials: None identified a | (A40) Allow clarification of whether the results which indicate efficacy of the biomarker-directed approach to treatment are caused due to a true effect of the biomarker status or to an improved treatment irrespective of the biomarker status. | ||
| (A41) Same as (A36) | |||
| Reverse marker-based strategy (4 papers) [86,92,93,109] (see Figure 15) | Enables testing the interaction hypothesis of treatment and biomarker in a more efficient way as compared to the first (i.e., Biomarker-strategy design with biomarker assessment in the control arm) and third biomarker-strategy subtype design (i.e., Biomarker-strategy design with randomization in the control arm and the marker stratified design) | (A42) Can estimate directly the marker-strategy response rate. | (L34) It has been claimed by Baker, 2014 [93] that other designs than the reverse marker-based strategy are more appropriate in order to investigate questions which include both treatment effect of biomarker-defined subgroups and the biomarker strategy treatment effect. These designs should allow the estimation of treatment effects within biomarker-defined subgroups as well as the estimation of the global treatment effect. |
| Also called: None found | (A43) Allows the estimation of the effect size of the experimental treatment compared to the control treatment for each biomarker-defined subset separately. | ||
| Examples of actual trials: None identified a | (A44) There is no chance that the same treatment will be tailored to biomarker-positive patients who are randomized either to the biomarker-based strategy arm or the reverse marker strategy. Also, there is no possibility of the same treatment assignment to biomarker-negative patients who are randomly assigned to the two biomarker-based strategy arms. | ||
| (A45) It has been demonstrated by Eng, 2014 [92] that this new type of design is more than four times more efficient for testing the interaction between treatment and biomarker compared to Biomarker-strategy design with biomarker assessment in the control arm, Biomarker-strategy design with randomization in the control arm and the marker stratified design. | |||
| A specific randomized phase II trial design that can be used to guide decision making for further development of an experimental therapy. (1 paper) [71] (see Figure 16) | Recommended when we want to conduct a Phase II randomized trial which allows decisions to be made about which type of Phase III biomarker-guided trial should be used. | (A46) Works well in providing recommendations for phase III trial design. | None found |
a Although not found within the review, the design may be implemented in ongoing trials.