Principles for Codevelopment of an 1 In Vitro Companion Diagnostic

6.  Preanalytic Procedures and Testing Protocols 

646

Many IVD companion diagnostics require a number of preanalytic steps to prepare the 

647

analyte(s) for measurement (e.g., tissue fixation, DNA and RNA extraction, melanin 

648

removal, whole genome amplification, bisulfite modification).  Preanalytic reagents and 

649

instrumentation are typically considered to be part of the test system and should be validated 

650

with the IVD.   

651

652

Variations in preanalytical steps at different testing sites may make it difficult to interpret 

653

analytical performance studies.  Thus, for all steps of preanalytical specimen handling and 

654

preparation, sponsors should have a detailed standard operating procedure (SOP) or protocol 

655

that is followed at each site that performs any of the preanalytical steps.  The sponsor should 

656

ensure that all sites handling the specimens are trained to use the specific method, follow the 

657

SOPs, and record any deviations from the SOP.   

658

659

FDA bioresearch monitoring (BIMO) personnel may, and in some cases (e.g., when a PMA 

660

for an IVD is under review) generally do, examine laboratory records to determine whether 

661

protocols have been followed (see also Section III. F.1.iii. of this guidance).  In cases where 

662

there is significant and/or uncontrolled deviation from the specimen testing protocol , FDA 

663

may be unable to approve the regulatory submission because it  may deem the data derived 

664

from poorly controlled testing to be unreliable and non-representative of the IVD companion 

665

diagnostic’s performance under its proposed instructions for use. 

666

667

7.  Planning Ahead for Analytical Validation Studies 

668

The IVD sponsor should consider the types of studies needed for analytical validation to 

669

Contains Nonbinding Recommendations 

Draft - Not for Implementation 

 

 

20 

support marketing authorization of an IVD companion diagnostic and plan accordingly.

56

  

670

For example, if the analyte is labile, a plan to collect several specimens from a small 

671

number of subjects to assess lability to inform appropriate limitations on storage and 

672

transport durations may be appropriate.  Note that some analytical validation studies may 

673

not require use of samples from therapeutic product clinical trial subjects, although the 

674

studies should be conducted with samples from the same target population to ensure that 

675

the variability parameters defined are relevant to the population to be tested.  

676

677

It is important to ensure that appropriate specimens are collected and banked (where 

678

analyte stability allows) in sufficient quantities and maintained adequately to support the 

679

full range of analytical studies.  Collecting the appropriate pathologic-based annotation 

680

(e.g., tumor content, necrosis, adiposity, presence of large amounts of stroma, and other 

681

characteristics) for the samples may help to support conclusions about the performance of 

682

the assay.  Appendix 2 provides additional detail on specimen handling considerations.  

683

684

In cases where multiple markers will be detected/measured by the test, analytical validation 

685

of each reported marker may be required regardless of each marker’s prevalence.  When it is 

686

not possible for sponsors to obtain specimens containing a particular marker, validation 

687

studies with contrived samples may be permitted.

57

  Analytical validation studies may also be 

688

complicated for IVDs that have the potential to detect a very large number of markers, in 

689

which case it may be necessary for the study to use a representative sampling of markers.  

690

For example, for next generation sequencing panels, the ability of the IVD to detect single-

691

nucleotide polymorphisms, copy-number variations, inversions or deletions, and other 

692

relevant variant classes should be studied.  Sponsors who are concerned about the feasibility 

693

of conducting analytical validation studies for all markers detected by an investigational IVD 

694

should consult with FDA before beginning sample collection and analytical validation 

695

studies. 

696

D. Therapeutic Product Clinical Trial Design Considerations  

697

When planning therapeutic product clinical trials designed to rely on information 

698

provided by an IVD, whether for enrollment, stratification, dose, or other uses, sponsors 

699

should consider clinical trial designs that can be used to support the claims for both the 

700

therapeutic product and IVD companion diagnostic, and consider whether the IVD 

701

companion diagnostic development strategy is aligned with the approval goals for the 

702

therapeutic product.   

703

704

Understanding the population of subjects enrolled in a clinical trial is critical.  It is 

705

conceivable, for example, that assessment of preclinical or early clinical studies indicates a 

706

                                                 

56

 Sponsors may find it helpful to consider resources on analytical validation studies, e.g., Mansfield, E., et al. 

“Biomarkers for pharmacogenetic and pharmacogenomic studies: Locking down analytical performance.”  

Drug Discovery Today: Technologies. 2007, Vol. 4, No. I, pp. 17-01.

 

57

 For example, see FDA guidance “Guidance on Pharmacogenetic Tests and Genetic Tests for Heritable 

Markers” 

(

http://www.fda.gov/MedicalDevices/DeviceRegulationandGuidance/GuidanceDocuments/ucm077862.htm

). 

 

Contains Nonbinding Recommendations 

Draft - Not for Implementation 

 

 

21 

therapeutic product may be beneficial in the test-positive subgroup

58

 and harmful in a test-

707

negative subgroup.  In such cases, subjects with false-positive results may be harmed by the 

708

therapy, and subjects with false-negative results may be deprived of beneficial therapy.  

709

Additionally, false-positive results could lead to underestimation of effect size, whereas 

710

false-negative results could lead to underestimation of the proportion of subjects who are 

711

more likely to respond.  Therefore, the therapeutic product and IVD sponsors should work 

712

closely to understand how the IVD’s analytical performance affects the selection of subjects 

713

in the trial.  To minimize the proportion of incorrect test results (i.e., false positives and false 

714

negatives that would result in misclassification),

59

 sponsors should ensure that the 

715

appropriate analytical validation studies are carried out and that the level of analytical 

716

validation of the proposed IVD(s), in relation to its specific role in the clinical trial, has been 

717

adequately assessed.  This is especially important when progressing from the versions of the 

718

test used in a trial to the candidate IVD companion diagnostic (see Section III.E.3. of this 

719

guidance).  

720

721

Sponsors should also be aware of, and plan to address, potential sources of bias or error 

722

associated with IVD development such as prescreening, preanalytical processing 

723

(discussed in Section III.C of this guidance), and bridging studies when necessary (see 

724

Section III.E of this guidance).   

725

726

The following sections discuss considerations for the design of clinical trials for a 

727

therapeutic product for use with a developmental IVD companion diagnostic. 

728

729

1.  General Considerations for Early Therapeutic Product 

730

Development  

731

Performing tests for exploratory purposes (referred to as exploratory testing) to identify 

732

potential biomarkers in early therapeutic product development may lead to a codevelopment 

733

program.  Sponsors should be aware that using exploratory testing that is not sufficiently 

734

analytically validated or is validated with inappropriate analysis methods may produce 

735

spurious associations.

60

  This could result in the failure of a codevelopment program if, for 

736

example, a late-phase clinical trial enrolls only “marker-positive” subjects, when positivity is 

737

based on flawed exploratory programs.  When using exploratory testing, it is advisable for 

738

sponsors to establish procedures that specify the process for sample acquisition and handling 

739

                                                 

58

 Note that the terms “test-positive” and “test-negative” are often used interchangeably with the term “marker-

positive” and “marker-negative;” however, it is important to be aware that tests for the same marker that have 

different performance characteristics may identify different subpopulations of “marker-positive” patients. 

59

 For example, molecular tests that are intended to select for one target but have undetected cross-reactivity 

with other targets may result in selection of a substantial number of patients with the cross-reactive target but 

not the target of interest.  

 

60

 Sponsors should consider principles laid out in the National Cancer Institute publication, “Criteria for the use 

of omics-based predictors in clinical trials,” McShane, et al., Nature. 2013, Vol 502, pp. 317-320; and FDA 

guidance for industry “Clinical Pharmacogenomics: Premarket Evaluation in Early-Phase Clinical Studies and 

Recommendations for Labeling” 

(

http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM337169.pd

f

). 

Contains Nonbinding Recommendations 

Draft - Not for Implementation 

 

 

22 

and the testing and analysis plans so that the preliminary evidence that is generated is most 

740

likely to be informative. 

741

742

Some early therapeutic product clinical trial designs employ testing for multiple markers to 

743

assign subjects to one of multiple different therapeutic arms with the goal of testing multiple 

744

hypotheses under one study protocol.  Sponsors of these clinical trials should consider the 

745

pathway for continued development of selected therapeutic products with accompanying 

746

IVDs in the event that such trials support further development of a candidate IVD companion 

747

diagnostic. 

748

749

2.  General Considerations for Late Therapeutic Product 

750

Development 

751

When a clinical trial is properly designed to establish the safety and effectiveness of a 

752

therapeutic product in a population based on measurement or detection of a marker, the 

753

results of the clinical trial can also be used to establish the clinical validity of the IVD 

754

companion diagnostic.

61

  There are a variety of clinical trial designs that may be used to 

755

study a developmental IVD companion diagnostic in combination with a therapeutic 

756

product in premarket codevelopment programs.  The appropriate clinical trial design to 

757

support the diagnostic strategy depends on the proposed claim(s) for the IVD and what 

758

has already been established about the predictive, prognostic, or other critical properties 

759

of the marker.

62

  The success of a clinical trial design strategy depends on many factors, 

760

including but not limited to the following: a) the characteristics of the marker as applied 

761

to the target population for whom the therapeutic product will be indicated, specifically 

762

the mechanistic rationale for selecting the marker, its predictive/prognostic/other utility 

763

and its intrinsic properties (e.g., variability and specificity with respect to the disease); b) 

764

the nature of the disease; and c) the need to fully characterize the therapeutic product’s 

765

benefits and risks, such as the safety profile (e.g., taking into account a possible lack of 

766

benefit in the test-negative population), and the degree of observed benefit, if any, in the 

767

population for whom the therapeutic product may not be indicated (e.g., test-negative 

768

subjects).   

769

770

Two marker-based clinical trial designs that are commonly used are illustrated in Figure 

771

1; however, other designs could be appropriate and should be discussed with the 

772

appropriate therapeutic product review center.

63

 

773

774

                                                 

61

 For IVDs, clinical validity typically refers to the accuracy with which the test identifies, measures, or predicts 

the presence or absence of a clinical condition or predisposition in a patient.  In the case of an IVD companion 

diagnostic, clinical validity typically refers to the accuracy with which the test identifies the patients for whom 

use of the therapeutic product is safe, effective, or both. 

62

 See Section III.D.3. and Section III.G.1 for additional discussion of predictive and prognostic markers.

 

63

 For additional trial designs and further discussion, please also refer to FDA draft guidance “Enrichment 

Strategies for Clinical Trials to Support Approval of Human Drugs and Biological Products” 

(

www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM332181.pdf

). 

FDA draft guidance represents FDA’s proposed approach on this topic.  When final, this guidance will 

represent the FDA’s current thinking on this topic.

 

Contains Nonbinding Recommendations 

Draft - Not for Implementation 

 

 

23 

Figure 1.  Clinical Trials Involving Markers.  Trial design A, called an interaction or 

775

biomarker-stratified design, is designed to evaluate treatment and marker effects, and 

776

their interaction, by stratifying randomization based on marker status, as determined by 

777

an IVD.  Trial design B, called a targeted or selection design, is designed to evaluate 

778

treatment effects in a targeted population by selecting only those who are test-positive.  

779

Key: test-positive, +; test-negative, -; randomize, R.  Treatment A is typically the 

780

experimental arm and Treatment B is typically standard-of-care or placebo. 

781

782

783

In many efficacy trials, it is generally desirable to obtain information about the safety and 

784

effectiveness of the therapeutic product for all subjects (rather than for only those 

785

subjects with a particular marker status), to ascertain the appropriateness of restricting the 

786

therapy to a patient population on the basis of a marker.  However, this does not mean all 

787

subjects, regardless of marker status, should be randomized.  The study could enroll 

788

marker-positive subjects and include only a sample of marker-negative subjects, e.g., 

789

when marker-positive subjects are only a small percentage.  Testing for the presence of 

790

particular markers may provide information on prognosis, prediction of response (i.e., 

791

response, non-response, or toxicity), or both.

64

  The clinical trial design depicted in 

792

Figure 1A, in which both test-positive and at least some test-negative subjects are 

793

enrolled and randomized, is the most informative design because treatment by marker 

794

interaction, as well as the prognostic versus predictive value of the marker, can be 

795

assessed.  This approach may be particularly valuable when the biological plausibility or 

796

medical relevance of the biomarker is not well understood (e.g., based on findings from 

797

exploratory studies or post-hoc analyses in other trials).  Other variations on this design 

798

exist, such as those including interim futility analysis where, for example, further 

799

enrollment could be limited to test-positive subjects if harm or lack of efficacy is 

800

                                                 

64

 A purely predictive marker will predict that patients, given a particular marker status, will have better or 

worse outcomes than patients without the marker, solely as a result of having received the investigational 

therapeutic product; that is, there is a clear therapy-marker interaction.  A prognostic marker would suggest that 

patients with the marker would, as a consequence of the natural history of the disease, have better or worse 

outcomes even absent treatment with the investigational therapeutic product; that is, the marker has little or no 

interaction with the therapy.  Some markers may have both predictive and prognostic properties in a given 

disease/therapy setting.  For example, the presence of HER-2 protein overexpression indicates a poorer 

prognosis in patients with breast cancer than in patients who do not overexpress HER-2, but the same marker 

also predicts greater likelihood of response to the drug trastuzumab (Herceptin).  Thus, it is important to 

understand the role the marker is expected to play in the therapeutic product trial.  The prognostic value of the 

marker, if unknown at the time of the therapeutic product trial, should be assessed in clinical trials that are 

stratified by marker status.  

 

Contains Nonbinding Recommendations 

Draft - Not for Implementation 

 

 

24 

identified in the test-negative population.

65

 

801

802

In the approach depicted in Figure 1B, only a subgroup identified by the marker status is 

803

enrolled (e.g., only subjects deemed positive by the test are enrolled into the clinical 

804

trial).  With this design, the predictive value of the test cannot be determined because 

805

there is no information on the treatment effect in the test-negative population.  Likewise, 

806

there is no information about whether the assigned assay cutoff adequately distinguishes 

807

those who will respond from those who will not.  FDA does not object to this approach 

808

categorically because it may be appropriate in some situations (see also Section III.D.3 of 

809

this guidance).  A modification of the design, however, could stratify by assay cutoff.   

810

811

Sponsors planning to evaluate the safety and effectiveness of a therapeutic product only 

812

in a subset of subjects identified by an IVD should consider whether there is persuasive 

813

evidence (e.g., evidence from strong preclinical data, preliminary clinical data, or from 

814

clinical trials with similar therapeutics) for the marker as a predictive measure of 

815

response or non-response.  Although the sponsor may select any cutoff , FDA 

816

recommends that sponsors choosing a marker-positive only approach assure that the 

817

chosen marker and assigned assay cutoff are relevant to the disease under study (i.e., 

818

known prevalence of marker positivity in the general patient population) within the 

819

context of likelihood of a subpopulation’s response (e.g., biologic plausibility, 

820

mechanism of action), and that sponsors make a persuasive case for use of the IVD to 

821

identify patients who are to be treated.   

822

823

3.  Prognostic and Predictive Markers  

824

In clinical trial designs, prognostic markers can be used either to identify the population 

825

to be enrolled or to stratify treatment randomization.  For putative prognostic markers, no 

826

difference in the effect size is expected in marker-negative versus marker-positive 

827

subjects.  Effect size may be measured in different ways, depending on the clinical trial.  

828

In oncology trials with time to death as an endpoint, a hazard ratio may be used.  

829

Potential study designs for markers expected to be predictive of therapeutic response are 

830

discussed elsewhere.

66

   

831

832

With respect to a predictive marker, the clinical trial can stratify by the marker test result 

833

and randomly assign subjects with the same marker status to the experimental treatment 

834

and control (Figure 1A).  If there is little possibility of any effect in marker-negative 

835

subjects, however, only marker-positive subjects might be randomly assigned to 

836

treatment (Figure 1B), but this provides no formal test of whether the marker predicts 

837

                                                 

65

 See note 63.  Sponsors may also find it helpful to consider resources on this topic, e.g., Wang SJ, O’Neill RT, 

Hung HMJ.  “Approaches to evaluation of treatment effect in randomized clinical trials with genomic subset.” 

Pharmaceutical Statistics  Vol. 6, pp.227-244. 

 

66

 See note 63.  Additionally, sponsors may find it helpful to consider resources on clinical trial designs, e.g., 

Fridlyand, J. et al. “Considerations for the successful co-development of targeted cancer therapies and 

companion diagnostics.” Nat Rev Drug Discov. 2013. Vol. 10, pp. 743-55; Temple, R. “Enrichment of clinical 

study populations.” Clin Pharmacol Ther. 2010. 88(6), pp. 774-8. 

Contains Nonbinding Recommendations 

Draft - Not for Implementation 

 

 

25 

treatment benefits only in such marker-positive subjects.  In clinical trial designs depicted 

838

in Figure 1 above, for a continuous marker for which a firm cutoff has not been 

839

determined, there could be randomization at varying degrees of marker positivity, or less 

840

formally, there could be a post-hoc analysis of the treatment effect at a range of cutoff 

841

values.  As noted, if the marker is both prognostic and predictive, then post-hoc analyses 

842

of response by marker positivity in the clinical trial designs depicted in Figure 1A or 1B 

843

are likely to be confounded, and stratification by degree of marker positivity is strongly 

844

recommended.   

845

846