Table 1. Invited Peer Reviewer Comments

Disposition of Comments

Comments received from Systematic Reviews on Selected Pharmacogenetic Tests for Cancer Treatment.

Disposition of Comments

Table 1. Invited Peer Reviewer Comments

Reviewer1Section2Reviewer CommentsAuthor Response3
Peer Reviewer 1CYP2D6The authors do a nice job of collecting relevant studies which analyze the association between cyp2d6 status and outcomes on tamoxifen in the adjuvant breast cancer setting (and the one study in the metastatic setting). As they point out, these studies are difficult to compare and combine because they used varying definitions of �extensive,� �intermediate� and �poor� metabolizing alleles. Moreover the analyses compare different combinations of these groups. This topic is further complicated by the use of SSRIs in many of these women which can change their metabolizer status. Many of the studies reported did not control for the use of SSRIs in their analyses.We thank the reviewer for the kind comments. We agree that there is irreconcilable heterogeneity between studies and definitions used; this is the main reason we refrained from quantitative analyses.
While the authors do present the problems associated with variable definitions in these studies, their conclusions are not helpful. While the evidence for routine testing for metabolizer status may not yet be warranted, the question is far from fully vetted. Tamoxifen is one of the most widely-used drugs for treatment of breast cancer, and the question of efficacy modulation is still an important one. It is compelling that many of these studies across different ethnicities did have positive associations. And instead of dismissing these studies, a proposal to standardize definitions of metabolizer status could be added to the conclusion. The role of chance versus true associations should also be further discussed.

The TA summarizes the evidence on existing studies and reaches conclusions based on the available evidence. We agree that standardization of definitions, further studies and individual-patient data meta-analyses would benefit the CYP2D6 field.

Regarding false positive associations (type I error) we have added the following comment:

�First, most studies are relatively small and thus probably underpowered to detect what would be a plausible effect size for modification of response to tamoxifen and susceptible to type I error (false positive findings).�

The authors mention that no formal interaction analysis was performed using women who were not treated with tamoxifen. The Schroth et al. publication did report the lack of association among women who did not receive tamoxifen. This finding should be added on page 23.

The study by Scroth did not perform (or at least did not report) and interaction test comparing the pharmacogenetic effect in tamoxifen treated and non-treated women. Results for the no tamoxifen arm are not presented, precluding us from conducting the interaction analysis.

We have added a note that �Differences were not observed in the control group�.

Peer Reviewer 2GeneralIn general, I found the reports to be clear, well organized, and appropriate in scope and methods.We thank the Peer Reviewer for his kind comments.
Peer Reviewer 2SummaryIn the second paragraph of the background section in the summary and on page 2, paragraph 1 of the main text, reference is made to the Medicare beneficiary population. Consideration of PICO framework � (P) patient population; (I) intervention; (C) comparison: (O) outcome _ might be helpful � I, C and O components seem fine.[no response needed]
Peer Reviewer 2CYP2D6Search was limited to MEDLINE until 24 August 2009. There appear to be a few more relevant articles since that time, for example, for CYP2D6, tamoxifen and breast cancer, which I found using the HuGENet Published Literature database, limited to pharmacogenomics, mammary neoplasms and CYP2D6. What was the rationale for not considering other databases such as EMBASE? I accept that these tend not to add much for genetic association studies compared with Medline/PubMed, but in the context of cancer therapy, it is possible that evaluations of treatment outcomes done by organizations such as EORTC, in which genotyping might have been performed secondarily, might be published in journals indexed in EMBASE and not in MEDLINE.We have updated the search strategies and data extraction to include data up to March, 2010 for the CYP2D6 and the KRAS systematic reviews. Regarding the EMBASE database, we agree that for biomarkers and genetic association studies EMBASE may not substantially increase the sensitivity of searches while reducing specificity.
Peer Reviewer 2GeneralI was a bit unsure of the expected audience. Some of the information that is specific to cancer therapeutics might be unfamiliar to a group that was charged with more generic decision making about health technologies, while some that is specific to genetic testing might be unfamiliar to health professions handling cancer treatment decisions (e.g. on page 2 of main text, final paragraph, there is the clarification that an enzyme is a protein, but not of �chimeric oncogene�.

We agree that the understanding of pharmacogenetic tests will require knowledge from diverse fields including clinical epidemiology, genetics and medicine.

We have provided a �plain words� explanation of �chimeric oncogene�.

Peer Reviewer 2Summary

In the summary and the introduction, it is stated

�The challenges in the integration of cancer pharmacogenetics and targeted therapies in clinical practice require proof of benefit to the healthcare system, incorporating patient preferences, improving provider education, and anticipating potential ethical and social implications.�

I agree, but it reads as if benefit to the patient is not a consideration! I also suggest referring to balance between benefits, harms and affordability, especially in contexts of ageing populations facing escalating health costs, and in some subsets at least, drift towards more interventions.

We have revised the sentence to read:

�The challenges in the integration of cancer pharmacogenetics and targeted therapies in clinical practice require proof of benefit to the patients (a favorable balance of harms and benefits of testing), cost-effectiveness for the healthcare system, incorporating patient preferences, improving provider education, and anticipating potential ethical and social implications.�

Table S1 � it would be useful to have not only numbers of studies, but also indication of volume of evidence (i.e. number of patients overall and/or in smallest subgroup)Given the extensive overlap between studies these estimates would not be representative of the true amount of evidence and may be .
Page S-5 � second bullet point under �study design issues� _ Agree with point about �repurposing� already completed RCTs. What about use of case-only design embedded in such a repurposed RCT?In epidemiological studies of disease causation the case-only design, which requires only diseased subjects, allows for estimation of multiplicative interactions between factors known to be independent in the study population. This design requires assumptions to be made, namely an independence assumption between the environmental factors and the genetic marker. When this assumption holds, interactions can be assessed based only on affected individuals. In most of the studies we reviewed all individuals were �cases�, i.e. treated. One �equivalent� of the case-only design in the �repurposed RCT� framework would be to only consider the �treated patients� and assess interactions with the treatment effect. As we extensively discuss in the Report, the majority of studies made this assumption and did not utilize any �controls�, i.e. individuals not receiving the treatment of choice.
Page 2 � it is stated that tamoxifen metabolites are �biotransformed through a complicated metabolic pathway, in which CYP2D6 is a leading enzyme�. I did not find this clear. In what sense is CYP2D6 �leading�?

We agree that the term �leading� may be inappropriate for metabolic pathways. We have rephrased the sentence to read:

�They are biotransformed through a complicated metabolic pathway, in which CYP2D6 is a key enzyme�.

Sentence starting at bottom of page 2: �Therefore there may be pharmacogenetic associations of mutations of the BCR-ABL1 gene with potential impact on management decisions.(4)� This was unclear.

We have revised the sentence to read:

�Based of these observations, the detection of mutations of the BCR-ABL1 gene has been proposed as a pharmacogenetic test with potential impact on management decisions.�

Page 3, key question 2: As �gender� appears to refer to biological aspects of treatment response, I think the appropriate term is �sex�We have used �sex� as a more appropriate term.
Page 3: example under key question 3 unclearSeems ok to me. Are we allowed to change the KQs?
Peer Reviewer 2CYP2D6Page 9, para 2: are aromatase inhibitors ever given concurrently with tamoxifen?To the best of our knowledge co-administration of tamoxifen and aromatase inhibitors is not an approved use of the drugs.
Para 3, last line � what is point about �(rare)� after mutations?We have revised the sentence to read: �Tamoxifen resistance has been extensively investigated and a variety of biological mechanisms are considered as potentially mediating treatment resistance, including cross talk of the ER/PR-activated pathway and growth-factor signaling pathways, activation of alternative (non-ER-dependent) signaling pathways, loss of ER expression and ER mutations (a rare cause of resistance).�
Page 12: last para � I understand the reluctance to perform quantitative mata-analysis. However, could this not have been done using the �simple� algorithm described in the last paragraph of the previous page, and then the effect of dropping each study in turn done as an influence analysis?Given the heterogeneity of outcomes reported, genotypes investigated, and results reported performing a meta-analysis of the set of CYP2D6-relevant studies would not be a valid approach.
Page 13, Fig 2: what does asterisk after �irrelevant� in bottom left box refer to?

There is a note at the bottom of the graph explaining the term �irrelevant�. The note reads:

* �Irrelevant� includes publications with no primary data, studies on healthy population, and studies on medications that inhibit CYP2D6.

Last paragraph - Avoid use of term �Caucasian� See Wikipedia on �Caucasian�, and that it has generated a lot of debate. Also Bhopal R, Donaldson L. Am J Pub Hlth 1998; 88: 1303-7; Ma IW et al. Journal of Clinical Epidemiology 60 (2007) 572e578; Comstock RD et al. Am J Epidemiol 2004;159:611�619.We have used the term �White� instead.
Page 14, line 1 � was there really no information about dosing in the RCTs?We have qualified this statement to read: �Tamoxifen dosing was not reported in the majority of studies�.
Last paragraph � I think it would be helpful to split off the study in the metastatic cancer setting � this was both the smallest study and had the shortest follow-up.For the descriptive characteristics and genotyping methods and results, the data items we extracted from the metastatic setting study were the same as for the adjuvant setting studies. We agree with the Peer Reviewer that outcomes are different in the two settings and as such the outcomes tables are separate.
Page 15, Table 1 � Has overlap between the two Goetz studies been excluded? Could information be given TAM vs non_TAM for these?There is substantial overlap between the two studies. For the CYP2D6 part of the report, given that no quantitative analysis was performed, we included all studies in the summary tables. We have made this explicit in the Methods and Results sections.
Page 16, col 2, rows 2-4, what is �RCS�?There is a footnote at the bottom of the table spelling out these initials: Non-RCS = non-randomized comparative study.
Page 18 � further to comment on page 12, I accept that inferring metabolizer phenotype on basis of genotype complex, but would it not have been possible also to do meta-analysis of *4/*4 vs. wt/wt (5 studies), 10/10 vs wt/wt (3 studies) and 41/41 vs wt/wt (also three studies), or are you arguing that this is pointless? (I accept that wt depends on what is tested for, but for 4/4, looks uniform apart from Schroth and Newman studies)Given the heterogeneity of outcomes reported, genotypes investigated, and results reported performing a meta-analysis of the set of CYP2D6-relevant studies would not be a valid approach. We also suspect there is substantial potential for reporting bias. As such, any meta-analysis would be prone to identify spurious associations.
Page 19, Table 2: testing for departure from Hardy-Weinberg equilibrium mentioned without explanation. If variation in CYP2D6 of etiologic importance in breast cancer, would equilibrium really be expected? (in case-control studies, often done in controls to indicate whether gross problem with genotyping, population stratification, other selection bias, but departure capable of many interpretations)The review did not evaluate CYP2D6 variants as a causative factor for breast cancer. We evaluated these markers as predictors of response to treatment. It is true that if CYP2D6 polymorphisms were also associated with breast cancer development departure from HWE could occur. On the other hand, if the variants are not causally associated with breast cancer risk, then departure from HWE could indicate the problems highlighted from the Peer Reviewer or biased sampling from the study base.
Many footnotes incomplete in this table. Footnote a was unclear � what was the p value a test of?Footnote was unclear and was removed.
Page 20, Kiyotani study � col 3 indicates available sample 67, genotyping success 100%, but total number in last column adds up to 58. Other totals seem to add up.

We have added the following footnote to explain the apparent discrepancy:

�Other genotypes [CYP2D6 *1/*4 (2%), *1/*5 (6%), *5/*10 (3%), *5/*41 (2), and *10/*21 (2%)] were also detected but not used in analyses because of their low frequency�

Footnote a � is this irrespective of genotype?Cannot locate this comment
Pages 22-24: It would be helpful to comment on effects of adjustment in Tables 3-4, in view of comments on need for adjustment (Mendelian randomization principle) later.

We have added the following two comments in the relevant sections:

�Many of the studies presented regression-adjusted estimates of the effect of CYP2D6 genotype on mortality risk, frequently for factors that could not confound the genotype-response association.�

�Many of the studies presented regression-adjusted estimates of the effect of CYP2D6 genotype on disease recurrence risk, frequently for factors that could not confound the genotype-response association.�

Page 27, para 1, last line � �preventive setting� confusing.This is statement verbatim from the American Society of Clinical Oncology and refers to the use of tamoxifen for the prevention of breast cancer occurrence in women at high risk of the disease.
Page 28 � is an issue the size of the study � if small, might there be departure from Mendelian randomization, just as can be inequaklities in co-variate diostributions between the arms of small RCTS?�Mendelian randomization� is general principle governing segregation of alleles at miosis. Random fluctuations away from the proportions expected under Hardy Weinberg equilibrium, a related problem, is indeed a concern. Yet it can be argued than when such problems are due to small study sample sizes, adjustment for covariates will also be problematic due to sparse data.
Peer Reviewer 2KRASIn this section, numerous times spaces between words missing, and more typos than in previous sections. Needs thorough proof reading.We have reviewed the section and attempted to correct all typographical errors.
Page 30, penultimate line � specify the tissue in which KRAS testing done � also applies to key question 1, page 31Tables 10 and 11 report the tissue and type of material (fresh-frozen versus paraffin-embedded) used for DNA isolation. We have also collected information on whether metastatic or primary tumor foci were used for tissue collection and � from the few studies that reported relevant information � we have also collected whether primary and metastatic foci examination leads to the same mutational analysis results.
Page 34, line 3 � suggest changing �pre-treated� to �with metastatic disease who had previously been treated�The suggested change was implemented.
Line 4 � �these� refers to what?

The sentence has been clarified. It now reads:

�In studies conducted in the metastatic setting, the majority of patients had received prior treatment with at least one chemotherapy regimen; both the number and types of treatment regimens administered varied across studies�

Penultimate line before Table 7: specify drug dosing � may have changed by time the report is read

This sentence was rephrased to read:
�Given that many of the patients in these studies were participants in larger, multicenter clinical trials, drug dosing in the studies included in this report can be expected to be similar to that employed in the prospective trials.�

Page 36, Fig 5: nice, but text in ellipses virtually impossible to readWe have generated a new figure given the large number of studies captured by our search updata. The new figure is of much higher resolution.
Table 9: I like the way the table is ordered by study design. However, within the single arm studies, not sure of what order reflects. I thought it might be descending order of sample size, but last study in table larger than the previous six, so I�m flummoxed.We have maintained the table structure based on design. All tables have been rearranged by year of publication. Studies with the same publication year are arranged by author name. In the rare case where year and author name are the same, we arranged studies by decreasing sample size.
Page 48, Table 10. After the three RCTs, I think a heading �single arm� and then Bengala reference missing.

We have corrected the Table. In addition, after obtaining author confirmation, we have re-classified the studies by Yen et al as first line studies. The studies did not report le specific line of treatment but Dr Wang, a corresponding author in one and author on both kindly provided this information.

[Personal communication, Professor Jaw-Yuan Wang, MD, PhD
Department of Surgery
Kaohsiung Medical University and Hospital
100 Tzyou 1st Road
Kaohsiung 807, Taiwan]

Based on this re-classification, we have re-arranged the tables and repeated the meta-analyses.

More generally in Table 10, need (ref) after name and year, as in Table 9. This comment also applies to other tables in this section.For all sections we have applied the policy of adding references after study authors only on the study characteristics Tables (Tables 1, 8 and 9, 20, 23 and 30-32). In all other tables the order of tables follows the order of the descriptive ones.
Page 54. Description of results of studies in first and second line studies mixed up, particularly for RCTs. Was difference in effect of treatment by presence of KRAS mutation in the same direction in the van Custem study as it was in the other?

The relevant quantitative analysis sections have now been updated to include the new studies identified by our search update.

We have added the following comment:
�The interaction test was non-significant in the study by van Cutsem 2009 (p=0.44) but the direction of effects was consistent.�

Pages 55-57: I appreciate need to report �null results� etc, but could there be a better way to signal this than entries in the tables where everything except author, year and study arm is �NR�?We can remove all the NR, NR (but I would prefer not).
Pages 61-64, Table 15: I may be over-interpreting the limited information I see in the Table, but median survival times seem to be longer in the three studies in which panitumumab was used (Amado, Freeman, Muro) than the other studies. Is this worth a comment?Although this observation is accurate, there are substantial differences in the populations included in the studies we analyzed (high potential for confounding). As such, comparisons of median survivals between studies may not be valid and we refrain from including specific comments in the report.
Page 73, Fig 7. Needs (ref) after author, year. The order of studies (year, then alphabetical first author) is different from the tables � I would have liked to see grouping by design and whether 1st line or salvage treatment. I see at different points Yen, 2008 and Yen, 2009, but I think there is just one reference � please check.

The studies in the forest plot are ordered by year of publication and then by author name. An explanatory footnote has been added. Results by specific subgroups are reported in the subgroup analysis table (Table 19). It is imposible to present all subgroup analyses in one figure.

Following our update of the search strategy, there are now two studies by Yen et al., one published in 2009 and one in 2010. We have corrected the publication year in the figures and tables.

Page 78: Is there any possibility that publication bias is relevant?It is hard to discern what the Peer Reviewer refers to. The consistency and magnitude of effects in the KRAS topic is reassuring regarding the threat of publication bias.
Peer Reviewer 2BCR-ABLPage 79, line 2 � this is annual number of newly diagnosed cases, not incidence

The sentence now reads:
�Chronic myeloid or myelogenous leukemia (CML) is a relatively uncommon hematological malignancy with approximately 5,000 new cases diagnosed annually.�

Page 85, last 3 lines, Fig 12 and Fig 13 (p.90, with brief ref in text in last line of p. 87). I think it would make more sense to deal with these points when describing the second line TKI studies. For the 1st line and 3rd line studies, could be deal with briefly in text.We adopted the Peer Reviewer�s suggestion. Now the relevant sentence reads:
�Most publications particularly in 2nd-line TKI treatments originated from MD Anderson Comprehensive Cancer Center.�
Page 87, para 1, line 3: does �noncomparative cohorts� mean that they were single arm studies � I would keep that nomenclature from section 2.We replaced non-comparative cohorts with single-arm studies, as suggested.
Page 95, para 2, line 2: �some studies� � as far as I understand from the tables, patients had received interferon in all but one of the studies.We changed the sentence as follows: In most studies, the vast majority of patients had also received other therapies such as interferon (Table 23).
Page 103, para 2, line 2: statement �17 to 71 percent for accelerated or blastic phase� is incorrect. This is range in Table 26 for miscellaneous phases. For the accelerated/blastic phase, range in Table 25 is 27-60%, very similar to chronic phase.The Page 103 pointed out by the Peer Reviewer should be Page 102. We corrected the percentage numbers accordingly. Thank you.
Last two lines: what studies do not conform to this pattern, and do their results look different?This is our supposition and there is no supporting evidence presented in the report; therefore, we deleted the relevant two sentences (P102 the last two lines).
Page 107, para 2: worth noting that the Jabbour study had lowest proportion of patients tested (Table 24)We inserted a sentence based on the suggestion:
�The study found no patients with the T315I mutation but assessed only 30% of the entire patient cohort for the presence of mutations .�
Page 108, line 1: Figure 15We corrected these converting errors. Thank you.
Page 114, para 1, line 6: Figure 16We corrected this accordingly. Thank you.
Page 21, penultimate line: change �0 cells� to �cells with zero entries�We made the suggested change.
Page 126, Table 33: It would be helpful if the footnotes more clearly related to the columns they are explainingThank you for pointing out. We corrected as per suggestion.
Page 127: a lot of abbreviations, not all of which are explainedWe added following abbreviations: AP (=accelerated phase), BC (=blastic phase), CCyR (=complete cytogenetic response), CHR (=complete hematologic response), CP (=chronic phase), CR (=complete reponse), CyR (=cytogenetic response), MCyR (=major cytogenetic response), OS (=overall response), PFS (=progression free survival), and RR (=relative risk).
Page 129, para 3: In general agree with conclusion that individual patient meta-analysis may be needed, but comment on last couple of lines about studies originating from limited number of referral centres raises question as whether more effort needed to assemble body of data from which greater generalizability will be possibleAgain, there is no evidence to support our claim of lack of generalizability in the report; therefore, we deleted the relevant two sentences (P129 the last two lines of the paragraph 3).
Peer Reviewer 2Cross-cuttingPage 132, 3rd bullet from end, factorsWe have corrected the typographical error.
Page 133: 2nd bullet _ insert �the reader� after �we remind�; further down �various biases� _ what are these?We added �the reader�. We clarified regarding biases.
Missing pages: I noted 66, 76, 92, 94, 128, 134These are even pages left black when there is a new section starting in an odd page.
Peer Reviewer 3GeneralOverall, this is an impressive and well-conducted evidence report � particularly in regard to the challenging analysis of the BCR-ABL mutations in CML.We thank the Peer Reviewer for the kind comment.
Key Questions: For future studies, could other �intermediate� key questions be asked � e.g., related to potential benefits or harms?For our review, the Key Questions were determined at the planning phase of the review.
Peer Reviewer 3�SummaryLack of clarity in BCR-ABL summary:
Summary (p. S-3): The first several sentences of the BCR-ABL section are extremely confusing and appear contradictory. Suggest re-writing, and defining TKI.
We rewrote the relevance sentence as follows:
� The presence of any BCR-ABL1 mutation (all mutations considered together) does not appear to predict differential response to tyrosine kinase inhibitor (TKI) treatments (defined as imatinib-, dasatinib-, and nilotinib-based regimens).�
Peer Reviewer 3CYP2D6Need to include recent, large studies on tamoxifen:
Association Between CYP2D6 Polymorphisms and Outcomes Among Women With Early Stage Breast Cancer Treated With Tamoxifen. JAMA, October 7, 2009; 302: 1429 - 1436.
Kiyotani K, Mushiroda T, Imamura CK, et al: Significant effect of polymorphisms in CYP2D6 and ABCC2 on clinical outcomes of adjuvant tamoxifen therapy for breast cancer patients. J Clin Oncol 28:1287�1294, 2010
We have updated the search strategies for all outcomes (including tamoxifen for breast cancer). The two studies suggested by the Peer Reviewer were among the new studies we identified.
Lack of pooling for tamoxifen studies:
p. 12. It is surprising that no attempt was made at meta-analysis for tamoxifen. Despite �irreconcilable� differences in definitions of metabolizer status, a random effects meta-analysis should be strongly considered.
p. 28. If the adjusted values presented in several studies are misleading, why not pool unadjusted results? Furthermore, classifying a relative effect of 1.55 as �modest� does nothing to confer the potential population impact. While the quality of the studies was clearly poor, the authors could conduct a random effects meta-analysis that would be more exploratory in nature. The results of such do not need to influence the overall conclusion of the report (and shouldn�t given the limitations), but the lack of synthesis here is an omission. If the authors choose not to conduct a random effects meta-analysis, a more specific explanation for their rationale should be provided.

Random effects meta-analysis (under the commonly utilized models) assumes that there is a distribution of true effects and the analytical method attempts to identify the mean and the uncertainty around it. Based on the heterogeneity of outcomes reported, genotypes investigated, performing a meta-analysis of the unadjusted results would not overcome the limitations discussed above. In addition, very few studies reported anadjusted estimates or crude event rates to allow calculation of the unadjusted estimates. As such any pooled analysis would be subject to reporting bias.

Since this report concerns pharmacogenetic tests, calculating the �population� effect would require use of the �average response rate� which is hard to calculate and cannot be representative of the patients from whom the relative effect was derived.

Peer Reviewer 3KRASp. 78: Although the authors briefly discuss the weight of evidence for cetuximab vs. panitumumab, it seems that stratified pooled analyses are warranted, and mention in the summary of any differences in effect sizes or amount of evidence � particularly for RCTs.We have conducted updated meta-analyses for the KRAS pharmacogenetic test and we have investigated the potential difference between panitumumab- and cetuximab-based studies when enough studies were available.
Peer Reviewer 3Generalp. 3: The statements �must be demonstrated� and �require proof of benefit� are somewhat extreme. �should be� and �evidence of� ?The suggested changes have been implemented.
Peer Reviewer 3SummaryTable S1: title should specify that numbers of studies is what is being listed.Given the extreme overlap between studies (particularly for the KRAS topic) patient numbers would not be representative and may be misleading.
Peer Reviewer 3CYP2D6Fig. 3 is very interesting and helpful. Suggest adding bar color coding legend, rather than defining colors in figure caption.This would not be feasible.
Peer Reviewer 3KRASFig. 5 is useful but legibility could be improvedWe have updated the figure to include the new KRAS-relevant studies and have also increased the resolution to improve legibility.
Table 10. Delete �could be color code�?We have removed this artifact of the editing process.
Peer Reviewer 4CYP2D6The SACGT definition focuses on heritable variation, which would only seem relevant to CYP2D6 polymorphisms in the context of this report. The �lumping� of tests based on somatic mutations (which differentiate disease processes) from heritable mutations (which differentiate individuals) is unfortunate and confusing to the average practitioner and patient.This is not accurate, the definition includes somatic mutations. (there is an �or� connecting several attributes of what falls under genetic testing, and somatic mutations are not explicitly excluded).
We have provided further clarification regarding the definition of genetic tests. In the Methods of the TA, the introductions, and results of the specific parts and the Discussion of the methodological topics, we highlight the distinction between germ-line and somatic variation.
In regard to CYP2D6, the timing of the literature review was such that the most important paper on this topic (Schroth, JAMA, 2009) was not included. This well done retrospective study convincingly demonstrates a relationship between CYP2D6 genotype and rate of recurrence. In addition, a recent study by Kiyotani (J Clin Oncol, 2010) provides further evidence supporting the importance of CYP2D6 genotyping.We have updated the search strategy and data extraction up to March 2010. The studies referenced by the Peer Reviewer were among the ones we identified in our update.
The lack of an association of genotype and mortality should be expected, given that tamoxifen�s primary effect is to delay progression rather than cure micrometastatic disease. For the same reasons, there is no basis for looking at overall recurrence rate. Thus, a more appropriate meta-analysis would focus on those studies that focused on time to recurrence or recurrence-free survival.We collected data both on mortality, overall survival, disease recurrence and time to recurrence. We did not perform meta-analysis for any of these outcomes because of the significant clinical heterogeneity.
The potential value of CYP2D6 genotyping is not well articulated. Randomized trials have demonstrated that letrozole is superior to tamoxifen for the overall population, but it is plausible that the two drugs have equivalent efficacy, except in CYP2D6 slow metabolizers. Thus, CYP2D6 genotyping could identify patients who would be spared the cost and toxicities of letrozole and other aromatase inhibitors.Although the hypotheses brought forward by the Peer Reviewer are plausible the systematic review methodology is geared towards answering specific clinical questions based on available evidence. Although prediction of response to tamoxifen would be very important from a public health perspective, this does not validate CYP2D6 polymorphisms as a predictive (pharmacogenetic) test.
The potential risk of CYP2D6 genotyping is also not well articulated. If CYP2D6 slow metabolizers do not have an inferior result, then the administration of tamoxifen (in lieu of an aromatase inhibitor) to extensive metabolizers would be harmful.Please see above.
Table 6 does not appropriately represent the issues regarding tamoxifen and CYP2D6 genotype. The underlying hypothesis is that all (or most) of the effects (both toxic and beneficial) of tamoxifen are a result of metabolites formed by CYP2D6. Thus, the assertion that the use of CYP2D6 inhibitors is not a relevant covariate is incorrect. If there is a relationship between CYP2D6 genotype and tamoxifen efficacy, that relationship would be masked if all patients were prescribed CYP2D6 inhibitors. Whether or not �confound� is the correct verb, the ideal analysis would exclude all patients who received CYP2D6 inhibitors, although that analysis would be confounded if there were more patients on CYP2D6 inhibitors who were extensive metabolizers (hypothetically due to a greater effect of tamoxifen in this population). For the same reason, adherence may vary among CYP2D6 genotypes, if the toxicity varies among CYP2D6 genotypes.

The use of the word �confound� is not a matter of terminology but one of substance. If a variable sis a potential confounder (and there are solid epidemiologic methods to identify potential confounders) then �conditioning� on the effect of the confounder (commonly achieved by performing multivariate regression including the confounder as a variable) would be necessary. On the other hand, if the variable of interest is not a confounder then simple inclusion in regression models will reduce precision and may also generate bias, thus adverse effects on study accuracy can occur when non-confounding factor is treated as such.

Because the in the scenarios described by the Peer Reviewer use of CYP2D6-inhibiting medications cannot be a confounder adjusting for their use in regression (commonly performed in the studies we reviewed) is a serious flaw.

Additionally, exclusion of individuals using CYP2D6 inhibiting medications for all analyses is problematic because it fails to account for all available information.

The most appropriate recommendation is the formation of a consortium to conduct an aggregated analysis of all data from the tamoxifen adjuvant trials. Ideally, all genotypes would be included, and there would be a quality control process to ensure that the genotypes are analytically valid. It would also be ideal to have uniform clinical data, particularly time to recurrence. However, it will be difficult to have consistent information regarding concomitant CYP2D6 inhibitors and adherence for these retrospective studies.

We agree that this would be a step forward for the CYP2D6 pharmacogenetics field. We have added the following comment in the discussion section:

�Efforts to standardize the definitions of metabolizer groups based on genotype information would allow uniform reporting and facilitate patient-level synthesis of results across studies.�

The recommendation regarding �repurposing� of randomized clinical trials is a good one. In fact, one could consider recommending that federal funding agencies require collection of germline DNA on all federally funded clinical trials, unless the investigator provides an adequate scientific justification for not doing so (e.g., small sample size). (This is similar in concept to NIH requirements to study women, minorities, and children.) There may be an imbalance in any randomized trial in potential covariates. Thus, it is appropriate to consider adjusting for such covariates, but this should be prospectively specified in order to avoid the data dredging that is so prevalent in the pharmacogenomics literature. The emphasis on analyzing data from randomized trials is important, particularly when there is no a priori hypothesis based on other pharmacological or biological data. However, for CYP2D6, there is no basis for hypothesizing any relationship between CYP2D6 and prognosis (in the absence of treatment) and thus this concern is less relevant (albeit still valid). In contrast, one would anticipate that somatic mutations may have an effect on prognosis.[no response needed]

1 Peer reviewers are not listed in alphabetical order.
2 If listed, page number, line number, or section refers to the draft report.
3 If listed, page number, line number, or section refers to the final report.

Return to Contents

 

Page last reviewed August 2014
Internet Citation: Table 1. Invited Peer Reviewer Comments: Disposition of Comments. August 2014. Agency for Healthcare Research and Quality, Rockville, MD. http://www.ahrq.gov/research/findings/ta/pharmacogenetic-tests/table1.html