Subpopulation Treatment Effect Pattern Plot Research Articles

Two Techniques for Investigating Interactions between Treatment and Continuous Covariates in Clinical Trials

There is increasing interest in the medical world in the possibility of tailoring treatment to the individual patient. Statistically, the relevant task is to identify interactions between covariates and treatments, such that the patient's value of a given covariate influences how strongly (or even whether) they are likely to respond to a treatment. The most valuable data are obtained in randomized controlled clinical trials of novel treatments in comparison with a control treatment. We describe two techniques to detect and model such interactions. The first technique, multivariable fractional polynomials interaction, is based on fractional polynomials methodology, and provides a method of testing for continuous-by-binary interactions and by modeling the treatment effect as a function of a continuous covariate. The second technique, subpopulation treatment-effect pattern plot, aims to do something similar but is focused on producing a nonparametric estimate of the treatment effect, expressed graphically. Stata programs for both of these techniques are described. Real data for brain and breast cancer are used as examples.

Identification of a subpopulation of metastatic breast cancer patients with very high HER2 expression levels and possible resistance to trastuzumab

1059 Background: Many HER2-positive patients with metastatic breast cancer (MBC) fail to respond to trastuzumab. We previously reported that precise quantitation of HER2 expression (H2T) by the HERmark assay identified a sub-population of IHC 3+, FISH(+) (positive) patients with low H2T levels that responded poorly to trastuzumab (Lipton, San Antonio Breast Cancer Symposium 2008, abs #32). Here we identify a sub-population of FISH(+) patients with very high H2T levels, that experience clinical outcomes that are indistinguishable from those of FISH(-) (negative) patients with low H2T levels. Methods: The HERmark assay was used to measure H2T in formalin-fixed, paraffin-embedded (FFPE) primary breast tumor specimens from 99 women treated with trastuzumab for MBC. Specimens were also tested by central FISH. A sub-population treatment effect pattern plot (STEPP) was generated to examine the progression-free survival (PFS) rate at 12 months after treatment with trastuzumab across the distribution of H2T. Kaplan-Meier (KM) analyses were performed comparing the PFS of FISH(-), H2T low (log10H2T < 1.25) patients with those of FISH(+), H2T high (log10H2T ≥ 1.95) and FISH(+), H2T intermediate (1.25 < log10H2T < 1.95) groups. Cutoffs were identified by lowest p-value in a positional scanning analysis. Results: The PFS rate improved gradually with increasing H2T in STEPP analyses. At the highest levels of H2T, an abrupt decrease in the PFS rate was observed, consistent with a reduction in susceptibility to trastuzumab. KM analyses demonstrated that patients who were FISH(+), H2T intermediate had a significantly longer PFS than patients who were FISH(-), H2T low (median PFS 12.6 vs. 4.5 mos; HR = 0.34; p < 0.0001). Patients that were FISH(+), H2T high experienced a PFS that was no better than patients that were FISH(-), H2T low (median PFS 4.6 vs. 4.5 mos; HR = 0.87; p = 0.68). Conclusions: Precise quantitation of HER2 expression levels allows the identification of multiple sub-populations of HER2(+) patients that have different clinical outcomes on trastuzumab. MBC patients with very high levels of H2T could represent a sub-group with de novo resistance to trastuzumab who may benefit from combined therapy. [Table: see text]

A small sample study of the STEPP approach to assessing treatment–covariate interactions in survival data

A new, intuitive method has recently been proposed to explore treatment-covariate interactions in survival data arising from two treatment arms of a clinical trial. The method is based on constructing overlapping subpopulations of patients with respect to one (or more) covariates of interest and in observing the pattern of the treatment effects estimated across the subpopulations. A plot of these treatment effects is called a subpopulation treatment effect pattern plot. Here, we explore the small sample characteristics of the asymptotic results associated with the method and develop an alternative permutation distribution-based approach to inference that should be preferred for smaller sample sizes. We then describe an extension of the method to the case in which the pattern of estimated quantiles of survivor functions is of interest.

Breast cancer patients with estrogen receptor-negative/progesterone receptor-positive tumors: being younger and getting less benefit from adjuvant tamoxifen treatment

Most breast cancer patients with estrogen receptor-negative/progesterone receptor-positive (ER-/PgR+) tumors are premenopausal cases, with few alternatives of adjuvant endocrine therapy but tamoxifen (TAM). The efficacy of adjuvant TAM on ER-/PgR+ patients is still controversial. In this study, we evaluated the efficacy of adjuvant TAM on patients with ER-/PgR+ tumors. Among all 1,836 consecutive patients with operable primary breast cancer, 798 cases were with ER+/PgR+ tumors and 205 with ER-/PgR+ tumors. By sub-grouping the patients according to ER/PR phenotypes and whether the patients had been treated with adjuvant TAM therapy or not, we investigated the differences of survivals between groups. Patients with ER-/PgR+ tumors were younger than those with ER+/PgR+ tumors (P = 0.021), and were mainly premenopausal (P = 0.013). ER-/PgR+ patients were related to more involved lymph nodes and later stage. In the absence of TAM treatment, ER+/PgR+ group had a similar survival to ER-/PgR+ group in terms of 5-year disease-free survival (DFS), as well as overall survival (OS). After TAM treatment, both groups had increased survival rates comparing with the baseline of non-TAM-treated groups. Moreover, significant survival differences were then observed between TAM-treated ER+/PgR+ group and TAM-treated ER-/PgR+ group either in DFS (P = 0.016) or OS (P = 0.007). Of the TAM-treated patients, by sub-dividing the chemotherapy-treated population into CMF (cyclophosphamide, methotrexate and 5-fluorouracil) group and CA(E)F (cyclophosphamide, doxorubicin/epirubicin and 5-fluorouracil) group, we found that ER-/PgR+ group got more benefits from CMF regimen than from CA(E)F. Subpopulation treatment effect pattern plot (STEPP) analysis showed that the ER-/PgR+ group had an obvious worse survival than ER+/PgR+ group in younger patients (<55 years). Axillary lymph nodes involvement was the only independent prognostic factor for ER-/PgR+ group. Our results indicate that patients with ER-/PgR+ tumors are mainly premenopausal and young. Although patients with ER-/PgR+ tumors are generally considered as candidates for endocrine therapy clinically, the ER-/PgR+ group gains less benefits from adjuvant TAM treatment than ER+/PgR+ group.

Letrozole Compared With Tamoxifen for Elderly Patients With Endocrine-Responsive Early Breast Cancer: The BIG 1-98 Trial

To explore potential differences in efficacy, treatment completion, and adverse events (AEs) in elderly women receiving adjuvant tamoxifen or letrozole for five years in the Breast International Group (BIG) 1-98 trial. This report includes the 4,922 patients allocated to 5 years of letrozole or tamoxifen in the BIG 1-98 trial. The median follow-up was 40.4 months. Subpopulation Treatment Effect Pattern Plot (STEPP) analysis was used to examine the patterns of differences in disease-free survival and incidences of AEs according to age. In addition, three categoric age groups were defined: "younger postmenopausal" patients were younger than 65 years (n = 3,127), "older" patients were 65 to 74 years old (n = 1,500), and "elderly" patients were 75 years of age or older (n = 295). Efficacy results for subpopulations defined by age were similar to the overall trial results: Letrozole significantly improved disease-free survival (DFS), the primary end point, compared with tamoxifen. Elderly patients were less likely to complete trial treatment, but at rates that were similar in the two treatment groups. The incidence of bone fractures, observed more often in the letrozole group, did not differ by age. In elderly patients, letrozole had a significantly higher incidence of any grade 3 to 5 protocol-specified non-fracture AE compared with tamoxifen (P = .002), but differences were not significant for thromboembolic or cardiac AEs. Adjuvant treatment with letrozole had superior efficacy (DFS) compared with tamoxifen in all age groups. On the basis of a small number of patients older than 75 years (6%), age per se should not unduly affect the choice of adjuvant endocrine therapy.

Interactions Between Treatment and Continuous Covariates: A Step Toward Individualizing Therapy

Interactions Between Treatment and Continuous Covariates: A Step Toward Individualizing Therapy

Chemoendocrine Compared With Endocrine Adjuvant Therapies for Node-Negative Breast Cancer: Predictive Value of Centrally Reviewed Expression of Estrogen and Progesterone Receptors—International Breast Cancer Study Group

To centrally assess estrogen receptor (ER) and progesterone receptor (PgR) levels by immunohistochemistry and investigate their predictive value for benefit of chemo-endocrine compared with endocrine adjuvant therapy alone in two randomized clinical trials for node-negative breast cancer. International Breast Cancer Study Group Trial VIII compared cyclophosphamide, methotrexate, and fluorouracil (CMF) chemotherapy for 6 cycles followed by endocrine therapy with goserelin with either modality alone in pre- and perimenopausal patients. Trial IX compared three cycles of CMF followed by tamoxifen for 5 years versus tamoxifen alone in postmenopausal patients. Central Pathology Office reviewed 883 (83%) of 1,063 patients on Trial VIII and 1,365 (82%) of 1,669 on Trial IX and determined ER and PgR by immunohistochemistry. Disease-free survival (DFS) was compared across the spectrum of expression of each receptor using the Subpopulation Treatment Effect Pattern Plot methodology. Both receptors displayed a bimodal distribution, with substantial proportions showing no staining (receptor absent) and most of the remainder showing a high percentage of stained cells. Chemo-endocrine therapy yielded DFS superior to endocrine therapy alone for patients with receptor-absent tumors, and in some cases also for those with low levels of receptor expression. Among patients with ER-expressing tumors, additional prediction of benefit was suggested in absent or low PgR in Trial VIII but not in Trial IX. Low levels of ER and PgR are predictive of the benefit of adding chemotherapy to endocrine therapy. Low PgR may add further prediction among pre- and perimenopausal but not postmenopausal patients whose tumors express ER.

Prognostic and Predictive Value of Centrally Reviewed Expression of Estrogen and Progesterone Receptors in a Randomized Trial Comparing Letrozole and Tamoxifen Adjuvant Therapy for Postmenopausal Early Breast Cancer: BIG 1-98

To evaluate locally versus centrally assessed estrogen (ER) and progesterone (PgR) receptor status and the impact of PgR on letrozole adjuvant therapy compared with tamoxifen in postmenopausal women with early breast cancer. Breast International Group (BIG) 1-98 randomly assigned 8,010 patients to four arms comparing letrozole and tamoxifen with sequences of each agent. The Central Pathology Office received material for 6,549 patients (82%), of which 79% were assessable (6,291 patients). Prognostic and predictive value of both local and central hormone receptor expression on disease-free survival (DFS) were evaluated among 3,650 assessable patients assigned to the monotherapy arms. Prognostic value and the treatment effect were estimated for centrally assessed ER and PgR expression levels using the Subpopulation Treatment Effect Pattern Plot. Central review confirmed 97% of tumors as hormone receptor-positive (ER and/or PgR > or =10%). Of 105 tumors locally ER-negative, 73 were found to have more than 10% positive cells, and eight had 1% to 9%. Of 6,100 tumors locally ER positive, 66 were found to have no staining, and 54 had only 1% to 9%. Discordance was more marked for PgR than ER. Patients with tumors reclassified centrally as ER-negative, or as hormone receptor-negative, had poor DFS. Centrally assessed ER and PgR showed prognostic value. Among patients with centrally assessed ER-expressing tumors, letrozole showed better DFS than tamoxifen, irrespective of PgR expression level. Central review changed the assessment of receptor status in a substantial proportion of patients, and should be performed whenever possible in similar trials. PgR expression did not affect the relative efficacy of letrozole over tamoxifen.

Aromatase inhibitors (AI) for elderly patients: Efficacy, compliance and safety according to patient age in the BIG 1–98 trial

9033 Background: Because elderly patients (pts) are under-represented in clinical trials, care should be taken in extending overall results to that patient group. We compared letrozole (L) with tamoxifen (T) with respect to disease-free survival, treatment compliance, and adverse events (AEs) according to age in the continuous therapy arms of the BIG 1–98 trial. Methods: Between March 1998 and May 2003, 8,028 postmenopausal women with hormone receptor-positive breast cancer were enrolled in the BIG 1–98 trial. This report includes the 4,922 pts allocated to continuous therapy with either T or L for 5 years. Subpopulation Treatment Effect Pattern Plots (STEPP) were used to look at the patterns of differences in disease-free survival and incidences of AEs according to age. In addition, three categorical age groups were defined: “younger postmenopausal” &lt; 65, “older” 65–74, and “elderly” = 75. Cox proportional hazards models were used to compare T and L according to time to first AE, adjusting for relevant pre-existing risk factors. The median follow-up was 40.4 months. Results: STEPP analysis showed the L provided consistent benefit compared with T irrespective of age. Pts 75 and older were less likely to complete trial treatment, but similarly in the two treatment groups. Adjusted treatment comparisons of AEs according to age groups are given below. (T,L: significantly more AEs on T,L; NS: no significant difference) Conclusion: Adjuvant treatment with L had superior efficacy compared with T in all age groups. Physicians should base clinical decisions on the individual patient profile with respect to reduction of risk of recurrence from L as well as existing comorbidities. [Table: see text] No significant financial relationships to disclose.

Detecting an interaction between treatment and a continuous covariate: A comparison of two approaches

In clinical trials, there is considerable interest in investigating whether a treatment effect is similar in all patients, or that some prognostic variable indicates a differential response to treatment. To examine this, a continuous predictor is usually categorized into groups according to one or more cutpoints. The treatment/covariate interaction is then analyzed in factorial fashion using multiplicative terms. The use of cutpoints raises several difficult issues for the analyst. It is preferable to keep continuous variables continuous in such a model. To achieve this, the MFP algorithm for multivariable model-building with fractional polynomials was recently extended to a new algorithm called multivariable fractional polynomial interaction (MFPI). With the latter, covariates may be binary, categorical or continuous, and cutpoints are avoided. MFPI is compared with a graphical technique, the subpopulation treatment-effect pattern plot or subpopulation treatment effect pattern plot (STEPP). Differences between MFPI and STEPP are illustrated by re-analysis of a randomized trial in kidney cancer. The stability of the two procedures is investigated by using the bootstrap. The Type I error probability of MFPI to ‘detect’ spurious interactions is estimated by simulation. MFPI and STEPP are found to exhibit similar treatment/covariate interactions. The tail-oriented variant of STEPP is found to give more stable and interpretable results than the sliding window variant. The type 1 error probabilty of MFPI is found to be close to its nominal value.

Using clinical trial data to tailor adjuvant treatments for individual patients

The 2005 St. Gallen Consensus Panel provided recommendations for the treatment of early breast cancer which rely on target identification first. The foremost advance since 2005—demonstration of trastuzumab efficacy for patients with HER2-positive disease—was realized because an effective treatment was being evaluated and because the trial patients had the targeted disease as determined by quality-controlled assessment prior to study entry. The BIG 1-98 trial provides a striking reminder of patients’ benefit from reliable pathological tumor assessment so that targeted therapies are effectively utilized. Several statistical methods facilitate use of clinical trial data for individualizing treatment. Subgroup analyses summarized using forest plots are essential for better understanding the disease and its treatment. The HERA trial illustrates the interpretation of relative and absolute treatment effects and estimating hazard rates over time as means to distinguish relevant differences across subgroups. Overview analyses, joint analyses, the STEPP (subpopulation treatment effect pattern plot) method and recursive partitioning are valuable tools.

Patterns of treatment effects in subsets of patients in clinical trials.

We discuss the practice of examining patterns of treatment effects across overlapping patient subpopulations. In particular, we focus on the case in which patient subgroups are defined to contain patients having increasingly larger (or smaller) values of one particular covariate of interest, with the intent of exploring the possible interaction between treatment effect and that covariate. We formalize these subgroup approaches (STEPP: subpopulation treatment effect pattern plots) and implement them when treatment effect is defined as the difference in survival at a fixed time point between two treatment arms. The joint asymptotic distribution of the treatment effect estimates is derived, and used to construct simultaneous confidence bands around the estimates and to test the null hypothesis of no interaction. These methods are illustrated using data from a clinical trial conducted by the International Breast Cancer Study Group, which demonstrates the critical role of estrogen receptor content of the primary breast cancer for selecting appropriate adjuvant therapy. The considerations are also relevant for general subset analysis, since information from the same patients is typically used in the estimation of treatment effects within two or more subgroups of patients defined with respect to different covariates.

Tailoring adjuvant treatments for the individual breast cancer patient

Background: Chemotherapy, tamoxifen and ovarian function suppression have all demonstrated their effectiveness for treating women with early breast cancer. Treatment selection for individual patients, however, requires estimates on the magnitude of treatment effects to be achieved from the application of each modality. Unfortunately, information currently available is insufficient to properly tailor adjuvant treatments. Methods: We consider predictive factors to improve our understanding about selection of adjuvant therapies, reassessment of data from previous clinical trials and design of future studies. Results: Estrogen receptor (ER) and progesterone receptor (PgR) are the primary measures available today to tailor adjuvant therapies. Patient age/menopausal status (ability to obtain treatment effects via ovarian function suppression), measures of the metastatic potential of the tumor (such as number of positive axillary lymph nodes), and concurrent use of chemotherapy and tamoxifen are other factors that modify the magnitude of relative effect associated with chemotherapy and endocrine therapies. The Subpopulation Treatment Effect Pattern Plots (STEPP) method displays the patterns of treatment effects within randomized clinical trials or datasets from meta-analyses to identify features that predict responsiveness to the treatments under study without relying on retrospective subset analysis. Confirmation of hypotheses using independent clinical trial databases is recommended. Discussion: All findings from clinical trials and meta-analyses should be presented primarily according to steroid hormone receptor status and patient age. Future studies should be designed as tailored treatment investigations, with endocrine therapies evaluated within populations of patients with endocrine responsive tumors, and chemotherapy questions focused within populations of patients with endocrine nonresponsive disease.

Features that predict responsiveness to chemotherapy and endocrine therapies

<h2>SUMMARY</h2> Prognostic factors, characterizing the background level of risk of relapse, and predictive factors, characterizing the degree of responsiveness to a specific treatment, are both used to select adjuvant therapies for patients with early-stage breast cancer. Determining how best to utilize available factors is challenging. We review various prognostic and predictive factors and present examples to illustrate how these factors can be used to improve our understanding about selection of adjuvant treatments, re-evaluation of data from previous clinical trials and design of future studies. Steroid-hormone-receptor status of the primary tumour and patient age/menopausal status (primarily reflecting the robustness of ovarian function) are the key features that predict responsiveness to chemotherapy and endocrine therapies. Qualitative interactions between these factors, and effects of combining chemotherapy and endocrine therapies, may confound treatment comparison. The STEPP (Subpopulation Treatment Effect Pattern Plots) method, by investigating the patterns of treatment effects within randomized clinical trials or datasets from meta-analyses, will help to identify features that predict responsiveness to the treatments under study without the pitfalls of selective retrospective subset analysis. Subset analyses according to steroid-hormone-receptor status and patient age should now be considered as prospectively defined. Future clinical trials should be designed as tailored treatment investigations, with endocrine therapies being evaluated within populations of patients with endocrine-responsive tumours, and chemotherapy questions being addressed within populations of patients with endocrine non-responsive disease.