Assessing the generalizability of all-cause mortality from the MOSAIC trial to older adults diagnosed with stage II and III colon cancer in SEER-Medicare
Jennifer L. Lund
University of North Carolina at Chapel Hill, Department of Epidemiology, Gillings School of Global Public Health
Dr. Lund receives research funding through the UNC Oncology Clinical Translational Research Training Program (K12 CA120780).Dr. Sanoff has received research funding paid to the University of North Carolina from Bayer, Novartis, Merck, and Precision Biologics. Dr. Meyer has received honorarium from the National Cancer Institute, the American Society for Clinical Oncology, and Merck.TS receives investigator-initiated research funding and support as Principal Investigator (R01 AG023178) from the National Institute on Aging (NIA), and as
Co-Investigator (R01 CA174453; R01 HL118255, R21-HD080214), National Institutes of Health (NIH). He also receives salary support as Director of the Comparative
Effectiveness Research (CER) Strategic Initiative, NC TraCS Institute, UNC Clinical and Translational Science Award (UL1TR001111) and as Director of the Center for Pharmacoepidemiology (current members: GlaxoSmithKline, UCB BioSciences, Merck) and research support from pharmaceutical companies (Amgen,
AstraZeneca) to the Department of Epidemiology, University of North Carolina at Chapel Hill. Dr. Stürmer does not accept personal compensation of any kind from
any pharmaceutical company. He owns stock inNovartis, Roche, BASF, AstraZeneca, and Johnsen & Johnsen.
Underrepresentation in RCTs may be problematic when patient subgroups experience different harms and benefits from treatment but there are too few patients within subgroups to accurately estimate treatment effects. Older adults may experience different treatment effects (compared with younger adults). Underrepresentation of older adults in cancer trials is of concern, as the median age at diagnosis is 67 and adults over 65 represent 60% of cancer diagnoses and 70% of cancer deaths. Despite this, a 2006 study found that those participating in cancer trials were generally younger and had better performance status than those that did not enroll.
As a result, substantial interest has focused on developing ways to estimate how treatment outcomes (e.g., risk of mortality) and effects (e.g., outcomes comparing one treatment to another) may differ when given to older adults treated in routine clinical practice versus the clinical trial setting - i.e., generalizability of trial findings. In the past, researchers have tried to evaluate the generalizability of trial findings of older metastatic cancer patients with solid tumors by directly comparing their outcomes to older patients treated in routine clinical practice (e.g., using the Surveillance, Epidemiology, and End Results (SEER)-Medicare database). However, due to the small numbers of older adults enrolled in the specific RCTs, these researchers were unable to account for differences between RCT participants and cancer patients treated in community settings with respect to variables like age and sex, which might plausibly influence outcomes and treatment effects. A new statistical method to address this shortcoming involves the use of weighting techniques similar to survey sampling weights (i.e., generalizability weights).
In this study, we propose to apply these generalizability weighting methods using data from older adults (age 65-75 years) enrolled in the MOSAIC randomized controlled trial (RCT) and SEER-Medicare observational cohort (a linkage of cancer registry and Medicare enrollment and claims data). Specifically, we aim to: (1) estimate the expected effects of adjuvant chemotherapy with oxaliplatin versus 5-fluorouracil (5-FU) alone in routine care populations of older adults (age 65-75 years) diagnosed with stage II and III colon cancer and (2) assess whether we can use the MOSAIC data to predict all-cause mortality in an observational SEER-Medicare cohort of older adults.
We will conduct a cohort study using patient-level MOSAIC trial data and aggregated information from the SEER-Medicare database, providing the unique opportunity to pursue our aims. The results of this study will be used to inform policymakers about the population-level impact of adjuvant chemotherapy with oxaliplatin (versus 5-FU alone) among older adults treated in routine care settings. We will plan to present our research findings at scientific conferences and publish our results in a peer-reviewed journal.
[{ "PostingID": 4261, "Title": "SANOFI-EFC3313", "Description": "Multicenter International Study of Oxaliplatin/ 5FU-LV in the Adjuvant Treatment of Colon Cancer" }]
Measures of interest:We will reweight the entire MOSAIC trial population (i.e., both arms) who are aged 65-75 years at randomization to reflect the distribution of key covariates of patients aged 65-75 years with stage II and III colon cancer in the SEER-Medicare cohort initiating oxaliplatin (i.e., age at diagnosis, sex, stage of cancer (AJCC stage, substage), tumor size, (T1-4), lymph node involvement (0, 1, or 2+), differentiation (well, moderately, poor), and perforation (yes vs. no). We will then estimate the 3- and 5-year risk difference and 3- and 5-year risk ratio for all-cause mortality corresponding to the estimated “treatment effect in the treated” overall and by stage. We will use a similar weighting method to estimate a treatment effect in all patients who receive any treatment (or the estimated “treatment effect in the population”).For our second aim, we will estimate three different measures of association in our evaluation of differences in all-cause mortality comparing patients receiving oxaliplatin in the MOSAIC trial and patients receiving oxaliplatin in the observational SEER-Medicare cohort: • a log-rank test examining differences in the overall survival comparing MOSAIC trial patients and SEER-Medicare cohort patients• 3- and 5-year cumulative risk differences for all-cause mortality comparing MOSAIC trial patients and SEER-Medicare cohort patients• 3- and 5-year cumulative risk ratios for all-cause mortality comparing MOSAIC trial patients and SEER-Medicare cohort patients• We will then replicate the above three tests after applying the generalizability weights derived using the SEER-Medicare cohort data, which will include age at age at diagnosis/randomization, sex, stage of cancer (AJCC stage, substage), tumor size, (T1-4), lymph node involvement (0, 1, or 2+), differentiation (well, moderately, poor), and perforation (yes vs. no). • We will further restrict our population based upon predicted an individual's probability of frailty or disability (based upon two Medicare claims-based models) in an additional set of analyses.Methods to control for bias:We will use two main methods to control for potential factors biasing our estimates:• Standardized mortality ratio (generalizability weights) weighting of the MOSAIC trial population to reflect the distribution of covariates in the SEER-Medicare cohort initiating adjuvant oxaliplatin with respect to the joint distributions of the above-mentioned variables.• Restriction of frail or disabled individuals from the SEER-Medicare cohort using Medicare claims-based models of frailty and disability.Analyses and planned adjustments for covariates:For our first analysis (estimating a generalized effect of oxaliplatin versus 5FU alone on all-cause mortality), we will apply generalizability weights to both arms of the MOSAIC trial (restricted to those age 65 years and older). These generalizability weights will be estimated after combining our aggregated observational data from SEER-Medicare with the MOSAIC trial data, including data on distributions of age at diagnosis/randomization, sex, stage of cancer (AJCC stage, substage), tumor size, (T1-4), lymph node involvement (0, 1, or 2+), differentiation (well, moderately, poor), and perforation (yes vs. no). With this combined dataset, we will use multivariable logistic regression to estimate each individuals' probability of being “sampled” into (i.e., a participant in) the trial as a function of their individual characteristics (e.g., age, sex, cancer stage). This is called the sampling probability and is calculated analogously to the propensity score. Next, we will use each individuals' sampling probability to construct their sampling weight (sw), as follows:If an individual is enrolled in MOSAIC trial: sw=P(MOSAIC=1)/P(MOSAIC=1)|C) If an individual is in the SEER-Medicare cohort: sw=0The quantity P(MOSAIC=1) is the number of patients enrolled in MOSAIC divided by the number of patients in the SEER-Medicare cohort. The quantity P(MOSAIC=1|C) is the probability of being sampled into the MOSAIC trial versus the SEER-Medicare population, given an individual's characteristics, C, which are defined above. This weighting scheme creates a weighted pseudo-trial population that reflects the distribution of patient characteristics in the SEER-Medicare target population. In this this weighted trial population, we will use Kaplan-Meier methods and Cox proportional hazards models to re-estimate the effect of adjuvant oxaliplatin versus 5-FU on all-cause mortality and report hazard ratios and 3- and 5-year risk differences and risk ratios and 95% confidence intervals using a robust variance. These estimates will reflect a generalizable treatment effect relevant in a target population of older adults in SEER-Medicare. We will estimate 3- and 5-year risk differences and risk ratios for all-cause mortality and 95% confidence intervals using bootstrapping with 200 replicates. We will also perform three distinct analyses in our comparison of all-cause mortality outcomes in the MOSAIC trial versus SEER-Medicare populations. In the first analysis, we will directly compare all-cause mortality among MOSAIC trial patients aged 65+ years that received oxaliplatin (i.e., active arm) and SEER-Medicare stage II and III colon cancer patients age 65+ years that received oxaliplatin in routine practice. In the second analysis, we will apply generalizability weights similar to the approach described above such that patients in the MOSAIC trial are weighted to reflect the distribution of patient characteristics in the SEER-Medicare oxaliplatin-treated cohort (e.g., according to age, sex, stage, etc.). The all-cause mortality of this weighted MOSAIC population (again restricted to those age 65+ years) will then be compared with the observational SEER-Medicare population. For our third analysis, we will use two different approaches to identify and remove individuals from the SEER-Medicare cohort who are likely to be frail or disabled using two Medicare claims-based models. We will then re-weight the MOSAIC trial population using generalizability weights to reflect the distribution of covariates in this restricted SEER-Medicare population as in the second analysis. Statistical approach, sample size, and missing data:In the first aim of this project, we use the MOSAIC trial and identify the 400 patients aged 65 years and over in the oxaliplatin arm and 380 patients 65 and over in the comparator (5-FU) arm. We expect a total population of SEER-Medicare patients with stage II and III colon cancer treated with 5-FU or oxaliplatin of greater than 7,000 (this population is used to establish the distribution of covariates for weighting). An intent-to-treat, complete case analysis will be conducted. Given the relatively low quantity of variables that we are standardizing by and their ease of collection in both the MOSAIC trial and SEER-Medicare observational cohorts, missingness in covariates will be limited. We will use Kaplan-Meier methods to account for censoring when constructing survival curves. In our analyses estimating the effects of adjuvant oxaliplatin versus 5-FU on all-cause mortality (in the population of adults age 65+ years), we expect 80% power to detect a risk ratio of 0.64 in 5-year overall survival based upon reported numbers of older adults in the MOSAIC trial and a ratio of 0.4:0.6 of stage II to stage III cancer in those older adults, assuming 5-year overall survival in the control group of 0.70. As this is an exploratory approach to estimating generalizable treatment effects in populations in routine clinical care using both RCT and observational cohort data, this low level of power does not impact our decision to move forward with analysis. In our second aim of this project, we will assess both qualitative and quantitative differences in all-cause mortality comparing the MOSAIC trial and SEER-Medicare cohorts who receive oxaliplatin and evaluate changes in these differences as we incrementally apply generalizability weights and cohort restrictions. We will perform log-rank tests to compare overall survival curves and use bootstrapping to generate 95% confidence intervals for our estimates of 3- and 5-year mortality risk difference and risk ratio comparing the two groups in all four of the above analyses. We will use a frequentist statistical approach and assess statistical significance through p-values (a=0.05) and 95% confidence intervals. We cannot know the precision of our estimates of the difference between the two populations receiving oxaliplatin without knowing the joint covariate distributions in the SEER-Medicare and MOSAIC populations, but the sample size appears large enough to see trends in changes in the estimates, as we apply generalizability weights and restrict the cohort.General References for Proposal:1. Sardar M., Badri M., et al. Underrepresentation of Women, Elderly Patients, and Racial Minorities in the Randomized Trials Used for Cardiovascular Guidelines. JAMA Intern Med. 2014;174(11):1868-1870. doi:10.1001/jamainternmed.2014.47582. Hernán MA, VanderWeele TJ. Compound Treatments and Transportability of Causal Inference. Epidemiology. 2011;22(3):368-377. doi:10.1097/EDE.0b013e3182109296.3. Hutchins L, Unger J, Crowley J, et al. Underrepresentation of patients 65 years of age or older in cancer-treatment trials. N Engl J Med. 1999 Dec 30;341(27):2061-7.4. Hurria A., Levit L., et al. Improving the Evidence Base for Treating Older Adults With Cancer: American Society of Clinical Oncology Statement. Journal of Clinical Oncology. 2015 Sept; 33(25): 2815-2820.5. Elting L, Cooksley C, Bekele B, et al. Generalizability of cancer clinical trial results: prognostic differences between participants and nonparticipants. Cancer. 2006 Jun 1;106(11):2452-8.6. McCleary NJ, Meyerhardt JA, Green E, et al. Impact of Age on the Efficacy of Newer Adjuvant Therapies in Patients With Stage II/III Colon Cancer: Findings From the ACCENT Database. Journal of Clinical Oncology. 2013;31(20):2600-2606. doi:10.1200/JCO.2013.49.6638.7. Sanoff HK, Carpenter WR, Stürmer T, et al. Effect of Adjuvant Chemotherapy on Survival of Patients With Stage III Colon Cancer Diagnosed After Age 75 Years. Journal of Clinical Oncology. 2012;30(21):2624-2634. doi:10.1200/JCO.2011.41.1140.8. Yothers G, O'Connell MJ, Allegra CJ, et al. Oxaliplatin As Adjuvant Therapy for Colon Cancer: Updated Results of NSABP C-07 Trial, Including Survival and Subset Analyses . Journal of Clinical Oncology. 2011;29(28):3768-3774. doi:10.1200/JCO.2011.36.4539.9. Cole SR, Stuart EA. Generalizing Evidence From Randomized Clinical Trials to Target Populations: The ACTG 320 Trial. American Journal of Epidemiology. 2010;172(1):107-115. doi:10.1093/aje/kwq084.10. Stuart EA, Cole SR, Bradshaw CP, Leaf PJ. The use of propensity scores to assess the generalizability of results from randomized trials. Journal of the Royal Statistical Society Series A, (Statistics in Society). 2001;174(2):369-386. doi:10.1111/j.1467-985X.2010.00673.x.11. Lamont EB, Schilsky RL, He Y, et al. Generalizability of Trial Results to Elderly Medicare Patients with Advanced Solid Tumors (Alliance 70802). JNCI Journal of the National Cancer Institute. 2015;107(1):dju336. doi:10.1093/jnci/dju336.12. Andre T., Boni C., et al. Improved overall survival with oxaliplatin, fluorouracil, and leucovorin as adjuvant treatment in stage II or stage III cancer in the MOSAIC trial. Journal of Clinical Oncology. 2009 Jul 1;27(19): 3109-16.13. Schmoll H., Cunningham D., et al. Cediranib With mFOLFOX6 Versus Bevacizumab With mFOLFOX6 As First-Line Treatment for Patients With Advanced Colorectal Cancer: A Double-Blind, Randomized Phase III Study (HORIZON III). Journal of Clinical Oncology. 2012 Sept; 30(29): 3596-3603.14. Faurot K., Jonsson Funk M., et al. Using claims data to predict dependency in activities of daily living as a proxy for frailty. Pharmacoepidemiology and Drug Safety. 2015 Jan; 24(1): 59-66.15. Davidoff A., Gardner L., et al. Validation of disability status, a claims-based measure of functional status for cancer treatment and outcomes studies. Medical Care. 2014 Jun;52(6):500-10.16. Westreich D., Edwards J., et al. Causal Impact: Epidemiological Approaches for a Public Health of Consequence. American Journal of Public Health: 2016 Jun; 106(60: 1011-1012.17. Williams G., Deal A., et al. Feasibility of geriatric assessment in community oncology clinics. Journal of Geriatric Oncology. 2014 Jul; 5(3): 245-51.18. Meyerhardt JA, Li L, Sanoff HK, Carpenter Wt, Schrag D. Effectiveness of bevacizumab with first-line combination chemotherapy for Medicare patients with stage IV colorectal cancer. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. Feb 20 2012;30(6):608-615.19. Rosenbaum PR, Rubin DB.The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70 (1): 41-55.
Lund JL, Webster-Clark MA, Hinton SP, Shmuel S, Stürmer T, Sanoff HK. Effectiveness of adjuvant FOLFOX vs 5FU/LV in adults over age 65 with stage II and III colon cancer using a novel hybrid approach. Pharmacoepidemiol Drug Saf. 2020 Oct 4.
https://doi.org/10.1002/pds.5148