Staff Collaborate Conference Room

Michael Martens, PhD

Michael Martens, PhD

Michael Martens, PhD

Assistant Professor

Locations

  • Institute for Health & Equity

Contact Information

Education

PhD, Biostatistics, Medical College of Wisconsin, Milwaukee, WI, 2017
MS, Mathematics, Northern Illinois University, DeKalb, IL, 2010
BS, Mathematics, Minor in Computer Science, Aurora University, Aurora, IL, 2005

Honors and Awards

F31 predoctoral training grant (#F31HL134317), NIH/NHLBI, 2016
Distinguished Student Paper Award, ENAR, 2016
Biostatistics Fellowship, Medical College of Wisconsin, 2012-13
Mathematical Sciences Fellowship, Northern Illinois University, 2008-10
Spartan Scholarship, Aurora University, 2004-05

Research Interests

Research Vision

Clinical trials are conducted to evaluate the potential benefit of proposed therapies. Because these studies can incur tremendous costs and require multiple years to conduct, statistical techniques that can reduce the expense and duration hold great value, including group sequential testing, adaptive design, and covariate adjustment. Time to event outcomes are often the primary interest in these trials, particularly in oncology; but these endpoints, in comparison to binary or continuous ones, require increasingly complex methodology in order to incorporate these techniques. In addition, accurate sample size calculation is crucial for both clinical trials and observations studies to permit them to meet their objectives while avoiding wasting resources.

My research goals include providing innovative group sequential tests and adaptive design methods for time to event endpoints in clinical trials, improving the precision of treatment evaluation and flexibility of the study as well as and reducing patient burden and study duration. Furthermore, I am interested in developing a general framework for accurate sample size calculation applicable to a wide variety of regression models and studies that accounts for any correlation among covariates and between subjects. These aims for future research are inspired by my experiences as a senior biostatistician at The Emmes Company (Emmes), providing statistical guidance and support for clinical trials research in blood and marrow transplant, cellular therapies, and ophthalmology.

Statistical Methodology Research

My previous work in statistical methodology includes group sequential testing of time to event outcomes, sample size determination for generalized linear models and time to event regression models, and nonparametric Bayesian inference. Research on the first two topics was funded by a training grant awarded by the National Heart, Lung, and Blood Institute (NHLBI) to support my dissertation research. Subsequent efforts have refined this research and brought it to publication.

1. A Group Sequential Test for Treatment Effect Based on the Fine-Gray Model
Competing risks endpoints arise when patients can fail therapy from several causes. Analyzing these outcomes allows one to assess the direct benefit of treatment on a primary cause of failure in a clinical trial setting. Group sequential design and regression modeling boost the efficiency of treatment evaluation, the former by allowing early stopping for efficacy and/or futility and the latter by accounting for other influential covariates. This paper proposes a group sequential test for treatment effect that uses the Fine-Gray model for covariate adjustment. This test is demonstrated through an analysis of data from BMT CTN 0402, a clinical trial evaluating an experimental therapy to guard against adverse outcomes following blood and marrow transplantation.

2. Group Sequential Tests for Treatment Effect on Survival and Cumulative Incidence at a Fixed Time Point
Medical research frequently involves comparing an event time of interest between treatment groups. Rather than comparing the entire survival or cumulative incidence curves, investigators may prefer to evaluate these probabilities at a fixed point in time, especially if the time point has a strong clinical importance. Performing a covariate adjusted analysis can improve efficiency, even in randomized clinical trials. This paper introduces covariate adjusted, group sequential, pointwise comparisons of survival and cumulative incidence probabilities derived from the direct binomial regression and stratified Cox models. Examples show the use of the proposed methods through reanalysis of survival and competing risks endpoints in the BMT CTN 0402 clinical trial.

3. A Unified Approach to Sample Size Determination for Common Nonlinear Regression Models
To enable a study to address its research goals properly, the sample size must be determined appropriately. Covariate adjustment via regression modeling permits more precise estimation of the effect of a primary variable of interest at the expense of increased complexity in sample size estimation. The existence of correlation between the main variable and other covariates, commonly seen in observational studies, further complicates this process. This manuscript introduces concise formulas for sample size determination with the generalized linear, Cox, and Fine-Gray models that account for this correlation. These sample size calculations are derived using a general, cohesive theory applicable to a broad class of regression models and covariate distributions. The formulas’ usage is exhibited through the hypothetical design of studies evaluating association between biomarkers and outcomes following blood and marrow transplant.

4. Low Information Omnibus (LIO) Priors for Dirichlet Process Mixture Models
Dirichlet process mixture (DPM) models offer flexible modeling of a data distribution as an infinite mixture from a given family of distributions. But, specifying an appropriate DPM model prior for an individual dataset can be challenging. This paper presents a scheme that requires only simple scaling information from the investigator. This is used to transform the data to a fixed scale on which a low information prior is applied that generates flexible distributions for the transformed data. Posterior samples are then back-transformed for inference on the original data. Using Gaussian and Weibull DPM models as examples, we show that the method provides accurate estimation for a diverse collection of data distributions that includes skewed, multimodal, and highly dispersed members.

Collaborative Biomedical Research

The primary biomedical research areas to which I have contributed include blood and marrow transplant and cellular therapies for blood cancers and other blood diseases. The following summarizes my work in this field professionally and as a student.

1. The Emmes Company - Data and Coordinating Center for the Blood and Marrow Transplant Clinical Trials Network (BMT CTN)
My primary role at Emmes is as a senior statistician for the Blood and Marrow Transplantation group, which serves as a data and coordinating center for the BMT CTN. Sponsored by the NHLBI, the BMT CTN conducts multicenter clinical trials to evaluate treatments impacting the prognosis of patients with blood cancers and other blood diseases. I am the lead statistician for eight late-phase clinical trials in this organization, supporting study conduct through sample size and power calculations; overseeing statistical programming of study progress reports; preparing and presenting safety data reports to the safety monitoring board, study sponsor(s), and FDA; planning and executing statistical analysis of study results; and abstract and manuscript preparation. Two of these studies are cosponsored by pharmaceutical companies seeking FDA registration of novel first-line and steroid refractory therapies for acute graft-versus-host disease (GVHD) following allogeneic transplant. Moreover, I analyzed an NHLBI-sponsored secondary study of BMT CTN data that investigated factors affecting immune suppression discontinuation following transplant. Using multistate modeling and landmarking techniques, we constructed predictive landmarking models for the probability of being immunosuppression-free and GVHD-free following allotransplant; a Shiny app allows easy computation of these probabilities, particularly at the point of care .This work with the BMT CTN has produced two journal articles and two conference presentations to date ([5]-[8]).

2. Center for International Blood and Marrow Transplant Research (CIBMTR)
I worked with the CIBMTR as a PhD student at MCW, providing statistical support for blood and marrow transplant studies using data from the CIBMTR database. Study topics included the increasing use of transplant in patients aged 70 years or older, the impact of donor relatedness and HLA matching on post-allogeneic transplant outcomes for ALL patients, and demographic and clinical factors associated with post-transplant outcomes in multiple myeloma and Waldenstrom Macroglobulinemia patients. I also performed statistical design and analysis for two secondary studies of the BMT CTN 0902 clinical trial’s data, evaluating the potential association of quality of life measures and exercise frequency with post-transplant outcomes including overall survival and hematologic recovery. This work produced six publications. ([9]-[14]).

Future Research Plans

Moving forward, I will continue to develop techniques for using interim analysis and covariate adjustment to improve the flexibility and efficiency of clinical trial research. Also, I will extend the sample size estimation methods previously developed to accommodate nonlinear mixed regression models, guiding the design of clinical trials and observational studies with correlated observations. Furthermore, I will conduct a comprehensive simulation study comparing performance of common statistical methods used for safety monitoring in clinical trials.

1. Interim Design Adaptations for Clinical Trials with Time to Event Endpoints
Adaptive designs offer increased flexibility for clinical trials by permitting modifications to the design mid-study while maintaining its validity. Potential changes include modifying the sample size, dropping treatment arms, and changing the primary endpoint of interest. Because control of the type I error rate is crucial for the validity of clinical trials, the analysis method must provide this control. Adaptive methods for time to event data are in great demand, particularly those that allow interim design modifications based on all interim data available, which includes (1) event times from patients that experienced the event and (2) preliminary information from others, such as surrogate endpoints. We will develop analysis methods that permit use of the Cox and Fine-Gray models for covariate-adjusted treatment evaluation of time to event endpoints, while allowing use of the full interim data for modifications to enhance the flexibility of these trials. This will use the conditional rejection probability principle introduced by Muller and Schafer 2004.

2. Efficient Covariate-adjusted Tests for Treatment Effect in Sequential Clinical Trials
When comparing study therapies in clinical trials, it is sometimes preferred to evaluate the marginal, population-wide treatment effects. A test or regression model is often employed without adjustment for potentially influential covariates in performing this assessment. But, the failure to account for these factors can adversely impact type I or II error rates, even in randomized trials. Incorporating covariate information is known to boost efficiency and avoid undue covariate influence. Zhang, Tsiatis, and Davidian 2008 proposes a method for evaluating the marginal treatment effects with covariate adjustment, using semiparametric theory to demonstrate consistency, efficiency, and robustness of the adjustment technique. We will extend usage of this method to the group sequential setting, making available the benefits of both sequential analysis and this covariate adjustment method for clinical trials. Tests will be derived for continuous, binary, and time to event endpoints. Extensive simulations will compare the efficiency gains provided by the proposed tests to existing methods.

3. A Unified Framework to Sample Size Determination for Generalized Linear Mixed and Frailty Models
The presence of correlation between observations often arises in clinical trials and observational studies. This may arise through repeated measurements on subjects over time or clustering produced by factors shared between subjects, such as treatment center or heredity. Regression models with mixed effects are often employed to account for this correlation while assessing a main variable of interest. Sample size calculations in this setting are often performed using simulation, requiring extensive computation and making assumptions about the data that are difficult to verify at the design stage. The class of existing sample size formulas for this setting is small, however, and so a simulation-based approach is often unavoidable. We will extend our previous work on sample size estimation for generalized linear and time-to-event regression to their mixed model counterparts, generalized linear mixed and frailty models, by introducing formulas for these models and providing theoretical support for their usage. The proposed formulas will be able to account not only for the mixed effects, representing correlation of outcomes within/between subjects, but also for any correlation between the main variable of interest and other covariates.

4. A Modified Sequential Probability Ratio Test for Safety Monitoring in Clinical Trials
Monitoring of safety events is an essential component of clinical trial support, with the observation of an excessive number of events used as a trigger for holding or closing the study. This is often implemented using a sequential testing procedure that compares an expected, null event rate to an alternative rate considered unacceptably high. The BMT CTN has used a modified Sequential Probability Ratio Test (SPRT) for monitoring its trials where only the upper stopping boundary, rejecting the null and indicating excessive risk, is used. One version of this scheme, the censored exponential SPRT, accounts not just for occurrence of adverse events but also their timing by assuming an exponential distribution for these events. In this study, we will describe and demonstrate this modified SPRT using real clinical trial data. Its operating characteristics will be compared to other common frequentist and Bayesian methods used for safety monitoring.

References

[1] Martens MJ and Logan BR. A Group Sequential Test for Treatment Effect Based on the Fine-Gray Model. Biometrics. 2018;74(3):1006-13.

[2] Martens MJ, Logan BR. Group Sequential Tests for Treatment Effect on Survival and Cumulative Incidence at a Fixed Time Point. Lifetime Data Analysis. 2019. doi: https://doi.org/10.1007/s10985-019-09491-z

[3] Martens MJ, Logan BR. A Unified Approach to Sample Size Determination for Common Nonlinear Regression Models. Under review by Statistics in Medicine.

[4] Shi Y, Martens M, Banerjee A, and Laud P. Low Information Omnibus (LIO) Priors for Dirichlet Process Mixture Models. Bayesian Analysis. 2019;14(3):677–702. doi:10.1214/18-BA1119.

[5] Pidala J, Hamadani M, Dawson P, Martens M, Alousi AM, Jagasia M, Efebera YA, Chhabra S, Pusic I, Holtan SG, Ferrara JLM, Levine JE, Mielcarek M, Anasetti C, Antin JH, Bolanos-Meade J, Howard A, Logan BR, Leifer E, Pritchard TS, Horowitz MM, MacMillan ML. Randomized Multicenter Trial of Sirolimus vs. Prednisone as Initial Therapy for Standard Risk Acute GVHD: BMT CTN 1501. Blood 2020; 135 (2): 97-107. doi: https://doi.org/10.1182/blood.2019003125

[6] Pidala J, Martens M, Anasetti C, Carreras J, Horowitz M, Lee SJ, Antin J, Cutler C, Logan B. Factors Associated with Successful Discontinuation of Immune Suppression after Allogeneic Hematopoietic Cell Transplantation. JAMA Oncology. 2020;6(1):e192974. doi:10.1001/jamaoncol.2019.2974

[7] Pidala J, Hamadani M, Dawson P, Alousi AM, Jagasia M, Efebera YA, Chhabra S, Pusic I, Holtan SG, Ferrara JLM, Levine JE, Anasetti C, Pritchard TS, Martens M, Horowitz MM, MacMillan ML. Randomized Multicenter Trial of Sirolimus vs. Prednisone as Initial Therapy for Standard Risk Acute GVHD: BMT CTN 1501. Presentation at the 2019 Transplantation & Cellular Therapy Meetings. Houston, Texas.

[8] Reshef R, Saber W, Bolanos-Meade J, Chen GL, Chen Y-B, Ho VT, Ponce DM, Nakamura R, Martens MJ, Hansen JA, Levine JE. Acute Gvhd Diagnosis and Adjudication in a Multicenter Trial - a Report from the BMT CTN 1202 Biorepository Study. Presentation at the 2019 Transplantation & Cellular Therapy Meetings. Houston, Texas.

[9] Muffly L, Pasquini MC, Martens M, Brazauskas R, Zhu X, Adekola K, Aljurf M, Artz A, Bajel A, Ballen KK, Battiwalla M, Beitnjaneh A, Cahn J-Y, Carabasi M, Chen Y-B, Chhabra S, Ciurea SO, Copelan EA, D’Souza A, Edwards J, Freytes CO, Fung HC, Gale RP, Giralt SA, Hashmi SK, Hematti P, Hildebrandt GC, Ho VT, Jakubowski AA, Lazarus HM, McCarthy PM, Olin RL, Olsson R, Rezvani A, Rizzieri DA, Seftel M, Seo S, Sorror ML, Szer J, Wood, WA. Increasing Use of Allogeneic Hematopoietic Cell Transplantation in Patients Aged 70 Years and Older in the United States. Blood. 2017;130(9):1156-64.

[10] Segal E, Martens M, Wang HL, Brazauskas R, Weisdorf D, Sandmaier BM, Khoury HJ, de Lima M, and Saber W. Comparing Outcomes of Matched Related Donor and Matched Unrelated Donor Hematopoietic Cell Transplants in Adults with B-Cell Acute Lymphoblastic Leukemia. Cancer. 2017;123:3346-55.

[11] Wingard JR, Wood WA, Martens M, Le-Rademacher J, Logan B, Knight JM, Jacobsen PB, Jim H, Majhail NS, Syrjala K, Rizzo JD, and Lee SJ. Pre-transplant Exercise and Hematopoietic Cell Transplant Survival: a Secondary Analysis of Blood and Marrow Transplant Clinical Trials Network (BMT CTN 0902). Biology of Blood and Marrow Transplantation. 2017;23(1):161-164.

[12] Cornell RF, Bachanova V, D’Souza A, Ahn KW, Martens M, Huang J, Al-Homsi AS, Chhabra S, Copelan E, Diaz MA, Freytes CO, Gale RP, Ganguly S, Hamadani M, Hildebrandt G, Kamble RT, Kharfan-Dabaja M, Kindwall-Keller T, Lazarus HM, Marks DI, Nishihori T, Olsson RF, Saad A, Usmani S, Vesole DH, Yared J, Mark T, Nieto Y, and Hari P. Allogeneic Transplantation for Relapsed Waldenstrom Macroglobulinemia and Lymphoplasmacytic Lymphoma. Biology of Blood and Marrow Transplantation. 2016;23(1):60-66.

[13] Knight JM, Syrjala KL, Majhail NS, Martens M, Le-Rademacher J, Logan BR, Lee SJ, Jacobsen PB, Wood WA, Jim HSL, Wingard JR, Horowitz MM, Abidi MH, Fei M, Rawls L, and Rizzo JD. Patient-reported Outcomes and Socioeconomic Status as Predictors of Clinical Outcomes Following Hematopoietic Stem Cell Transplantation: A study from the BMT CTN 0902 Trial. Biology of Blood and Marrow Transplantation. 2016;22(12):2256-2263.

[14] Dhakal B, D’Souza A, Martens M, Kapke J, Harrington AM, Pasquini M, Saber W, Drobyski WR, Zhang MJ, Hamadani M, and Hari PN. Allogeneic Hematopoietic Cell Transplantation in Multiple Myeloma: Impact of Disease Risk and Post

Publications