Articles

Newsletter Summer 2019 - Alan Fisher, Experiences as a Data Monitoring Committee Biostatistician

Experiences as a Data Monitoring Committee Biostatistician 

Alan Fisher, DrPH 

Biostatistics 

This email address is being protected from spambots. You need JavaScript enabled to view it.  

Data monitoring committees (DMCs) monitor the data that accumulateover the course of clinical trials that address major health outcomes. The intent of a DMC is to assure the safety of study subjects and the integrity of the study and study results. I have had the opportunity to be involved in several DMCs, ranging from Phase II safety studies with about 100 subjects to Phase III and IV efficacy studies with several thousand subjects. 

Virtually all DMCs include physicians with clinical expertise in the medical area being studied and abiostatistician experienced in clinical trials and group sequential techniques and witha clinical understanding of the disease area. The committee may also include a bioethicist and a representative of the patient community. 

DMC biostatisticians are generally perceived to primarily interpret the results of interim analyses.However, DMC biostatisticians encompass much more than that. Below, I describe some of my experiences that illustrate the biostatistician’s importance as an integral team member. 

In one of my experiences,I was the reporting statisticianfor a major Phase III trial in which time to death was a primary endpoint. Unlike most current DMC charters in which members are unblinded throughout the course of the study, the DMC charter in this situation specified that the members and the reporting biostatistician be blinded until the DMC voted that it was necessary to break the blind.   

As the reporting statistician,I ran the listings, figures, and summary tables for each interim analysis. For each analysis,I received from the CRO responsible for the data management and randomization an updated randomization file list that displayed whether subjectswere randomized to group A or B. One of the quality control procedures I employed was to check the output for each interim report analysis to that of prior reports. In one of the analyses,I saw that some of the previously entered subjects were now assigned to a different treatment group.This blatant error of an inaccurate randomization list was quickly recognized and then rectified. (Unfortunately, and particularly with the advent of IVRS (interactive voice randomization systems), and now IWRS (interactive web-based randomization systems), the occurrence of incorrect randomizations that result in destroying the integrity of the study isunfortunately not that uncommon. This is further exacerbated by errors in on-demand shipping and kit randomization lists.) 

Many members of this DMC were also serving on NIH committees, and I was exposed to best practices. For example, there was an open session, in which the sponsor and key investigators were present, and a closed session that included the reporting statistician and the DMC members. The DMC requested that the pages of the open and closed reports be bates stamped (a technique now replaced by converting files in a report to PDF and merging the files and numbering the pages with Adobe Acrobat). This allowed for an expedient review of the data during the meetings.   

As another example, amajor focus for the committee was to ensure data accuracy and timeliness and to adjudicate endpoints on an ongoing basis. I preparedan addendum of any important events, such as deaths and serious adverse events (SAEs) that occurredbetween closure of the database and the meeting. More importantly, having a DMC in place allowed us to conduct a sample size re-estimation toward the end of the study. It was recognized that the discontinuation rate, one of the assumptions in the sample size calculation for survival analysis techniques, was lower than expected. By taking this into account, and without breaking the blind, the sample size was reduced and the study was terminated earlier than anticipatedsimply by modifying this one parameter. 

In other situations, in which I was the DMC voting biostatistician, the reports to the DMC were prepared by a group within the same company, either a pharmaceutical firm or a CRO. This reporting group was also responsible for the final study report. In each situation, this group had to be explained to that the standard data listings, tables, and listings that are in a final clinical study report do not meet the needs of a DMC. A DMC often needs visual output, particularly box and whisker plots for safety data. Data must be current, a need that is met by electronic data capture (EDC) systems, whichfocus on entering data immediately. Moreover, for DMC reports, data sets need not be CDISC compatible and analysis data sets may not be required, which accelerates the time to produce a DMC report.   

In my experience, the initial set of output prepared by the reporting group often has programming errors and errors in the calculations of statistical summaries. I therefore recommend that templates be prepared before the first report that is presented to the entire DMC,and that these reports include an initial set of data from the study. In some situations, I have been able to work with the reporting group so that output presented to the DMC avoids such issues. 

Another situation I was in reinforced the need to always be skeptical about the accuracy of the data.In reviewing the safety data for a Phase II safety study, it was noted that there was a dramatic doseresponse relationship for a specific important laboratory parameter. If the data were valid, the concern for the safety of subject participants might have required a recommendation to either halt enrollment or terminate the study, both ofwhich would be major issuesfor the company. Instead, and particularly since the data were entered at the site by electronic data capture (EDC), we requested copies of the case report forms (CRFs). Interestingly, we uncovered transcription errors at several sites, and these errors were predominantly for subjects of the high-dose group. After correcting these transcription errors, there ended up being no dose response relationship. The DMC also recommended to the sponsor that the data be reread at a central lab. 

In these situations, the DMC biostatistician serves, to some extent, as an independent biostatistician to the client. DMC charters typically mandate review of the protocol and focus on the integrity of the study and the study results. The DMC statistician mustnot be a consultant to the company for the studies in which he/she is on a DMC, as this presents a conflict of interest. In my experience, I have resigned from being a consultant for projects in which I was a DMC member. 

In conclusion, I and my colleagues on the DMCs that I have served on recognize the gravity of our assignment. In each situation, the relationship between the client and the DMC has been cordial, and the client has recognized the value added by incorporating a DMC into their study program. We at Kinexum can advise you onhow to achieve such a relationship in the development of your projects.