Home » Additional Features, ASA Leaders Reminisce, Featured

ASA Leaders Reminisce: Jonas Ellenberg

1 June 2016 237 views No Comment
In the 18th installment of the Amstat News series of interviews with ASA presidents and executive directors, we feature a discussion with 1999 ASA President Jonas Ellenberg.

Jonas H EllenbergJonas H. Ellenberg earned his BSc in economics from the University of Pennsylvania’s Wharton School in 1963, his AM in mathematical statistics from Harvard University in 1964, and his PhD in mathematical statistics from Harvard University in 1970. He joined the biostatistics faculty at the University of Pennsylvania in the fall of 2004 as professor of biostatistics and associate dean for research program development in the school of medicine.

Ellenberg’s collaborative research has focused on neurological diseases, and more recently, HIV/AIDS and cardiovascular disease. He spent 26 years at the National Institute of Neurological Diseases and Stroke, NIH, with 11 of those as chief of the biometry branch.

With medical colleagues, Ellenberg performed extensive analyses of the Collaborative Perinatal Project, a longitudinal study intended to identify the etiology of serious childhood neurological illnesses and conditions. In addition to major medical findings related to causes of cerebral palsy and the significance of febrile seizures, this collaborative work led to important methodological insights regarding the conduct of longitudinal research, particularly relating to selection bias and generalizability of study results.

Ellenberg recently completed his leadership of the Clarification of Optimal Anticoagulation through Genetics, or COAG, study, which evaluated the use of genetic-guided dosing of Warfarin for use in anticoagulation, and the Prematurity and Respiratory Outcomes, or PROP, a structured follow-up study of very premature infants to assess pulmonary function. From 1995 to 2004, he served as vice president of Westat, Inc., and headed its biostatistics group. His collaborative research during this period focused on HIV in adolescents.

Ellenberg is an elected fellow of the ASA, Society for Clinical Trials, and AAAS, as well as an elected member of the International Statistical Institute. He served as president of the American Statistical Association in 1999 and the International Biometric Society in 1988. He is now professor emeritus in the department of biostatistics and epidemiology at the Perelman School of Medicine, University of Pennsylvania.

You began your education by studying economics and eventually moved into statistics. What motivated you to change disciplines?

My family business background, steeped in the New York City 7th Avenue textile trade, led me to the Wharton School at the University of Pennsylvania as an undergraduate. I chose statistics as a major, as I had always enjoyed math. Wanting to earn extra money while at Penn, I signed on to the University of Pennsylvania Periodic Health Examination project under the tutelage of Stanley Schor. The project was designed to evaluate the ability of longitudinal exams to detect undiagnosed cardiovascular and malignant diseases in corporate executives and the predictive ability of clinical and laboratory tests to quantify lethality of disease. The study was exclusively male since, at the time, the concept of female executives was incorrectly considered by many as an oxymoron. This experience introduced me to the application of statistics to medicine. You can find details in the article “Periodic Health Examination. Nature and Distribution of Newly Discovered Disease in Executives” in volume 172 of the Journal of the American Medical Association.

W. G. Cochran served as your thesis advisor at Harvard. What was his general approach to working with students on their thesis research? Does any single lesson you learned from him endure in your memory?

Cochran was both British and in the forefront of biostatistics when I entered the math stat department at Harvard. This and the fact that I had never taken advanced calculus or above prior to entry made me quite intimidated and [put me] in catch-up mode with my cohort. In spite of his warm, welcoming, and cozy demeanor, I was afraid of him and reluctant to ‘pester’ him with issues on my thesis—a test for outliers in multivariate regression. As a result, it was only after major milestones in my work that I felt comfortable meeting with him. This, of course, was a great error on my part, and while it served as a lasting lesson about self-confidence and assertiveness, it was my great loss to not have worked with him closely.

Beginning in the late 1970s, you and Karin Nelson published an extensive series of papers out of the National Institutes of Health (NIH) on the etiology of neurologic disorders in children. What impact did these results have on your career and the medical community?

My first position out of graduate school was at the then-named National Institute for Neurological Diseases and Blindness, now the National Institute for Neurological Diseases and Stroke, or NINDS. These results were the beginning of about 25 years of collaboration with Karin Nelson, a major pediatric neurological scientist and close colleague, as well as other medical colleagues on the evaluation of prenatal, perinatal, and early developmental risk factors for cerebral palsy and convulsive disorders from the Collaborative Perinatal Project, or CPP, database that recruited participants from 1959 through 1965.

First a digression to talk about the CPP. In the CPP, approximately 54,000 pregnant women admitted to 12 selected hospitals across the United States were followed through pregnancy, and their offspring were followed through seven years of life. The large longitudinal data set on the women and their children’s detailed examinations over years of life was, at the time, extraordinary. One element that highlights the meticulous and insightful planning of the CPP was the inclusion of a CPP nurse in the delivery room who was responsible only for capturing the critical data during delivery and birth. Using the data from the CPP, our collaborations forced the rethinking of many established medical paradigms.

Now on my data analytic involvement. My involvement with the CPP data analysis was in the early ’70s, and this was my first encounter with very large longitudinal data sets and the unique statistical issues they presented. Setting the stage in which we worked then: data were entered manually on punch cards, and lugging around 11 x 14-inch continuous computer printouts was a charm of yesteryear. Both the speed of computers and the availability of software to implement the then-new and ground-breaking statistical methodological developments were relatively primitive. The privacy of patient medical records in panel studies mostly relied on the integrity of investigators. Such protections were later codified into law by the HIPAA legislation in 1996.

Next month, we will feature an interview with 1996 American Statistical Association President Lynne Billard.

The general research, which began with the papers you ask about, led to the conclusion that cerebral palsy, which at that time was widely believed to be due to problems during labor and delivery, was in fact due largely to factors occurring prior to labor and delivery—thereby upsetting the large cadre of lawyers whose incomes sprang from litigation against obstetricians for bad pregnancy outcomes.

At the time, the conventional wisdom was that fetal loss of oxygen, or asphyxia, during labor and delivery caused brain damage that was highly related to cerebral palsy—read caused. The data supporting this wisdom was generated largely from retrospective studies with highly selected samples and heterogeneous definitions of both putative risk factors and outcomes. A perhaps natural, but in hindsight, inappropriate response to this conventional wisdom was the increased use of C-sections and the introduction and rapid growth of the use of electronic fetal monitoring during labor. Both of these expensive actions were justified by the perceived need to prevent asphyxia and the eventual outcome of CP. Also note that C-sections are not without risk.

In the area of convulsive disorders, we showed that a febrile seizure, a convulsion occurring in the presence of very high fever with a fairly common occurrence in infancy—approximately 1 in 20 children—was not a risk factor for epilepsy, seizure disorders, or mental retardation, except in a very small and well-defined subset of children. This subset consisted of those children who were neurologically abnormal prior to their first febrile seizure, had a family history of seizures, or had a first complex febrile seizure. This result called into question the then-common practice of treating all children with febrile seizures with neuroactive drugs such as phenobarbital, or Pb, for extended periods—as long as two years—to prevent further seizures. Pb was not a benign drug in children; it causes hyperactivity—it is not calming, as it is in adults—so was not the best thing for children who were going through the terrible twos. These findings in observational data, while highly concerning, required confirmation, so we persuaded the institute leaders to provide funding for a randomized trial to study the efficacy and safety of Pb in the prevention of febrile seizures. The clinical trial results showed, first and somewhat surprisingly, that Pb actually did not prevent febrile seizures and, second, and just as importantly, that after a year of treatment, the children receiving Pb had lower IQ scores than the children who had had the good fortune to be assigned to receive the placebo.

On a personal note, my mother, who had expected me to go into business like my father and uncles, never understood what I did for a living. When asked what her son was up to, she just reported that he “worked for the government.” When our results on febrile seizures were reported by Gina Kolata and appeared above the fold on the front page of The New York Times on February 8, 1990—right next to the article describing the fall of the Soviet Union, which had occurred the day before—she received numerous excited phone calls from everyone she knew in New York City. After that, she still didn’t understand what I did for a living, but recognized that I might be doing something worthwhile, even though I wasn’t “in business.”

With regard to the impact on the medical community, our febrile seizures research showed that these events are basically benign for all but a tiny subgroup of children who have them, and that this subgroup can be readily identified. In addition, our clinical trial showed that active medical treatment was both ineffective and harmful with respect to its serious cognitive side effects. Despite this hard evidence and the replication of our results over the years, it took almost two decades before these results were incorporated as part and parcel of standard of care in the management of febrile seizures.

The impact of our results on the etiology of cerebral palsy is interesting. Investigators continue to proffer data that are claimed to be contradictory to the CPP findings in terms of the putative increased risk of cerebral palsy related to labor and delivery issues (e.g., asphyxia). However, these new results use the same questionable design approaches as the older studies. C-section rates and the use of electronic fetal monitoring, or EFM, have both increased despite the lack of evidence for their effectiveness in preventing harm; many would say their increased use reflects a defensive medicine practice to combat potential malpractice suits that continue to arise following the delivery of an infant with cerebral palsy. Despite the routine use of EFM and the increase in nonmedically required C-sections, the rate of cerebral palsy has not decreased over time. The litigation against obstetricians continues, although perhaps less frequently, due to the Supreme Court decision on Daubert v. Merrell Dow Pharmaceuticals, Inc. disallowing evidence based on ‘junk’ science into the courtroom.

This CPP observational database was extraordinary for its time, and is so even today. An attempted sequel—the National Children’s Study, or NCS—was a recent multi-billion-dollar effort to reinvent the CPP with a focus on the impact of environmental factors on childhood development. It was authorized by the Children’s Health Act of 2000 and undertaken with a goal of developing a randomized approach to sampling pregnant women that would have allowed the most reliable conclusions. Ultimately, the NIH leadership decided to close the study, commenting that “… When recruitment ended in July 2013, the Vanguard Pilot Study had enrolled approximately 5,000 children in 40 locations across the country. The planned NCS Main Study would have followed 100,000 children from before birth to age 21. However, the NIH director decided to close the NCS on December 12, 2014, following the advice of an expert review group.”

Mike O’Fallon, left, with Jonas Ellenberg

Mike O’Fallon, left, with Jonas Ellenberg

I began with great hopes and expectations for the NCS and was a great supporter, and even an early-stage contractor—note that my contractual role was not continued by the NCS during a re-envisioning of the study. I later served three years on and then resigned from the NCS Advisory Board, or NCSAB, after a polite but somewhat rocky tenure related to study design issues. My resignation from NCSAB was reported in Science in March of 2012: “In an email dated 16 March, the University of Pennsylvania’s Jonas Ellenberg submitted his resignation to NCS Director Steven Hirschfeld at the National Institute of Child Health and Development, or NICHD. His note contains no explanation but says: ‘I strongly urge that the NCS be reviewed a second time by the Institute of Medicine, since I believe that the current NICHD view of the NCS does not reflect the parameters of study design reviewed and endorsed by the IOM in 2008.’ The reference is to an IOM report that commended NCS’s plan to recruit pregnant women living in a statistical sample of about 100 U.S. counties.” I should mention that this was to be my first and only mention of any sort in Science.

In addition to statistical expertise, what other skills did you need to succeed in leadership positions at the NIH’s Neurology Institute and Westat? Did you feel adequately prepared, and if not, what did you do to develop these additional skills?

Biostatisticians often work within organizational structures that may not fully recognize the benefits of statistical collaboration, and as such may not accord organizational stature, resources, or recognition to our input. With the exception of Westat, with its focus on the statistical arena, I have always worked within medical hierarchies. Winning over both leadership and my medical and other nonstatistical colleagues was often difficult, and I did not feel well prepared for this aspect of statistical collaboration by my doctoral training in mathematical statistics. What worked well for me in many circumstances was to use examples in demonstrating the worth/value of statistical input, in contrast to making arguments based on abstract statistical paradigms. I developed, from the literature and from my own experience, a series of examples of completed studies with poor statistical design that resulted in the waste of resources—both human and monetary—and/or resulted in less-than-useful conclusions. This approach tended to be more persuasive than arguments from basic statistical principles.

What I remember about mid-level leadership at NIH came in two arenas. The first was staffing. I knew that support of junior colleagues was of paramount importance and also provided major gratification. My two final hires, Paul Albert and Lisa McShane, were wonderful colleagues who have both made and continue to make extraordinary contributions to statistical and medical science—unfortunately for NINDS, in other NIH institutes.

The second arena was dealing with resources, both staff positions and funding for our research. I took the approach in lobbying for both resources of being honest and direct about our potential accomplishments and projecting expenses as exactly as possible, allowing for the possibility of exceeding or missing goals in both areas. I can’t say this was an enormously successful approach, but it seemed to provide the branch with the resources it needed.

What were the highs and lows of your term as president of the ASA?

I was delighted to be able to follow through with an initiative undertaken during my presidency of the International Biometric Society in 1988 and 1989 to make the Journal of Agricultural, Biological, and Environmental Statistics a reality—primarily through Linda Young’s efforts. The IBS did not then have the resources to maintain this journal on its own; it seemed to me important for the ASA to support a journal that could report innovative applied work in important scientific areas that would probably not find a home in more theoretical journals such as JASA and Biometrics.

Less substantively, I am also pleased that my initiation of wearing formal attire at the Tuesday evening JSM session has been taken up by most ASA presidents since. I wanted a visible way to express the statistical community’s acknowledgements of the year’s awardees, particularly the newly elected fellows of the ASA.

With benefit of hindsight, I wish I had put more emphasis on what is now an element in the ASA Strategic Plan under education strategies: Develop and implement a plan to influence the inclusion of statistical thinking in science and computer science. In considering our professional involvement in genetics and data mining and other new areas of science, I believe the following questions remain:

  • Are the current organizational homes for either statistical genetics or data mining appropriate for attaining high-quality statistical input as true collaborators; or in the extreme, is statistical thinking in these areas considered merely a technical assist on an as-needed basis? I note in this regard a current headline on the ASA website stating that according to Careercast.com, “the best job of 2016 is data scientist, while statistician comes in at number two.” Why isn’t data science a subgroup of statistics?
  • Is our profession in a position to aggressively lay claim to and lead the collaborative development in new areas such as these? This is neither a new concern nor one that is being ignored by the profession. In discussion with colleagues, there is the strong belief that we must be present for these emerging team science domains, especially for statistical design, inferential framework, and analytical methods that investigate confounding and causality.
1 Star2 Stars3 Stars4 Stars5 Stars (No Ratings Yet)

Leave your response!

Add your comment below, or trackback from your own site. You can also subscribe to these comments via RSS.

Be nice. Keep it clean. Stay on topic. No spam.

You can use these tags:
<a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

This is a Gravatar-enabled weblog. To get your own globally-recognized-avatar, please register at Gravatar.