Dr. Robert Byers

CHAMBERLAND: This is Dr. Mary Chamberland, and I'm here with Dr. Robert Byers at the Centers for Disease Control and Prevention (CDC) in Atlanta, Georgia. Today is Tuesday, June 12, 2018. I'm interviewing Dr. Byers as part of the oral history project The Early Years of AIDS: CDC's Response to a Historic Epidemic. Bob, welcome to the project. Do I have your permission to interview you and to record this interview?

BYERS: You do.

CHAMBERLAND: Bob, you first came to CDC in 1984 as a mathematical statistician in the AIDS [acquired immunodeficiency syndrome] Program. You stayed nearly 20 years before retiring in 2005, so quite a run! To start off our conversation this morning, let's talk a bit about your pre-CDC life. Could you tell us where you grew up and about your early family life?

BYERS: My family moved to Atlanta when I was six years old. I grew up in the hinterlands out near Avondale [Estates] and went to Avondale High School. After 00:01:00that it was natural to go to Emory [University] to study. I studied sociology there and started graduate school in sociology. I discovered that I was really interested in what they called methodology, which is basically statistics. As luck would have it, the Department of Biometry and Statistics was just getting started. That has grown into the Rollins School of Public Health since then, but at the time it was Biometry and Statistics.

CHAMBERLAND: That's right. At that time the School of Public Health didn't exist.

BYERS: Right.

CHAMBERLAND: Exactly. Had any one in your family had any statistical background, or was this just something, as you said, that appealed to you personally?

BYERS: I was lucky to have some really good math teachers at Avondale High School, including Ms. Mary Lallerstedt, who was so well known among the mathematicians at Emory that they let me skip the first algebra course. He said, "Oh, you had Mary Lallerstedt. Okay, you can go right into calculus." I said, "Okay." That was fun.

CHAMBERLAND: Then after you obtained your Ph.D., you stayed on at Emory for--

BYERS: Ten years.

CHAMBERLAND: Ten years. What sort of projects were you working on at Emory?

BYERS: I was working for the Emory University Computing Center. I would consult with people who had come to get their data analyzed, so there were all kinds of projects over a ten-year period. I remember one time I had 15 different projects going at one time, and I was working half-time --

CHAMBERLAND: Half time?

BYERS: Yes, to finish my dissertation. I finally got that done in '74. Okay, then it was ten years after that that I was called up at CDC.

CHAMBERLAND: You said, "called up." I guess you need to explain a little bit about the circumstances of your arrival at CDC.

BYERS: At the Emory University Computer Center, we had a lot of really smart students, some even in high school, who were working on our computer and doing programming for us. One of those was [Dr. W.] Meade Morgan, who called me up one morning and said, "How would you like to work at the CDC on the AIDS project?" This was like 8:00 in the morning on a Monday. I think I was signed up by Friday, which was really fast, I found out. Yes, Meade called me because he knew me. They had a statistician who had just left, so as far as I know, I'm I guess 00:04:00technically the third statistician on the project.

CHAMBERLAND: Meade at that time was the lead, if you will, for statistics and computing in the AIDS Program.

BYERS: Yes.

CHAMBERLAND: And he had been a student--

BYERS: Yes.

CHAMBERLAND: -- at Emory.

BYERS: At Emory.

CHAMBERLAND: It sounds like you didn't think too long and hard about making the move.

BYERS: I'd been looking for a couple of years.

CHAMBERLAND: Okay. Okay. Did you know a lot about the AIDS Program before you got there?

BYERS: I didn't even know anything about AIDS. It was quite a learning curve at the beginning.

CHAMBERLAND: What was Meade's pitch to you?

BYERS: Nothing much. He just said, "Would you be interested in doing this?" and I said, "Yes, let's talk." He and I went out to lunch, and we talked about it. He told me what they were doing, and I just wanted to leave Emory. I'd been there ten years; that was long enough.

CHAMBERLAND: When you came on board, it was essentially you and Meade, when you first started, just the two of you.

BYERS: Yes. Right.

CHAMBERLAND: You really were the pioneers of getting that program up and running. Maybe this is a good time to pause a little bit, before we go into the specifics of your work with the AIDS Program, to have you set the scene for us so to speak. Tell us a little bit about what the state of information technology (IT) and computing was in the mid- to late-1980s. I'm specifically interested, upon your arrival at CDC, what you found, and how that compared to the work environment you were used to at Emory.

BYERS: Just before I left Emory, Emory had spent a million dollars to upgrade 00:06:00their computer with one megabyte of memory. Within ten years you could put a megabyte on a floppy disc. The memory was called core memory, because it was little metal donuts with wires running through it. It was very expensive stuff. When I moved to CDC, they had a much bigger computer. I was very impressed. They had four megabytes of memory. It was just at the beginning of desktop computers, the IBM [International Business Machines] PC [personal computer]. Windows had not been invented yet. Mostly the desktop computer was used as an interface to the mainframe.

6:49

CHAMBERLAND: Most of us in 1984, most of the rank and file, so to speak, at CDC did not have a desktop computer. We were in the position of vying for time on 00:07:00the secretary's word processor, the first generation of which CDC had were Wangs [Wang Laboratories computer company]. We all, as I said, on her lunch break or after hours, would try and get a chance to sit down and type up some documents. I'm assuming you had some sort of computer at your desktop.

BYERS: I did, and the most wonderful thing about it was a spreadsheet, because I could do arithmetic on it.

CHAMBERLAND: That's pause for thought, because these are all things now that have just catapulted in terms of capacity and diversity of software and the like, so these were early days.

BYERS: You bet.

CHAMBERLAND: I have very fond memories of working in Building Six, which is where the AIDS Program was housed in those days, and of Meade Morgan literally 00:08:00crawling through the ceiling tiles stringing what I guess -- computer cables? -- to get us connected. Did you ever get involved in some of these "hardwiring projects?"

BYERS: Actually, I missed all that, and I'm glad, because it doesn't sound like fun to me. But Meade enjoyed doing things like that.

CHAMBERLAND: He did?

BYERS: He really did, and being sure that the nuts and bolts were working. At one point he had his own computer in his own little computer room, which was about the size of a closet. He always loved to check the small details in getting stuff going.

CHAMBERLAND: All right, that's the base that you're starting [at], and you're impressed by what facilities CDC had.

BYERS: Oh, one other thing. If you wanted to print something out, you either printed it on the mainframe, which were the big wide sheets of double-fold paper, or else you had a dot-matrix printer at your desk. If anything was printed out that I printed out, it's faded into oblivion by now, because that was not a really good solution for printing. Eventually we got Xerox machines that would give a nice publishable printout, but it was pretty primitive.

CHAMBERLAND: My recall also was those dot matrix printers were extremely noisy--.

BYERS: Oh, yes.

CHAMBERLAND: -- and very distracting. If you were sharing an office and someone was printing, it was pretty unbearable.

BYERS: And slow.

CHAMBERLAND: And very slow. That was the state of affairs with respect to IT. 00:10:00What was the state of affairs at that time with respect to mathematic modeling for infectious diseases? Had modeling really been used a lot for infectious diseases? Had you done modeling as part of the work you were doing at Emory?

BYERS: Yes, in a sense. I wasn't familiar with what had been going on in epidemiology or disease work, so I had a lot to learn. That's one of the things I loved about this job, that I was always learning something new. After 21 years [on the job] I was still learning new stuff. First I had to learn the jargon. What is HIV [human immunodeficiency virus], what is seropositivity, what is -- there was a lot to learn. Everybody was very kind and helpful to get me up to speed.

Then I started working with [Dr. William W.] Bill Darrow and [Dr.] Harold [W.] Jaffe. The first two papers I got my name on, the first one was with Harold Jaffe as the lead author, and the other one was with Bill Darrow. Basically, in those days-- this was before the virus had been isolated; this was before a test for infection came around. There wasn't a lot to do other than to figure out how people were getting infected and what were the risky behaviors. That's what those two papers were about. One was about the San Francisco City Clinic Cohort, which was a gold mine of information for the whole time I was there.

CHAMBERLAND: It sounds like working on this project, various analyses related to the San Francisco City Clinic Cohort -- was that the first piece of work you 00:12:00were given, essentially, to focus on?

BYERS: Yes.

CHAMBERLAND: Maybe we need to back up a little bit. Can you tell us about what was the San Francisco City Clinic Cohort? Who were the people in the cohort, and why was it such a gold mine?

BYERS: The San Francisco City Clinic Cohort was a group of 6,700 homosexual men who were recruited to study hepatitis B. They were given serial blood tests starting in 1978, and they did that for a couple of years. Then they dropped the project for a couple of years until somebody realized -- until the test for HIV came along, and then somebody realized these are perfect blood samples for testing for HIV. They got about 400 men to volunteer to let their blood be 00:13:00tested and to do follow-up interviews. It became the City Clinic Cohort for HIV Studies.

13:26

CHAMBERLAND: The access to the men who volunteered, the sample of men from that large cohort, coupled with the availability of this stored blood, as you said, was a veritable gold mine.

BYERS: Yes.

CHAMBERLAND: What were some of the questions that you and your co-investigators were trying to answer using this cohort?

BYERS: The first one was how fast are people getting infected. That was the first paper I did on my own. Looking back, I found a copy, and I found a stack 00:14:00of requests for reprints that was like six inches deep, and a form letter saying, "Dear Sir, we're sorry, we've run out of reprints." That was a very popular paper. Basically, I fit four survival curves to the data, which had a little problem because you never knew when somebody was infected. All you knew is the first positive test and the last negative test, and there might be 15 months between them. What are you going to do about that? That's when I started learning, and I learned that there was a model that will take care of that.

CHAMBERLAND: You, on your own, actually had to do some research into modeling and different types of models to try and get the best model to fit the data that 00:15:00you had, if I understand that correctly.

BYERS: Always.

CHAMBERLAND: Always. In terms of those first few papers, you found some pretty amazing stuff with respect to the proportion of people in the cohort that had become infected?

BYERS: Yes. In fact, even now I am shocked and appalled. By the time we did the study in 1984, it appeared that 60% of the people in the cohort were already infected with HIV, and that was before people even knew HIV existed.

CHAMBERLAND: That's right. The cohort had been recruited for, I believe it was hepatitis B vaccine studies back in '78, long before those first cases were 00:16:00officially reported in the 1981 MMWR [Mortality and Weekly Report]. It was a window onto, as you say, as to how fast people were becoming infected. What were people's reactions when you looked at the data?

BYERS: They were properly horrified. Nobody knew that that was going on, because most of these people did not have symptoms. The only way you knew they were positive is you tested their blood.

CHAMBERLAND: Even before that, as you say, there wasn't a licensed test, I believe until early 1985. The only indicator, if you will, was the development of clinical illness, AIDS or some of the associated conditions that were being recognized, like lymphadenopathy syndrome in these men.

BYERS: Kaposi's sarcoma.

CHAMBERLAND: Kaposi's sarcoma, one of the opportunistic infections. That, as you say, was properly horrifying, to find that level of infection in the cohort. Who were some of the investigators that you worked with? You mentioned Harold and Bill. I know Bill was very interested in some of the sexual practices and if those provided any clues on risk factors -- what would possibly make men more likely to be infected with HIV. Who were some of the other folks that you worked with on the cohort?

BYERS: Naturally we had to work with some people at San Francisco. Nancy [A.] Hessol was an epidemiologist there, and [Dr.] Frits van Griensven was over from Amsterdam. They were fun to work with, and I got to go to San Francisco a couple 00:18:00of times. We got some papers out of that. Nancy was a hard worker and got a lot of publications from the cohort.

CHAMBERLAND: What were you learning, do you remember, in terms of survival, once people got sick or got infected? You were able to come up with some early projections--not projections, estimates on how long people survived. Is that correct?

BYERS: Yes. Nancy did a paper with me and Frits on the relationship between what she called -- or what I call incubation time, and she called it latency -- the time from infection until the symptoms, and the relationship between that and survival. The idea was the longer you've been infected, the poorer your chances 00:19:00of survival are. We did not find a significant relationship. I think there is one. I think it's clear that there is one. But we were early in the epidemic, and we didn't know that. This led to this question, "What is the mean incubation time?" The first paper that came out about that was, again, shocking. The estimate was five years. You could be infected five years before you show up with AIDS symptoms. People were used to thinking of incubation time in days, not years, so that was quite interesting. After that there was a cottage industry estimating incubation time; everybody who had a cohort started working on it. 00:20:00For years, until treatment came along, people kept estimating incubation time, and the mean incubation time kept going up. About the time people quit doing that, it was up to 15 years. Nobody ever knows, and we hope nobody ever will know, how long it really is. To think that you could go 15 years and not even know that you're sick and infecting other people is pretty scary.

20:40

CHAMBERLAND: It takes some doing to recall that these very basic things were--

BYERS: Nobody knew.

CHAMBERLAND: We didn't know, and you and your colleagues were on the cusp of chipping away at some of this very basic information about the disease and just 00:21:00being shocked at what you were seeing and how different it was to people's experience beforehand. Pretty amazing stuff. Along with the work that you were doing in the San Francisco City Clinic Cohort, there was a lot of interest in these early days, and continues to this day, in projecting the future, forecasting the future: how many cases. We know how many cases meeting this surveillance definition that we had established nationwide had been reported to date. Again, it was a time when cases were supposedly doubling. Every few months we would have twice the numbers of cases that we did. People got very interested in trying to forecast the future.

The very first projections were presented in 1986 at a meeting that the Public Health Service convened at the Coolfont Conference Center in West Virginia. At that time CDC estimated that by the end of 1991, five years down the road, the cumulative number of AIDS cases was going to be something like about 270,000. At the time of the meeting, the number of reported cases was about 21,000, so more than a tenfold increase in five years. I think, for a lot of people, these projections really caught them off guard. The unfolding of the epidemic was just staggering to comprehend.

Let's talk a little bit about modeling and projections. Were you involved in these first projection calculations that were made? I know Meade was.

BYERS: I think Meade was just doing that on his own.

CHAMBERLAND: I think it was a very simplistic approach that was being used at that time. Let's try and break this apart a little bit, because I think there will be a lot of nonstatisticians that will be looking at this interview and hearing your presentation. If you want to make projections about the course of the epidemic, what are the types of information data needs? What do you need just to even begin to think about doing this?

BYERS: It would be lovely to know how many people are infected.

CHAMBERLAND: OK, prevalence.

BYERS: Prevalence.

CHAMBERLAND: The prevalence.

BYERS: And the AIDS counts were misleading because of the long incubation time.

CHAMBERLAND: Right, OK, that's another piece.

BYERS: That means if you see 10,000 AIDS cases, there are probably 50,000 [infected persons] out there that don't even know they're sick yet. But you don't know that. [Dr. Ronald S.] Ron Brookmeyer from Johns Hopkins--

CHAMBERLAND: Another modeler?

BYERS: Another modeler. He was an epidemiologist; he's still working. He came up with this idea called back calculation. You've heard of that.

CHAMBERLAND: I've heard of it. Why don't you tell us just a little bit, the watered-down version, if you will. What do you need for back calculation?

BYERS: If you knew how many people were infected and if you knew the distribution of incubation times-- Then a further complication is reporting delay, because sometimes the reports didn't come in for a year and a half or 00:25:00more. Those three things together could be used to project. But since you don't know how many are infected, Ron had the idea, let's estimate the incubation time from the data we have. Then let's apply it to the AIDS cases we have and go backwards in time, and try to predict how many people we needed in order to generate this many AIDS cases. Everybody was doing back calculation for a long time.

25:35

CHAMBERLAND: Did that calculation have some limitations, if you will, in terms of projecting the future?

BYERS: Yes.

CHAMBERLAND: Far out? By far out, I mean is this method good for just projecting short term, a few years?

BYERS: Right. You couldn't go out very far, because you didn't know the 00:26:00incubation distribution even past the mean, so you were limited that way. Then there was another limitation that nobody seemed to realize. Everybody who was doing incubation estimation was coming up with different answers, and nobody wanted to talk about that. I made a plot and went to a meeting here in Atlanta. I showed the plot, and it showed all of these curves going all over the place. I explained what I had done and how I'd done it, and I wanted to know what people thought about it. People thought they should shuffle their feet. One person said, " I've noticed that." But nobody wanted to talk about it. Obviously, if your incubation distribution is wrong, then your projections are going to be wrong. Being statisticians, we always do confidence intervals around our 00:27:00estimates, and the confidence intervals were reassuringly broad, like plus or minus 50%.

CHAMBERLAND: I guess that gets to the maxim that I remember hearing, that your model is going to be only as good as the data that you put into it, the data you have available at the time. This, though, is a very dynamic time, not only with respect to the epidemic, because the epidemic is moving into different populations -- IV drug users, heterosexuals, transfusion recipients.It's also very dynamic in that data are coming in from studies, like the San Francisco City Clinic Cohort and other studies, saying, "Hey, this what we found," as you 00:28:00said, for incubation period and or whatever.

You mentioned a meeting, and I understand that -- in preparing for this interview I was looking at old MMWRs -- and the [U.S.] Public Health Service actually convened at least a couple of meetings in the mid- to late-'80s, trying to bring different modelers together. I believe the National Academy and the Institute of Medicine tried to do that as well. Tell us a little bit more. What's it like when you've got a bunch of modelers in a room trying to reach -- is consensus the right word?

BYERS: Yes. Yes. Basically, each one would stand up and explain what he or she had done, and then people would throw bricks at them. Then everybody would go home, unfortunately.

CHAMBERLAND: So often these meetings didn't end with a lot of agreement.

BYERS: Correct.

CHAMBERLAND: Correct. Were the different groups in a sense competing a bit with each other? Was there a sense of rivalry?

BYERS: Unfortunately, yes. The thing that bothered me most was people felt like they had a right to their data, and nobody else should be able to look at it. That is just slowing up progress. I never could get behind that kind of idea.

CHAMBERLAND: I wondered about that; that's disappointing to hear. Some of the key data for doing these back calculations are national AIDS surveillance data. Correct?

BYERS: Right.

CHAMBERLAND: -- the number of cases of AIDS that are being reported. CDC is also doing studies to look at what the delay is, from diagnosis of AIDS in an 00:30:00individual to its actually making its way as an official case report to CDC via the health department. All of them in a sense were beholden to CDC.

BYERS: Yes. Yes.

CHAMBERLAND: Did these other groups have access to other types of data?

BYERS: For example, the New York City Health Department had a cohort of their own; Amsterdam had a cohort of their own. Also for example, CDC had funded one of the consulting firms in Washington, D.C. [District of Columbia] to do some analysis and collect some data. The person in charge of that project apparently felt like that was nice that we were giving her money, but she didn't feel she 00:31:00needed to report anything back. Finally, we had to send two [medical doctors] MDs up there to sit there and say, "We want those tapes right now."

CHAMBERLAND: Oh, gosh.

BYERS: Yes. That was extreme. There were a lot of people like that: "We'll tell you what we've found from our data, but we're not going to let you see our data."

31:32

CHAMBERLAND: That, as you say, made things a lot harder.

BYERS: Right.

CHAMBERLAND: The other thing that is happening is-- things are advancing, there are antiviral-- treatments starting to become available, I guess that's in the late 80s, early 90s. That's good, obviously, that's very good, although those 00:32:00early drugs certainly had their own limitations. As the epidemic is maturing and other things start to come in play, was that making your job harder or easier?

BYERS: Harder. Things were just getting more and more complicated.

CHAMBERLAND: In the sense that you have to start putting information about these different factors into your model?

BYERS: Yes. Yes. In fact, sometimes it would just make you cross your eyes, listening to some of these presentations and how complicated they were. I always was wondering, is that really buying us anything extra or not? I never did decide.

CHAMBERLAND: Do you remember some of the factors besides the effect of antivirals? I think you actually worked on that area specifically, didn't you?

BYERS: Mm-hmm.

CHAMBERLAND: What impact did that have on back calculation? Were there other very important factors that had to be taken into account in those models?

BYERS: Reporting delay was a difficult one, because it was different in different parts of the country. Geographic differences could mess you up. At one point we thought, "Wonderful, the number of AIDS cases is dropping drastically." Nope, they just hadn't been reported because somebody somewhere had cut the budget for the health department. That was where they had decided to draw the line: they quit reporting their AIDS cases. I don't want to point any fingers, but that was terrible.

CHAMBERLAND: What was the effect of having the virus spread into other risk 00:34:00groups? In the beginning gay men constituted --and continue to -- the largest proportion, but initially that was pretty much where the epidemic was centered. Then there's a period of incredible discovery with IV [intravenous] drug users, their partners, transfusion recipients. What impact does that have when you're trying to make a projection for the epidemic as a whole?

BYERS: If all you're looking at are gay men, you've got a handle on how many there are. But when you're talking about IV drug users, which is essentially illegal, they're not going to report themselves. All you know is how many AIDS 00:35:00cases there are, and they're not being tested either, so that was difficult. I don't think anybody's found a way around that.

CHAMBERLAND: Do you have to take into account [the effect] that each of these risk groups or different populations is having on the dynamics of--- since HIV transmission is a little bit different in terms of timing or maybe how efficient it is? You've got a big outbreak occurring in gay men, and then other outbreaks in drug users, transfusion recipients. It strikes me that the transmission dynamics might be a bit different in each of those groups. Did you have to come up with some way to take that into account for a comprehensive national model?

BYERS: We tried. I looked into a lot of that. Questions like, "Is the incubation period the same for IV drug users?" You can imagine that it might not be, but I was never able to come up with data that would support anything. I just was not able to find anything.

CHAMBERLAND: That raises an interesting question. In fact, preparing for this interview I came across a quote which I think is particularly apt. This was a report that the Institute of Medicine put out in 1988. It said, "Modelers, data collectors and policy makers should meet regularly to ensure that modelers are asking questions for which data can be collected and that it's collected in a 00:37:00way that it's useful to modelers and policy makers."

BYERS: Yes.

CHAMBERLAND: You agree?

BYERS: I totally agree.

CHAMBERLAND: Were you able to work with colleagues at CDC in the AIDS Program or others external to CDC to have discussions about, "Hey, these are the pieces of data we need to make the models better. How can we put our heads together to design a study to get that?" Did those kinds of discussions happen?

BYERS: Yes, at the national meetings especially, and at the statistical meetings people would get together and talk about what needed to be done.

CHAMBERLAND: OK. Were you ever in a position yourself personally, where you were part of a discussion with policy makers, either at high levels within CDC or within the Public Health Service, where modelers were sitting at the table 00:38:00trying to assist in some decision-making that had to happen, and these decisions makers were looking to statisticians to give them--

BYERS: I don't remember that. I really don't.

38:25

CHAMBERLAND: It points to, I think, a problem or a challenge--

BYERS: A challenge.

CHAMBERLAND: --which is, in developing these models, communicating what the model can do or not do, or tell you or not tell you, with respect to the questions being asked. Can you elaborate a little about the challenge of putting the outcome of a model, if you will, in perspective for decision makers or even 00:39:00epidemiologists? As we talked about, the model is only as good as the data, and there's going to be some limitation. Is that a hard thing to explain to non-modelers?

BYERS: Yes, and even people who do modeling often confuse the model with the reality. The model is basically some equations that are supposed to mimic what's going on in real life, but it's not the real thing. I remember the first guy who published on incubation time was trying to convince an M.D. that his model was predicting when a fetus would catch HIV from its mother. The M.D. was saying, 00:40:00"But you don't have any data about that," and he's saying, "But this is the truth because this is a model." I'm like, uh, really? Yes, that happens a lot, and it's tough to -- especially when you're working with the same model over and over. You start to think this is reality. But it's not. If you don't have good data and if your model actually doesn't fit the data as well as it should, then you always have to be skeptical.

CHAMBERLAND: We've talked a little about the different groups that were out there. It sounds like there were a lot of groups, different institutions, that were involved in modeling the AIDS epidemic. How big a player was CDC in this 00:41:00modeling effort, do you think?

BYERS: I like to think we led the way.

CHAMBERLAND: All right, go, Bob!

BYERS: I think Johns Hopkins was very good. San Francisco City Clinic had some good people. UCSF, University of [California] San Francisco. Yes. There was a lot of work being done overseas in Thailand and Africa. I wanted to mention Africa, because that was one of the first things I did. I went to Zaire, which is now the Democratic Republic of the Congo, to Kinshasa. I worked for two weeks with Dr. Jonathan Mann on a study he was doing of prostitutes. I guess we say sex workers now. He came up with the first hard data [to support] that using 00:42:00condoms reduces the risk of HIV transmission, and that resulted in a letter to the New England Journal of Medicine. That was really interesting. Kinshasa was interesting. It was the first time I had ever been in a country where I was a discriminated-against minority, and it was instructive.

CHAMBERLAND: Before you went, just to have you expand on this a bit, Jon Mann had been doing this study with prostitutes. Was this like a case-control study he was doing? I'm just trying to figure out how it was--

BYERS: It was pretty basic. He would give them like 50 cents to let him take their blood, and they were happy to take it. That was a lot of money to them. He 00:43:00would follow up, and then he would interview them about what they did, what kind of sex acts, and whether or not they used condoms, of course, which was the most important thing. We found out two things that were very significant. If a woman had unprotected sex during her period, she was very much more likely to be infected. And, of course, condoms worked.

CHAMBERLAND: You were doing some statistical analyses.

BYERS: Yes.

CHAMBERLAND: Did you do any sort of projections in terms of just how big a problem this might be, to the estimated number of sex workers that there might be in Zaire?

BYERS: No.

CHAMBERLAND: No. I'm getting the sense that you were a field utility player. You 00:44:00started off with San Francisco City Cohort, and then as projects came along--

BYERS: Correct.

CHAMBERLAND: Did you like that?

44:14

BYERS: I did. I hate to do the same things over and over again. I like learning new stuff. I liked the variety of projects that I got to work on.

CHAMBERLAND: I want to talk about a couple of those other projects, but before I do that, I want to talk about one more, or have you talk about one more thing that was a key factor in doing this modeling, this projection. We talked about estimating incidence being a really key component for these back calculations. If I understand correctly, you were working with data like the City Clinic 00:45:00Cohort, where there happened to be blood test results at two different points in time. You didn't exactly know when the seroconversion might have occurred, so that was another mini model, I guess, or mini estimate. There was a new testing strategy that came about in the late 1990s that proved, I think, very useful for estimating incidence, as well as for clinical and prevention purposes. I'm talking about this idea, about in a sense manipulating HIV antibody tests. I wonder if you could talk a little bit about what that was and how it helped estimate more precisely when people might have become infected.

BYERS: You're talking about the less sensitive assay?

CHAMBERLAND: I am, I am. Sorry to be so obtuse.

BYERS: There were several blood tests that were "de-tuned" so that they would return a number that was supposed to be proportional to how much HIV is in your blood. The amount of HIV after infection goes up and up and up. What [Dr. Ronald S.] Ron Brookmeyer thought is that, if we would have an arbitrary cut-off and define a window period from infection until the cut-off, maybe we could use that to estimate infection with a cross-sectional study. In other words, one test per person. Then if people were in their window period, you'd count them. If they weren't in the window period, you wouldn't, and then you'd put that into a formula and get an incidence estimate. That was really interesting. The key 00:47:00piece of information was the mean window period, and in order to do that you had to do serial testing on a sample from your population that you're going to work on. That kind of data was hard to get, because people don't get tested at regular intervals and they don't all have the same number of tests and like that. We figured out some fairly sneaky models to estimate the mean window period, and then once you knew that, you would put that into the estimate. You'd do another study where you would do one test per person, count the number of people who were in their window period, put that in the formula and come up with 00:48:00an estimate of incidence.

CHAMBERLAND: If I understand it correctly, there's an HIV antibody test, these enzyme immunoassay antibody tests. They're getting better and better over time so they're becoming more sensitive, meaning you can detect the presence of antibodies earlier in the course of infection. The window period, the so-called period between when someone actually gets infected and when you can detect it, starts to become smaller and smaller, so that's good. Then the bit about what you call the less sensitive or the de-tuned assay, if I understand it correctly in terms of how it relates to incidence, is that the closer you test a person in 00:49:00time to when they became infected, they're only going to test positive, if you will, on the most sensitive assay. As time moves on and you make the assay less sensitive, then they become positive on the less-sensitive assay?

BYERS: No. The less-sensitive assay gives you a number that is supposed to be proportional to the antibodies in the serum.

CHAMBERLAND: Exactly. OK.

BYERS: It gets bigger and bigger as time goes on, or at least it should. We did find some people where it didn't change at all.

CHAMBERLAND: In a population, say you're testing a cohort of gay men or IV drug users and you've got an HIV antibody test that's been drawn on them, using this sensitive and less-sensitive testing methodology, you can figure out at the time 00:50:00of that test who got infected fairly recently and who's probably been infected for a longer period of time?

50:10

BYERS: Yes. I don't really like the emphasis on recent HIV seroconversion. Just because they haven't reached a certain level doesn't really mean that they're recently infected. The variability was huge, and that was one of the problems. It might be three years, it might be six months, and you had people both ways. The distribution of the outcomes was quite wide.

CHAMBERLAND: Oh, I had not appreciated that. Still and all, was it useful for modeling, in terms of trying to get a better handle on incidence?

BYERS: Everybody thought so, but so far, I'm not aware that it's ever given an answer that people agreed with.

CHAMBERLAND: Oh, really?

BYERS: To me, there's only one place it can go wrong, and that's in the mean window period. There's so much variability, and like I said, some of the people, their test result didn't increase at all over the three-year period. Those people were just thrown out because we didn't know what to do with them. How do you put zero back into a mean? I think the problem was that it's difficult to come up with a mean window period because of the variability.

CHAMBERLAND: I see. Is there anything else that made your life easier as a modeler in terms of--- Can you remember any key pieces of data or advances?

BYERS: Software. Software, man. Statistical Analysis System (SAS), SPSS 00:52:00[Statistical Package for the Social Sciences], then S-PLUS, three statistical packages that really, really helped get things done in a hurry. Also, they would help you learn new techniques, because the programmers had provided packages that could do things that you might not ever have heard of.

CHAMBERLAND: So it was a continual learning curve in terms of technology and software and scanning the literature, I guess, to see what other people are doing.

BYERS: Yes. Yes.

CHAMBERLAND: I guess as an epidemiologist, I just thought you guys thought deep thoughts in your offices and came up with these equations. But it was a lot more complicated than that.

BYERS: There are a lot of people out there a lot smarter than I am, and they were publishing in Biometrics and several statistical journals. I tried to keep 00:53:00up with those. In fact, a lot of the stuff they published had to do with analyzing AIDS data, so I was able to ride on their coattails to get things done here at CDC.

CHAMBERLAND: Do you think the HIV-AIDS epidemic was a motivator for progressing the whole field of infectious disease modeling? Had there been a lot of infectious disease modeling before AIDS? What impact do you think it had?

BYERS: I'm sure it did. I'm sure it did, because there were lots and lots and lots of people working on stuff. People here at the Rollins School of Public Health. Just about every school of public health had an AIDS project they were working on.

CHAMBERLAND: Because certainly now we hear a lot more about modeling -- 00:54:00influenza pandemics, there's a big Ebola outbreak. It's the first question, one of the first questions people want to know is "How big is it going to get; how many more cases are we going to get?" I wondered if in a sense the AIDS epidemic helped to move that whole discipline along.

BYERS: I totally believe it did.

CHAMBERLAND: Interesting. I want to talk about another area that you did a lot of work on, and this was modeling in the pediatric population. You were involved in a number of studies that looked at the prevalence and incidence of mother-to-child transmission of HIV and following these trends over time. Can you tell us a little bit about this work and some of the creative modeling that 00:55:00was required?

BYERS: The first paper that I was involved in was with Dr. Susan [F.] Davis. Basically, we took all the information that was available on vertical transmission (from mother to child), and just analyzed it and published the paper in 1995. Then in 1997, I published a paper in the Statistics in Medicine, which is a much more technical journal, and it was a much more technical paper. What we discovered is the number of pregnant women with HIV was falling, already falling, and the number of infants born with HIV was also falling. At that time 00:56:00we found that 25% of mothers with HIV would transmit HIV to their newborn. By '97, [Dr. M.] Blake Caldwell and [Dr.] Mary Lou Lindegren and I published a paper [showing] that we found that vertical transmission dropped 64% between 1992 and 1997. That's my favorite project of all the stuff I did, because it was the only good news we ever got. I just totally loved that.

I had some interesting models to do. It wasn't clear what kind of distribution to use to model the infection. Of course, there was no incubation time to speak 00:57:00of, so that simplified things some. I fit some different models and learned how to compare different models on the same data, which was an interesting new question for me. Most of the time people would just say, "Okay, I'm going to use this distribution," and they use it and that's it. But I wanted to know, is that the best one? I didn't have any reason to pick that one, so I picked four that seemed reasonable and found that one was a lot better. That was fun, too. Basically, the plots, the projections were showing that the number of babies being born with HIV was just falling off a cliff.

57:48

CHAMBERLAND: There was a big factor that played into that.

BYERS: Oh, AZT [azidothymidine].

CHAMBERLAND: AZT.

BYERS: Yes. Once they started giving the mothers AZT, the probability the baby 00:58:00would be born [infected] dropped from 25% to 8%. That was good. I think they have other drugs now that maybe even work better.

CHAMBERLAND: Knowing that percentage drop, due to the mothers taking AZT, were you then able to model how many -- I guess you could model pretty straightforwardly how many infections you could prevent annually or in a given period of time?

BYERS: I don't remember that we did that specific thing, but, yes, you could do that.

CHAMBERLAND: You're certainly right, that was one of the really good news stories that came out of the AIDS epidemic. First, the clinical trial results, which I think completely shocked people, that AZT was so effective when taken by 00:59:00pregnant HIV-infected women in reducing transmission to their infant, and then your documenting and expanding through your models just how big an impact that had. What were some of the sources of data that were critical that you needed to build those models? Obviously, you had national pediatric AIDS case surveillance, but you probably had to have some other bits of information -- women were being tested as part of a serosurvey of childbearing women. BYERS: Oh, the survey with childbearing women. Yes, that was a key source of data, the survey of childbearing women. I don't know much about it.

CHAMBERLAND: Right. CDC had implemented a whole family of seroprevalence surveys. And pregnant women, sampling pregnant women nationally to determine-- 01:00:00and that was easy to do, because actually instead of sampling the women, you sampled the cord blood.

BYERS: Mm-hmm.

CHAMBERLAND: [The cord blood] perfectly reflected the mother's serostatus because her antibodies were passively transfused to the infant. Infants are getting tested for various metabolic diseases routinely, and so [the seroprevalence survey involved] just siphoning off a bit of blood from the cord blood and adding on an HIV serologic test. That was a huge chunk of data that you got in terms of understanding prevalence in women nationally. Then there must have been various epi studies, because you knew what the rate of transmission was pre-AZT, 25%.

BYERS: Yes.

CHAMBERLAND: That really involved putting together a lot of, I guess as in any model, lots of bits of information.

BYERS: The paper in Statistics in Medicine went on and on and on. I was looking 01:01:00at it just the other day and saying, "Did I really do all this?" Yes, it was complicated.

CHAMBERLAND: And on and on and on, having looked at some of your papers, means a lot of equations.

BYERS: Lots of equations, lots of graphs. I love graphs.

CHAMBERLAND: That's a good illustration of the power of collaboration with epidemiologists.

BYERS: Oh, you bet. Blake and Mary Lou and Susan were wonderful to work with and very smart. [They were always coming up with good questions, trying to stump me.

CHAMBERLAND: By this time the program, AIDS, the statistical modeling, I'm assuming it's grown a bit. It was you and Meade when you first arrived. How did the program expand over time?

BYERS: Okay, the statistics section --you wouldn't really call us a section when 01:02:00there are only two of us, but when I left, the statistics section had, I don't know, maybe half a dozen Ph.D. statisticians, and they were staying busy.

CHAMBERLAND: Was it unusual at the time for modelers to actually be embedded in the program? Did you have colleagues within CDC who were statistical modelers that were embedded on a day-to-day basis in other programs?

BYERS: Yes, that's pretty much the way it worked, I think, although there was always somebody who said, "I want my own statistician. I want one in my section." We always resisted that, because it was nice to be able to walk next door and say, "Hey, I've got a problem."

CHAMBERLAND: Right, within the AIDS Program you were your own unit. How did you 01:03:00divide the projects among you? You were obviously working on pediatrics. You had done the City Clinic early on. How, as topics and questions came along, how did you guys figure out --guys and gals, I don't know -- how did you figure out who was going to do what?

BYERS: I'm afraid it was kind of random.

CHAMBERLAND: That sounds statistical: random.

BYERS: Yes, yes. I knew people. I mean, you get to know people. Sometimes the section chief would assign projects, and then you would get to know those people. Then they would come back with the next project, and the section chief wouldn't hear about it for a year or so. It was kind of loose, but it worked.

CHAMBERLAND: Who were some of your companions in crime there in the statistical 01:04:00section as it grew larger? Who were some of the other folks that were there?

BYERS: Of course, there was Meade and then there was -- I can see them in front of me, but I don't do names very well anymore.

CHAMBERLAND: [Dr.] John [M.] Karon.

BYERS: John Karon, right, right. He's out in Albuquerque now, retired. [Dr.] Lillian [S.] Lin. Sorry.

CHAMBERLAND: There was a number of you.

BYERS: There was a number, yes. I can picture three others right now, but I can't remember their names.

1:04:45

CHAMBERLAND: That's fine. What I was trying to get at a little bit earlier was the idea that there was a group of actual modelers working embedded in a 01:05:00program, and I didn't know if that was unusual within CDC. Certainly, other programs had statisticians that helped them, but I wasn't sure if I could recall instances in which bona fide modelers, if you will, were embedded in programs.

BYERS: I wasn't very in tune to what else was going on at CDC outside AIDS.

CHAMBERLAND: It strikes me as a good idea. You grew with that program. You came in '84. That's still very early days, and information about the epidemic is growing by leaps and bounds, as are statistical software packages and techniques and the like. To have someone intermittently helicoptering in and out to help 01:06:00with projects doesn't strike me as an efficient way to do business, especially when you're on such a fast curve of progression.

BYERS: I agree.

CHAMBERLAND: I guess I was struck by that. Do you think you and your compatriots, the statisticians and modelers, were under a lot of pressure in those early days to produce?

BYERS: There were always deadlines, mostly having to do with conferences. We wanted to present the work we had done at a conference. You had to have a presentation put together in time to send it to the conference organizers so that they would put you on their list. It wasn't like you were going to lose your job if you didn't get it done in time, but you'd be disappointed.

CHAMBERLAND: I was thinking more from the aspect of this intense interest that grew up on projections, national case projections, which continues today. The epidemic is continuing to be projected in terms of incidence and prevalence. I wondered if there was a lot of pressure being brought to bear to your group collectively about this, whether you ever just felt, wow, they're really interested in this stuff.

BYERS: That was more exciting than being pressured. You know, we're where it's at. We liked learning. Everybody really took it all very seriously.

CHAMBERLAND: Did you continue to work on national case projections or HIV 01:08:00prevalence as time marched on? I know you were doing some of that early on.

BYERS: Yes. I think John Karon took over most of that. The last time I was involved was shortly after I retired. There was a conference, and I sat in on it. People had made different estimates. I checked the MMWR -- no, the CDC website -- and the number of -- the annual incidence estimates were still pretty much the same as they were then. Actually that is totally reasonable, because you've got a huge population at risk. Once they all get infected, you expect just a steady state, and that seems to be what has happened now.

CHAMBERLAND: Which sadly says something about the efforts to prevent infection 01:09:00in the first place.

BYERS: I wasn't going to say that, but that's what I was thinking, yes.

CHAMBERLAND: Yes, yes, yes. I've touched on some of the aspects of your work. Am I missing -- for example, I had no idea about your international work.Any other topics, memories that you have that we should touch on, in terms of areas of work that you were involved in or particularly enjoyed?

BYERS: I think that pretty well covers it.

CHAMBERLAND: Did you get to go on any more international trips?

BYERS: No. That was the only one.

CHAMBERLAND: When you were in Zaire, did they have any statistical support at all? Were you able to--

BYERS: (shakes his head "no")

CHAMBERLAND: None at all?

BYERS: I was it. They had a little IBM computer. I had to install the software on that. That took a day. It was software that I had never used before, so it 01:10:00took another day to figure out the software. The resources weren't great in Kinshasa at all.

CHAMBERLAND: Again, since it was very early days, ultimately the effort was made to train people, local people in these countries to be able to handle the epidemiologic studies and statistical computations. This was even before that was even possible, from what you described.

1:10:40

BYERS: Yes.

CHAMBERLAND: Gosh. The AIDS epidemic, you worked on it for 20-plus years at CDC. Do you have any sense of what the personal and professional impact of working on the AIDS epidemic had for you?

BYERS: I feel truly fortunate because of the atmosphere at CDC. Everybody was so 01:11:00willing to work together, willing to share the glory. All the papers had like a dozen people on them. Everybody who worked on it got credited. It was not like that everywhere. It sure makes your life a lot more pleasant, not to be fighting over things like who did this and who did that. Basically, if you wrote the paper, you were the first author, and [Dr. James W.] Jim Curran was the last author, and then the rest of it was up for grabs. As I mentioned, I was always learning new stuff. I would have gotten really bored in something else, I think.

CHAMBERLAND: Yes, that's interesting, because to work 20 years on one topic, AIDS, but--

BYERS: It just kept expanding. There was always something new.

CHAMBERLAND: It's interesting, because in the interview we talked about the San Francisco City Cohort and the shock, the devastation of learning such a high proportion of people were infected before really the epidemic itself was even recognized. At the other end of that spectrum was the pediatric incidence study, which documented and modeled through your work just what an impact AZT was having-- close to stopping the pediatric mother-to-child transmission route. It was two ends of the spectrum of despair and hope, I guess.

BYERS: Yes. One story about AZT: one of our number was convinced that AZT was killing people. We had a statistical model which was new, called the Cox Proportional Hazards Model, and it was very clever. You could put in a covariate that said, at this time something changed, and what is the relative hazard due to that change. We analyzed people and said, "Okay, we're going to put in a covariate for when they started AZT." Sure enough, every time somebody was put on AZT, their risk of dying went up six times.

1:13:58

Okay, let's publish a paper. I said, "Wait, I've got to think about this. Maybe 01:14:00it's the other way around. Maybe the people are getting put on AZT because they're very, very sick and about to die anyway." "Oh, no, no, no, it's AZT." "No, no, no, maybe it isn't." This person, who I will not name, would walk past me in the hall and not speak to me because I wouldn't sign off on this paper. Eventually, I thought of a way to do a graph where there was one horizontal line for each person in the study, with colors for when AZT started and when they died and when they were infected, or when we thought they were infected. Sure enough, the people who were dying after AZT were the people who were infected at least two years longer than everybody else. They were being put on AZT because 01:15:00they were very, very sick, not the other way around. Then this guy published a nice little paper going through everything. He just did a 180 and wrote it up differently, and it was good. Watch out for those people. Selection bias. Selection bias. Watch out for that.

CHAMBERLAND: Gosh, that's a really interesting story. I think it illustrates a powerful example of the importance of good collaboration between--

BYERS: Yes.

CHAMBERLAND: --your epidemiologists and your statistical modeling experts. How interesting. Had this also been an idea external to CDC? Were people also putting forward--

BYERS: Not that I know of.

CHAMBERLAND: This was just a home-grown--

BYERS: We knew that AZT was toxic to a degree, and that's why they didn't give it to people until they got sick. But nobody was saying it was killing people.

CHAMBERLAND: Good story, Bob. Got any more?

BYERS: I think that's about it.

CHAMBERLAND: It's been just a delight talking with you and getting a conversation included in our oral history project with a statistician-modeler. As you've illustrated so nicely for us, it really was a team effort. The example you just provided and others were absolutely essential to helping make sure that CDC got it right. Thank you. I enjoyed talking with you.

BYERS: Thank you for inviting me, because it's been great. I've really enjoyed it.