I knew there was a reason that I don’t often blog about politics, and yesterday reminded me of it. Maybe I should have just launched another enthusiastic debunking of the distortions and outright false information put out by antivaccination advocates like Dawn Winkler. Instead, I thought it might be educational to return to a topic that I haven’t revisited in a while, so-called HIV/AIDS “dissidents.” These cranks resemble antivaxers in their fast-and-loose approach to and cherry-picking of the data, along with some outright misrepresentations of studies.
They’re at it again.
This time around, they’re misrepresenting a paper:
Context Plasma human immunodeficiency virus (HIV) RNA level predicts HIV disease progression, but the extent to which it explains the variability in rate of CD4 cell depletion is poorly characterized.
Main Outcome Measures The extent to which presenting plasma HIV RNA level could explain the rate of model-derived yearly CD4 cell loss, as estimated by the coefficient of determination (R2).
Results In both cohorts, higher presenting HIV RNA levels were associated with greater subsequent CD4 cell decline. In the study cohort, median model-estimated CD4 cell decrease among participants with HIV RNA levels of 500 or less, 501 to 2000, 2001 to 10 000, 10 001 to 40 000, and more than 40 000 copies/mL were 20, 39, 48, 56, and 78 cells/µL, respectively. Despite this trend across broad categories of HIV RNA levels, only a small proportion of CD4 cell loss variability (4%-6%) could be explained by presenting plasma HIV RNA level. Analyses using multiple HIV RNA measurements or restricting to participants with high HIV RNA levels improved this correlation minimally (R2, 0.09), and measurement error was estimated to attenuate these associations only marginally (deattenuated R2 in the 2 cohorts, 0.05 and 0.08, respectively).
Conclusions Presenting HIV RNA level predicts the rate of CD4 cell decline only minimally in untreated persons. Other factors, as yet undefined, likely drive CD4 cell losses in HIV infection. These findings have implications for treatment decisions in HIV infection and for understanding the pathogenesis of progressive immune deficiency.
The HIV/AIDS “dissidents are, of course, leaping at the last sentences there in the conclusion that HIV RNA levels are poor predictors of the rate of CD4 cell decline in individuals, as if this observation somehow invalidated the hypothesis that HIV causes AIDS or negated the findings that initial viral load is a predictor of the rate of progression to AIDS.
Nick Bennett and Tara Smith are already all over this; so I’m not going to rehash everything they said. The easiest way to explain it is that the HIV “dissidents” crowing about this study seem to demonstrate a profound confusion over the concept that measurements that correlate with a certainclinical outcome in a population may not necessarily be so good at predicting outcomes in individual patients. As Nick Bennett put it:
It’s well known (at least among those who bother to read and understand the literature) that those people with higher viral loads tend to progress faster, as was shown by John Mellors back in the mid 1990s using the large Multicenter AIDS cohort study (MACS).
This study took things one step further. They replicated the original findings of Mellors by showing again that viral load roughly predicted how fast AIDS occurred in another large cohort composed of people from 3 seperate study sites. For example, in this new paper people with viral loads less than 500 had an average loss of CD4 cells of 20 per year whereas those with viral loads over 40,000 had an average loss of 78 a year (with a smooth change for values inbetween). Basically this data proved that viral load was a reasonable predictor of rate of progression! They compared this analysis with the original MACS cohort and it looks practically identical!
But then they tried to look at the individual rate of progression of each member of the cohort. Unsurprisingly they found that the rough-and-ready estimates of progression rate within a subgroup varied from one individual to another. When they ran complex statistical analysis on the effects of viral load on THIS data they found that only about 5-6% of the inter-individual variation can be explained by the initial viral load. In another words, while viral load predicts that you WILL lose CD4 count, and you can give an AVERAGE loss of CD4 cells per year based on that count, you can’t say for sure what the ACTUAL loss will be for any one person very accurately.
This sort of phenomenon is not at all uncommon in medicine, particularly when dealing with prognostic indicators. (It’s even not that uncommon in predicting response to treatments, as I will discuss below.) In this case, the authors actually confirmed that increased HIV loads do correlate with how rapidly a patient’s CD4 count falls every year. But we’re talking about aggregates here–populations. In individuals, the study found that CD4 counts are not nearly as predictive. As Tara points out:
What they apparently don’t realize is that again, this is common in most studies looking at predictive criteria such as these. For example, on average, elevated blood cholesterol levels correlate with an increased risk of cardiovascular disease (CVD). But, one person may have very high blood cholesterol level and show no signs of disease, while another may have fairly low cholesterol and still have CVD. It’s always tougher to apply these population-based measurements at the individual level, since population-based data by their nature average out these individual variations.
Tara’s is a good example. Now here’s where I put my two cents in that isn’t just rehashing other people’s explanations and argument. Does anyone in this day and age still believe that smoking doesn’t cause lung cancer? The epidemiological evidence of the association is bulletproof. However, the majority of smokers don’t get lung cancer. In fact, there are complex statistical models that allow a pretty accurate calculation of risk in populations based on how long and how much a smoker has been smoking. For example, heavy smokers have an approximately 20% to 25% chance of developing lung cancer in the course of their lifetimes. That’s actually a considerably lower chance of developing a lung cancer from smoking over 50 years or so than an HIV-positive patient has of progressing to full-blown AIDS in 10 years (approximately 50%); yet no one seriously disputes that smoking is a very strong risk factor for lung cancer. At the population level, the association is very strong. However, if I see a 65-year-old patient who has been smoking since age 18, I can’t tell him whether he will definitely get lung cancer. I can only quote probabilities. Between two heavy smokers, one may live to 80 and never get lung cacner while another may develop stage IV lung cancer at age 45. We have no good way of predicting which individual patient will be which.
Let’s look at another example: Breast cancer. Let’s look at stage IV disease, which is, in essence, 100% fatal eventually. We actually have pretty good estimates of median survival and what a patient’s chance of living 6, 12, 24, 36, and 60 months after diagnosis are. However, when faced with a single breast cancer patient with stage IV disease, we are pretty poor at predicting how long that particular patient will survive. We just can’t give a very accurate answer to this poor hypothetical patient’s question, “How long have I got left?” We can only quote probabilities, which is not what the patient really wants to hear. To echo the example of smokers above, one woman might deteriorate and die in 6 months after being diagnosed, while all of us involved in the care of breast cancer patients have seen the occasional patient who has lived with metastatic disease for several years and done mostly well. Again, we have no good way of predicting which patient will be which, although that may change in the future.
As physicians, we are used to thinking like this about biological markers and prognostic indicators. We are quite comfortable distinguishing between population-based studies and are aware of the perils of trying to extrapolate those studies to individual patients. Mathemeticians like Darin C. Brown, apparently, are not:
Now, after 25 years of very expensive research, ‘they’ have arrived at the conclusion that 90% of what we always thought was the most important factor (loss of CD4 cells) can’t be accounted for by the amount of HIV (assuming the viral load test is even accurate in the first place).
Never mind, the fundamental assumption cannot be wrong. We just need billions and billions more dollars to figure out the ever more enigmatic mechanisms by which HIV is responsible for CD4 loss while only 10% is accountable for by viral load. Just give us more money, and we’ll produce as many “mysteries” as we please, and you pay for.
Nor, unsurprisingly, are HIV “skeptics” like Hank Barnes.
Funny how HIV/AIDS “dissidents” like Dr. Brown and Hank claim that the “establishment” never questions the “HIV dogma,” when this paper questions the previously accepted utility of using viral load as a predictor of the rate of progression to AIDS for individual patients, isn’t it? Of course, the paper doesn’t question that HIV causes AIDS because, quite simply, the HIV/AIDS hypothesis does not rely on HIV RNA levels being predictors of disease progression. Indeed, the authors even say as much:
These findings represent a major departure from the notion that plasma HIV RNA level is a reliable predictor of rate of CD4 cell loss in HIV infection and challenge the concept that the magnitude of viral replication (at least as reflected by plasma levels) is the main determinant of the speed of CD4 cell loss at the individual level. The clinical implications are that in the majority of cases, an individual patient’s plasma HIV RNA level at the time of presentation for clinical care cannot predict, to a significant extent, the rate of CD4 cell decline that he or she will experience over the subsequent years and is therefore of limited clinical value in shaping the decision to initiate antiretroviral therapy. This is despite the fact that a group of individuals with an approximately similar level of plasma viremia will, on average, tend to lose CD4 cells at a faster rate than another group with a lower level of viremia, a previously reported finding5 that stands uncontested by our results.
So much for the dreaded conformity of scientists and their refusal to challenge accepted findings. Although this paper does question the utility of viral load as a predictor for progression to HIV in individual patients, it does not in any way question the well-supported findings that HIV causes AIDS in a high proportion of patients, as Nick puts it:
This result is very important in that it highlights the need to investigate other factors important in triggering or controlling rate of progression to AIDS, but it won’t really change the current paradigm in terms of understanding AIDS pathogenesis, nor will it change current treatment guidelines, because neither depends on the idea that HIV viral load is the be-all and end-all of AIDS.
In reality, what this result really suggests is that our understanding of the mechanism by which HIV causes the decline of CD4 cells is incomplete, and, like all good research papers, suggests areas for further research. As the authors state:
In humans, the predictive value of immune activation level on HIV disease course, independent of plasma HIV RNA levels, can be demonstrated even when measured during early infection or before actual seroconversion. Thus, immune activation may be a major determinant of T-cell turnover and CD4 cell depletion in chronic HIV infection both in human and animal hosts. Our results provide further support for additional studies exploring the relative contribution of immune activation to the pathogenesis of immune deterioration in treatment-naive, HIV-infected persons.
The results of our study challenge the concept that CD4 cell depletion in chronic HIV infection is mostly attributable to the direct effects of HIV replication. Future efforts to delineate the relative contribution of other mechanisms will be crucial to the understanding of HIV immunopathogenesis and to the ability to attenuate it.
Repeat after me: This study does not challenge the concept that CD4 depletion is caused by HIV. It simply suggests that the concept that this depletion is mostly due to direct effects of HIV replication is probably too simplistic a description of how HIV depletes CD4 cells.
In case Darin Brown still thinks that an only 10% predictive value in individual patients is not common clinically or are not clinically relevant, let me provide one last example from the world of oncology. In stage I breast cancer we often recommend chemotherapy after surgery. Here’s a question: If one group of 100 women with stage I breast cancer are given chemotherapy after surgery and another group of 100 are not, do you know how many more women in the chemotherapy group will survive compared to the group that didn’t get chemotherapy? (For reference, I’ll tell you that five year survival in stage I cancer is upwards of 90%.)
The answer? Three. That’s right. Three. In stage I disease, chemotherapy only confers approximately a 3% better chance of survival. And we have no real way of telling which three women will benefit. We end up giving 97 women chemotherapy that they either wouldn’t have needed (they would have been cured by surgery alone anyway) or that wouldn’t help them (they would have recurred regardless) in order to save three. Granted, looked at another way, this does represent an approximately 30% relative better chance of survival (3/10), but as an absolute percentage it is small. Yet we still do it, and some studies have suggested that many women with breast cancer would take chemotherapy for even a 1% better chance at survival. In any case, this modest benefit from adjuvant chemotherapy is why we are working on developing other methods of detecting which women will and will not benefit from chemotherapy. Right now there is a test known as the Oncotype DX assay that looks at he expression of a set of genes that distinguish stage I cancer at high risk of recurrence from that at low risk. Women at low risk would likely not benefit from chemotherapy and could safely forgo it.
In medicine, when faced with weak correlations between a marker and a clinical outcome on an individual level, physicians usually conclude quite reasonably that something else is complicating the link apparent at the population level. Given the complexity of human biology, such problems with predictive markers are very common. Indeed, a clean association between a marker and a clinical outcome in individuals is rare. In the case of HIV, other evidence that HIV causes AIDS is overwhelming, and all the results of this study show is that HIV viral load is not a reliable predictor of CD4 cell decrease and progression to AIDS in individual patients. Other factors must be important in terms of the mechanism by which HIV infection leads to CD4 depletion. That’s it. The attempt by HIV “skeptics” to twist this study to “show” that HIV doesn’t cause AIDS reveals them to be the equivalent of an antivaccination loon like Dawn Winkler. They’re not even all that much better at cherry picking and misrepresenting studies to imply that they show something they do not.