CROI: Rate of CD4 decline is a poor guide to the risk of AIDS, say investigators

This article is more than 17 years old. Click here for more recent articles on this topic

The rate at which an untreated HIV-positive person's CD4 cell count is declining is a poor predictor of the risk of AIDS or death in individual patients, the Fourteenth Conference on Retroviruses and Opportunistic Infections was told this week.

A study published last September (Rodriguez) caused some consternation in the scientific community by questioning the link between HIV viral load and CD4 decline.

Rodriguez found that although in broad terms high, medium and low baseline viral loads predicted high, medium and low rates of CD4 decline over the succeeding six months, when it came to individual patients their baseline viral load only predicted between 3-9% of the variability of CD4 decline.


inter-quartile range

The spread of values, from the smallest to the largest. The inter-quartile range (IQR) only includes the middle 50% of values and measures the degree of spread of the most common values.


The process of viral multiplication or reproduction. Viruses cannot replicate without the machinery and metabolism of cells (human cells, in the case of HIV), which is why viruses infect cells.

surrogate marker

An indirect indicator of something, such as measuring viral load to assess the treatment effect of a drug.



The origin and step-by-step development of disease.

He explained this by suggesting that there were many other factors which led to CD4 decline other than direct viral attack by HIV and commented that his findings “challenge the concept that the magnitude of viral replication is the main determinant of CD4 cell loss.”

The challenge was largely to an influential study of the MACS Cohort (Mellors 1997), a group of over 1,600 HIV-positive gay men without AIDS symptoms who were initially recruited in 1984 and 1985. John Mellors, the chief investigator on the MACS Cohort study, had analysed their earliest blood samples from the 1980s after viral load testing became available in 1995, and found a strong correlation between their original, baseline viral load and the speed of progression to AIDS.

Rodriguez’ study therefore challenged Mellors’ original findings and he announced last September that a re-analysis he was conducting of the MACS Cohort would refute Rodriguez’ paper.

He duly presented his findings at CROI (Mellors 2007). A second study (Lau) was also presented that broadly agreed with his conclusions.

Mellors first announced that he had in fact confirmed Rodriguez’ conclusions. His analysis found that only 3% of the variability in CD4 decline was accounted for by baseline viral load. He went further and took two other measurements: baseline CD4 count and a count of the cells with the CD38 marker, a measure of immune activation.

He found that baseline CD4 count only predicted 7.1% of subsequent CD4 decline and CD38 count only 1.4%. He also included age at baseline and found this had virtually no effect on CD4 decline.

He said he was intrigued by these findings and extended his analysis to ask the question: which baseline marker is most strongly predictive of the time taken to develop AIDS symptoms? (Mellors also analysed time to death but did not present these findings, though he said they supported the ones he did announce.)

He measured baseline viral load, CD4 and CD38 counts and also baseline CD4 decline, as defined by the decline in cells per year from the time patients joined the study to mid-1988.

He found that baseline viral load predicted 46.1% of the variability in time to AIDS, CD38 count 40.1% and CD4 count 29.4%.

When the baseline measurements were combined, their predictive power became stronger. Viral load and CD38 count together predicted 58% of the variability in time to AIDS, viral load and CD4 count 54% and CD4 and CD38 count 43%.

But the medium-term baseline CD4 decline was a very poor predictor: it only predicted 2.9% of the time to AIDS.

Why was this? Mellors found that although the medium-term CD4 decline averaged 65 cells a year, it had a huge range. The inter-quartile range was nine to 130 cells, in other words, about half of the CD4 declines measured were either greater than 130 cells or less than nine cells (or could even be increased in some cases).

This means that medium term CD4 decline – the quantity Rodriguez said was poorly predicted by viral load – is itself a useless predictor of progression to AIDS, because it is so immensely variable.

In fact it is so difficult to calculate the gradient of a medium-term CD4 decline which may in fact feature short-term, abrupt increases and decreases such that the average statistical error in the calculation is 55 cells – almost equal to the average annual decline.

“In a quarter of the observations, the error of the slope is greater than the measurement of the slope itself,” Mellors said.

“These results reaffirm the key role of viral replication as reflected by viral load in the pathogenesis of AIDS,” he added.

In other words, baseline viral load is a strong predictor of the long term CD4 decline which leads to AIDS.

A questioner commented that CD4 decline might accelerate as AIDS neared and that this might invalidate overall decline as a measurement. And UK statistician Professor Andrew Phillips commented: “If you wish to predict time to AIDS over a long period and can only measure one thing, you might want to measure viral load. But if you want to know the risk of AIDS within a year, you’d need to know the CD4 count.”

Benigno Rodriguez, who was in the audience, commented that his paper and Mellors’ broadly agreed on the importance of baseline viral load but said that Mellors' findings “provide a much needed piece of additional information, which is how well the CD4 slope correlates with progression.”

Mellors’ findings were broadly supported by a second paper (Lau). This found that a single measurement of viral load had virtually no predictive value for CD4 decline over the short term (less than nine months). This was because both viral load and CD4 slope were inherently unstable variables. However a single CD4 count measurement had a much greater predictive value over short term periods: its predictive value was 76% over a time period of nine months, supporting Andrew Phillips’ comment.

A single viral load measurement was only moderately predictive of the time to AIDS or death (23%) but a series of viral load tests was strongly predictive (61%).

What these two papers taken together demonstrate, then, is that certain surrogate markers can indeed predict the risk of AIDS or death with reasonable accuracy but that one has to select different markers according to the time period over which one is measuring the risk.


Rodriguez B et al. Predictive value of plasma HIV RNA level on rate of CD4 T-cell decline in untreated HIV infection. JAMA 296 (12): 1498-1506, 2006.

Mellors JW et al. Plasma viral load and CD4 lymphocytes as prognostic markers of HIV-1 infection. Annals of Internal Medicine 126: 946-954, 1997.

Mellors JW et al. Comparison of plasma HIV-1 RNA, CD4 cell count, and CD38 expression on CD8 T cells as predictors of progression to AIDS and CD4 cell decline among untreated participants in the Multicenter AIDS Cohort study. Fourteenth Conference on Retroviruses and Opportunistic Infections, Los Angeles, abstract 139, 2007.

Lau B et al. Predictive value of plasma HIV RNA levels for rate of CD4 decline and clinical disease progression. Fourteenth Conference on Retroviruses and Opportunistic Infections, Los Angeles, abstract 140, 2007.