HAART, heart disease, & lying with statistics
Posted by Henry Bauer on 2010/01/19
Brian Carter had asked: “More proof of drugs causing disease?”
in connection with the piece,
“Increased presence, severity of coronary artery plaques in HIV-infected men”
(ScienceDaily, 8 January 2010):
“The HIV-positive participants had longstanding infection, were generally healthy, and the great majority were receiving antiretroviral therapy” [emphasis added].
Seemed likely that the answer would be “Yes” to Brian’s question, so I got the original article, “Increased prevalence of subclinical coronary atherosclerosis detected by coronary computed tomography angiography in HIV-infected men” (Lo et al.). I was surprised to read that duration of HIV infection was significantly associated with coronary plaque build-up “after adjustment for age, traditional risk factors, or duration of antiretroviral therapy” [emphasis added]. If HAART was causing the coronary disease, then longer HAART should correlate with more disease, surely!
Well, consider that statement carefully: It doesn’t deny that duration of antiretroviral therapy is associated with plaque build-up; it just says that HIV infection is independently associated with plaque buildup. But reading the whole article reveals that there’s actually no good evidence that plaque build-up is associated with “HIV” in absence of HAART.
Many studies have indicted HAART for increasing coronary risks, notably the protease inhibitors which disturb fat metabolism. That was reported already within a year or two of the first use of protease inhibitors, which had also been reported to produce hyperglycemia and hemolytic anemia (Massip et al.). “Use of HIV PIs is associated with atherogenic lipoprotein changes and endothelial dysfunction. Because these metabolic and vascular changes predispose to atherosclerosis, monitoring and treatment of dyslipidemia in patients taking these medications is warranted (Stein et al.).
HAART doubles the risk of coronary heart disease in young adults (Currier et al.) and “the proportion of deaths attributable to non-AIDS diseases increased and prominently included hepatic, cardiovascular, and pulmonary diseases, as well as non-AIDS malignancies. Longer time spent receiving HAART and higher CD4 cell counts at HAART initiation were associated with death from non-AIDS causes. CD4 cell count at time of death increased over time” (Palella et al.). HAART also is associated with an increased thickness of carotid-artery deposits (IMT): “Antiretroviral-treated patients had a higher median IMT than the untreated patients (0.94 vs. 0.85 mm, P = 0.006; Fig. 3). Furthermore, among all HIV-infected participants, increasing duration of HAART (rho 0.20, P < 0.001), protease inhibitor use (rho 0.19, P < 0.001), and nucleoside analogue use (rho 0.23, P < 0.001) were each associated with thicker IMT. These relationships remained significant after adjustment for traditional cardiac risk factors and the duration of HIV diagnosis (P < 0.024 for all)” (Hsue et al.).
((However, this unequivocal statement is followed by equivocations in the article’s Discussion section — “Compared to uninfected controls, carotid IMT was higher among all groups of HIV patients, irrespective of antiretroviral treatment or the level of viremia”; yet also “antiretroviral therapy and advanced immunodeficiency likely contribute independently to atherosclerosis in HIV patients”. The equivocation may be inevitable since the HIV-negative “controls” had lower coronary risk factors, so the study could not possibly identify “HIV” as the cause of more coronary disease:
“The HIV-seropositive patients were older than the controls (48 years compared to 43 years) and more likely to have smoked in the past. The HIV-seropositive patients were also more likely to have had a prior history of coronary artery disease and a prior history of hypertension, whereas the controls had a higher LDL cholesterol [112 vs 103] and HDL cholesterol [47 vs 42.7]”. My cardiologists asserted that the risk factor is actually HDL/LDL ratio, desirably >0.3 and ideally >0.4. Here we have controls with 47/112 = 0.420, in other words no risk factor, and “HIV-positives” with 42.7/103 = 0.414, indistinguishable from controls. In other words, “HIV-positives” in this study had unquestionably higher risk factors, making any implication of “HIV” highly questionable. You can’t claim significant results if the controls aren’t proper controls, no matter how you dice and slice the numbers.))
The Treatment Guidelines issued by the National Institutes of Health surely represent an official mainstream consensus, and they are clear that HAART is associated with increased organ failure including adverse coronary events:
“In the era of combination antiretroviral therapy, several large observational studies have indicated that the risk of several non-AIDS-defining conditions, including cardiovascular diseases, liver-related events, renal disease, and certain non-AIDS malignancies [14-19] is greater than the risk for AIDS in persons with CD4 T-cell counts >200 cells/mm3; the risk for these events increases progressively as the CD4 T-cell count decreases from 350 to 200 cells/mm3” (emphases added; p. 21, 3 November 2008).
How then could Lo et al. claim that HIV infection was significantly associated with coronary plaque build-up “after adjustment for . . . duration of antiretroviral therapy”?
In point of fact, some of the reported data do show an effect of HAART on plaque build-up — albeit there’s a certain amount of statistical legerdemain to wade through, and the need to THINK about whether the statistical calculations make sense.
The authors’ affiliations are with programs in Nutritional Metabolism, Cardiovascular Imaging, and Internal Medicine, entirely appropriate for examination of arteries and the like, but not necessarily for statistical analysis. It may be, of course, that one or more of the authors is expert in statistics, but it isn’t confidence-building to read that
“Two-tailed probability values are reported and statistical significance was assumed when P value was less than 0.05. All statistical analyses were performed using JMP (SAS Institute Inc., Cary, North Carolina, USA) and SPSS (SPSS Inc., Chicago, Illinois, USA).”
Expert statisticians tend to groan and roll their eyes heavenward when non-statisticians use such readymade statistics software packages and feed data into them without thinking knowledgeably about whether a particular mode of analysis makes sense.
Lo et al. used imaging to measure degree of atherosclerosis “in HIV-infected men”, with a sample of 78 and 32 controls of HIV-negative men matched for cardiovascular risk factors, and they found significantly more plaque in the “HIV-infected”. Then, among the HIV-infected only, they compared those with plaque (32) and those without (46) and found statistical significance at the p<0.05 level for several parameters (Age, Race, Framingham risk score, Duration since HIV diagnosis, CD4+/CD8+ T-lymphocyte ratio, Total cholesterol, Triglycerides, Cytomegalovirus titer), but not for antiretroviral treatment:
Note that ALL of the measures of HAART duration are higher in the plaque group. That these differences do not reach significance at the p<0.05 level is no evidence, however, against the previously well-established fact that antiretroviral treatment causes coronary disease. At a minimum, one would say that while the numbers do not add statistically significant evidence, they are consistent with the expectation that HAART causes coronary disease:
— For one thing, the sample sizes are not large enough to produce statistical significance in a multivariate analysis with so many variables.
— Second, comparing mean years of treatment is not the best way to look for possible causation of coronary disease. One would like to see a graph or data set showing the raw data for each individual, because one wants to test for the presence of a threshold effect or some other non-linear dose-effect or duration-effect relationship: it is perfectly feasible, perhaps even to be expected, that a certain period of years on antiretroviral treatment would produce no discernible plaque but that accumulation might progress relatively rapidly once plaque has begun to form.
— Third: One would like the data for individuals also because of certain apparent incongruities, for instance the much shorter periods on protease inhibitors than on antiretroviral treatment as a whole, since HAART is generally thought to have protease inhibitor as one of the typical components.
— Above all, though, the manner in which these numbers are presented makes it easy to overlook that the best indicator of whether HAART causes coronary disease is to compare “HIV-positive” people who have never been on HAART at all with those who have been. Here, there is no significant difference found, with p=0.16 too large. But hidden in plain view is the fact that the number of individuals NOT on HAART was only 3 in the no-plaque group and 1 in the plaque group!
NOT on HAART were 3/32 (9.4%) in the no-plaque group and 1/46 (2.2%) in the plaque group. That looks quite significant, and very different from 90.6 vs. 97.8.
“How to Lie with Statistics” by Darrell Huff (1954/1993) and “Damned lies and statistics: untangling numbers from the media, politicians, and activists” and “More damned lies and statistics: how numbers confuse public issues” by Joel Best (2001,2004) point to many other devices for presenting numbers and statistical data to produce a desired impression not necessarily warranted by the data themselves. In the present case, one can legitimately state a priori that it is unlikely to be a good test of the influence of HAART on coronary disease when the proper control group — “HIV-positive” individuals who have never experienced HAART — constitutes only 4 of the sample of 78. Further, any possible effect is hidden rather than exposed when the data are presented only as averages, which leave salient questions unanswerable. For example, the average time since HIV diagnosis is 13.5 years, yet the average duration of antiretroviral treatment is 6.2 and 7.9 years for the two groups but only 4 individuals out of 78 have never had antiretroviral treatment: How many had very short courses? Why? What were the differences in individual treatments? And so on.
Given the small sample size, one might reasonably conclude that statistical analysis could not command the power to deliver statistically significant results, and one would make do with the numbers themselves. And those numbers are perfectly consistent with all the previous reports that antiretroviral treatment increases the risk of coronary disease (and also kidney and liver disease as well as certain cancers). After all, among people never on HAART, 75% (3/4) had no sign of coronary disease and only 25% did, whereas of those who have experienced HAART, 61% showed plaque (45/74).
Above all, the data presented by Lo et al. shows a pronounced increase in cardiovascular disease with longer duration of “HIV infection” AND 74 OF THE 78 PEOPLE IN THE SAMPLE HAD BEEN ON ANTIRETROVIRAL TREATMENT, only 4 had not, and 3 of those 4 showed no sign of cardiovascular disease. Those data are at least equally compatible with the inference that duration of HAART correlates with cardiovascular disease, and comparisons of those with and without plaques confirms that: Those with plaque had experienced antiretroviral treatment 27% longer (7.9 vs 6.2 years), particularly protease inhibitors (52% longer, 4.4 vs 2.2).
Lo et al.: Janet Lo, Suhny Abbara, Leon Shturman, Anand Soni, Jeffrey Wei, Jose A. Rocha-Filho, Khurram Nasir, and Steven K. Grinspoon, “Increased prevalence of subclinical coronary atherosclerosis detected by coronary computed tomography angiography in HIV-infected men” AIDS 24  243-53
Massip et al.: P. Massip, B. Marchou, E. Bonnet, L. Cuzin, & J. L. Montastruc, “Lypodystrophia with protease inhibitors in HIV patients”, Thérapie 52  615
Stein et al.: James H. Stein, Melissa A. Klein, Jennifer L. Bellehumeur, Patrick E. McBride, Donald A. Wiebe, James D. Otvos, & James M. Sosman, “Use of Human Immunodeficiency Virus-1 Protease Inhibitors is associated with atherogenic lipoprotein changes and endothelial dysfunction”, Circulation 104  257-62.
Currier et al.: Judith S. Currier, Anne Taylor, Felicity Boyd, Christopher M. Dezii, Hugh Kawabata, Beth Burtcel, Jen-Fue Maa, & Sally Hodder, “Coronary heart disease in HIV-infected individuals”, JAIDS 33  506-12
Palella et al.: Frank J. Palella, Jr., Rose K. Baker, Anne C. Moorman, Joan S. Chmiel, Kathleen C. Wood, John T. Brooks, Scott D. Holmberg, & HIV Outpatient Study Investigators, “Mortality in the Highly Active Antiretroviral Therapy era — Changing causes of death and disease in the HIV Outpatient Study” JAIDS 43  27-34
Hsue et al.: Priscilla Y. Hsue, Peter W. Hunt, Amanda Schnell, S. Craig Kalapus, Rebecca Hoh, Peter Ganz, Jeffrey N. Martin & Steven G. Deeks, “Role of viral replication, antiretroviral therapy, and immunodeficiency in HIV-associated atherosclerosis”, AIDS 23  1059-67
This entry was posted on 2010/01/19 at 2:10 pm and is filed under antiretroviral drugs, clinical trials, experts, HIV risk groups, HIV/AIDS numbers. Tagged: Amanda Schnell, Anand Soni, Anne C. Moorman, Anne Taylor, B. Marchou, Beth Burtcel, Brian Carter, Christopher M. Dezii, Donald A. Wiebe, E. Bonnet, Felicity Boyd, Frank J. Palella, HAART and atherosclerosis, HAART and coronary disease, Hugh Kawabata, J. L. Montastruc, James D. Otvos, James H. Stein, James M. Sosman Judith S. Currier, Janet Lo, Jeffrey N. Martin Steven G. Deeks, Jeffrey Wei, Jen-Fue Maa, Jennifer L. Bellehumeur, Joan S. Chmiel, John T. Brooks, Jose A. Rocha-Filho, Jr., Kathleen C. Wood, Khurram Nasir, L. Cuzin, Leon Shturman, lying with statistics, Melissa A. Klein, P. Massip, Patrick E. McBride, Peter Ganz, Peter W. Hunt, Priscilla Y. Hsue, Rebecca Hoh, Rose K. Baker, S. Craig Kalapus, Sally Hodder, Scott D. Holmberg, Steven K. Grinspoon, Suhny Abbara. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.