Early diagnosis and treatment is standard advice for many medical conditions -- and for good reason. Cancer, heart disease and other ailments are much more effectively treated if diagnosed and treated early. So it was good news, but not a great surprise, when a study of asymptomatic HIV-infected patients with evidence of immune suppression demonstrated lower rates of clinical progression among patients treated with the antiviral drug AZT compared to untreated patients (ACTG019, NEJM, 4/90). These results, analyzed at the Biostatistics Dept., encouraged early treatment of HIV infection, and created large markets of healthy-appearing patients for newly-available drugs. Infectious diseases, like cancers, are generally easier to treat in their early stages; and laboratory data from antiviral studies confirmed the effect of AZT in reducing viral burden and improving immune systems among all classes of patients. Yet conventional wisdom and theories about how HIV disease progresses are no substitute for hard evidence about benefit; how strong is the evidence in favor of treating asymptomatic patients with AZT or other available antiviral drugs?
The consensus about early treatment was shattered when a recent study conducted in England in France (Concorde) found no benefit to patients randomized to receive AZT immediately compared to those randomized to defer AZT use until the onset of clinical disease or evidence of immune decline (Lancet, 4/94). The ensuing medical controversy has had all the elements of drama: politics, profits, deadly viruses, and patients inclined to be suspicious. Because of our involvement in the AIDS Clinical Trials Group, Department faculty have had prominent parts on stage. And such participation shows the value of biostatistics in creating a framework for rational debate about policy, among investigators under duress.
What does this statistican make of the evidence from the two studies? The first, ACTG019, was stopped early because signficantly fewer patients on AZT developed clinical disease; consequently, the proportion of patients who developed disease during follow-up (mean 55 weeks) was small (58 out of 1338 patients). As the investigators cautioned, we don't know whether the benefit conferred to patients at high immediate risk of progression applies to patients at lower short-term risk, and the follow-up was too short to permit evaluation of the effect on survival. Further complicating interpretation, this study discouraged the use of drugs that could delay onset of some AIDS-defining conditions, without treating the underlying disease. The Concorde study was considerably longer (median follow up of 3.3 years) and proceeded until a higher proportion of patients developed AIDS or died (347 out of 1749 patients). In this study there was an indication of early, but transient, clinical benefit in favor of immediate AZT use, but it was not sustained; the study concluded that there was no benefit to immediate use. (Continued follow-up of patients from ACTG019, all of whom were offered AZT after termination of the randomized portion of the study, demonstrated that the benefit of AZT, though significant, was transient as well.)
Yet Concorde does not provide clear guidance about treatment either. Concorde permitted patients in the deferred group to begin AZT if they had evidence of declining immune system during the course of the study. Although overall, the deferred group took much less AZT and fared no worse than the immediate group, one cannot conclude that immediate use of the drug had no benefit until one resolves the following issue: What if the patients at highest risk of clinical disease in the deferred group were in fact heavily treated with AZT? This might well have happened even though overall, patients in the deferred group took AZT only 16% of the time before disease onset. If it did, then the power of the study to detect differences between groups would be small, even if AZT mattered, because the treatment practices would be similar among those most at risk. Do the lab data from these studies help tell patients what to do? Unfortunately, not much. From analyses undertaken by Department faculty that account for measurement error and missing data, it appears that treatment-mediated changes in blood levels of viral protein or in measures of immune function are not very useful predictors of clinical benefit. The issue of whether asymptomatic patients should take AZT cannot be considered resolved, and the availability of new antiviral drugs compounds the problem. Fortunately, analyses underway in the Department of a much longer-term placebo-controlled study of AZT in patients with very early HIV infection will soon provide additional information.
This experience illustrates the challenges faced by Department faculty working in collaboration with clinical investigators: how to hasten the research process, but develop valid evidence about efficacy; how to make new agents as available as possible, but protect patients from excess toxicity and expense; how to use laboratory data to understand drug activity, but maintain the distinction between improving lab values and improving lives. The financial and emotional pressures to find effective drugs underscore the importance of the academic topics we study and teach: design of studies (including rules for early stopping), analyses of repeated measurements and their relationship to subsequent events, and combining evidence from different studies. However, we need not only react to and analyze data; statisticians played a key role in initiating a pilot study of immediate versus deferred antiretroviral therapy, which has the potential of resolving this issue.
In medicine, as elsewhere in society, concensus is more easily reached than the truth. Widespread belief may be more a reflection of commonality of interest than reliability of evidence. In recognizing and asserting that what passes for "knowledge" often consists more of beguiling narrative than substantiating fact, statisticians build what educators used to refer to as "character." To paraphrase the 18th century British statesman, Edmund Burke, a statistician says not what people want to hear but what they ought to hear.
Last modified $Date: 1994/12/01 20:23:33 $ by Evelyn Ophir, firstname.lastname@example.org