Commentary on Gulati et al., p. 740
You cannot always get what you want. In epidemiology, this universal axiom can be translated as the question we want to answer is not necessarily the question we can answer. As we often lack the information needed to answer the desired question, we settle on a related, manageable question that can provide at least some insight into the problem at hand.
Consider the first decision a man faces on diagnosis with PSA (prostate-specific antigen) detected prostate cancer: treat or wait. Therefore, we want to answer “What are the consequences of postponing treatment?” However, as it is difficult, if not impossible, to currently answer this question, we could settle on the easier “What are the chances that the patient will never experience any clinical symptoms?” Although this prognostic question will not give us the exact answer we want, it would still provide useful information for a patient. Gulati and colleagues (1) can answer this question and offer the answer as part of a wider inquiry into the outcomes from never treating cancer. However, when the question desired and the question answered do not coincide precisely, it is important to evaluate our chosen replacement and remind ourselves of the true question, the information needed to answer that question, and the limitations of the provided answer.
A man diagnosed with prostate cancer has two options. The first option is to immediately treat with a primary treatment such as a prostatectomy or radiation. The second option is active surveillance, where he monitors his condition, possibly using secondary treatments, and reconsiders a primary treatment if his cancer progresses or PSA rises. Therefore, to be specific, the question he would want answered is “What are the consequences of postponing treatment if progression is carefully monitored?” The consequences of his choice are the differences that would result from the two treatments. How much more likely is he to die from prostate cancer if he practices active surveillance, as compared with pursuing immediate treatment? How much less time would he suffer from serious side effects, like impotence and incontinence, if he practices active surveillance? As do the authors in the original article, we believe that the side effects expected under both treatments have been well documented, so we will focus on the main outcome.
The best answer to any comparative efficacy problem would come from a randomized clinical trial of men similar to the patient. In one arm of the trial, we treat newly diagnosed individuals immediately. In the second arm, we define criteria for progression and only perform the primary treatment when the individual meets these criteria. Unfortunately, no such randomized trial has been performed in the PSA-screening era and there are no plans for such a trial in the near future.
It is the absence of this study that mandates settling on an easier question. Motivated by the possible similarity between active surveillance (i.e., postponing primary treatment until progression) and postponing any treatment, primary or otherwise, until the onset of clinical symptoms, one reasonable substitute would be “What are the consequences from taking no action until clinical symptoms?” Although not advertised as such, the ideal PSA-screening trial would effectively address this question. The screening arm would be a randomly selected group of individuals that receive immediate treatment for PSA-detected prostate cancer. The control arm would receive no treatment until the onset of clinical symptoms. Therefore, the likelihood of there being no consequences from postponing treatment is the difference between the proportion of individuals diagnosed with prostate cancer in the screening arm and the proportion diagnosed within the control arm. The increase in the mortality rate from postponing treatment is the increase observed in the control arm, as compared with the screening arm.
Unfortunately, for obvious ethical reasons, PLCO (2) and ERSPC (3), the two main PSA-screening trials, could not force individuals in the control arms to refrain from screening. Consequently, the possibly substantial cross-over complicates the interpretation of the results and the ability to answer the proposed question. The observed, unadjusted differences in diagnosis and mortality rates would likely underestimate their true values. Therefore, when answering the replacement question with trial results, it will be important to carefully account for the impurities in the screening arm.
Gulati and colleagues pose a slightly different question, asking about the outcomes from never treating prostate cancer. From the perspective of a newly diagnosed individual, this question would be too broad. He would have no reason to consider the possibility of never seeking treatment, only the possibility of postponing treatment until it was needed. Estimates of the overall mortality rates, which still assume no treatment after clinical symptoms, are not directly relevant. Nevertheless, in answering that broad question, Gulati and colleagues also address the previously defined replacement question and explore the results from taking no action until clinical symptoms. A patient would find great value in this part of their discussion.
Gulati and colleagues first estimate the likelihood that a patient would never experience any symptoms. Their approach avoids the potential bias encountered in a screening trial and allows them to tailor their estimates to the individual characteristics of a patient. In addition, Gulati and colleagues also estimate the likelihood of a metastasis occurring prior to clinical symptoms in an untreated individual. If a patient is willing to assume this estimate reflects the likelihood of a metastasis prior to progression, as defined for active surveillance, these rates would be equally valuable. The limitation of their approach is that their models can project only the natural history of progression and cannot incorporate treatment. Therefore, their discussion is limited to the outcomes, as opposed to the consequences, of postponing treatment. Their models cannot project, for example, metastasis rates if immediate treatment were chosen. For the needed comparison, we must make assumptions about the outcomes after treatment, such as primary treatment prevents any metastasis. Moreover, a patient should only use the model to project events that occur prior to the onset symptoms if he is likely to seek treatment upon their presentation, as treatment would obviously alter his prognosis.
Studying the costs from waiting until clinical symptoms is one of the more attractive ways to estimate the potential costs of active surveillance. In this commentary, we discussed two tools for this study, PSA screening trials and natural history models. As the screening trials mature, and more outcomes are reported, their contribution is likely to increase. The caveat is that we are making an assumption that the effects of blindly postponing treatment until symptoms and postponing treatment while monitoring progress are similar. Because this is a large assumption, there is a general need to search for other insightful questions, and, even more importantly, to start collecting the information needed to directly evaluate the consequences of active surveillance.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.