The red pill or the blue one?

I break for chemotherapy, and so can you.

It may surprise some, but there is often no definitive standard of care for those individual patients who vary from the Olympian athlete profile of the typical clinical trial participant. This is true of all fields of medicine, from anaesthesiology to trauma care. To therefore discuss the varied options requires a Sisyphean mental effort to make sense of all available data in the light of individual circumstances.

Many situations demand some improvisation. Consensus guidelines may be idols with feet of clay. There really is little purpose in recommending a cocktail of six different medications, if the little old lady staying alone is having problems with her eyeglasses, and thus prone to mixing up the red and the blue pills. Given inevitable modifications of best practice, a discussion of options is usually required for most patients.

Since the concept of informed consent entered respectable academic spheres in the 1960s, intelligent doctors in the Anglo-Saxon tradition have nodded sagely in agreement at bioethics conferences that our patients should fully understand the benefits and risks of all interventions. Informed consent can however only exist when there is a clear communication of the potential effectiveness and harm inherent in available treatment options. Such a discussion is particularly critical for my oncology patients, given that cancer treatment is viewed by many as being expensive, toxic and accompanied with only marginal gains.

What kind of outcomes?

How effective is a treatment, really? How harmful is it? A good understanding of outcomes is key to an educated discussion, and when life is limited, the measurement of these outcomes can be particularly complex. As cancer physicians, we seek to either cure the patient of cancer, or to provide palliation by controlling the disease extent. Throughout all these efforts, relief of symptoms is uppermost. In few other fields is the dictum of ‘to cure sometimes, to relieve often, to comfort always’ so evident in daily work.

When cure is the goal, discussion can be relatively simpler, since the goal is clear. We extend a hand in companionship – though one may walk through the valley of the shadow of death, one does not walk alone, for we are there with you, steadily walking together towards the crack of light at the end.

When cure is not an option though, things can get a little murky in terms of outcomes. With a specific treatment, who will benefit with prolonged life, who will be harmed with side effects, and how much overlap is there between these two samples from the same group?

Making a decision

Suppose we have an intravenous anti-cancer drug. Let’s call it Drug X for convenience. It doesn’t cure cancer, but it has a chance of controlling it. Ms. Lim has metastatic cancer, and her life is necessarily limited.

Median survival

If Ms. Lim chooses not to receive this drug, she will have a median survival commonly quoted as being 16 months. If she chooses to receive the drug, she will have a median survival of 20 months.

Essentially, if she is untreated, she has a 50% chance of living to 16 months and beyond. She also has a 50% chance of living less than 16 months. With drug X, she will have a median survival of 20 months, meaning a 50% chance of living to 20 months and beyond. Similarly, she also has a 50% chance of living less than 20 months, even with drug X on board.

In my experience, however, patients hang on to their doctor’s words, and discussions centring around median survivals will result in patients concluding (erroneously in many situations) that Drug X will prolong their lives by several months, and that they are destined to live 20 months. This will be true for some, but untrue for many, many more. The median is not the message!

The relative risk

Some doctors will choose a second option, the relative risk. The same data above may be presented as – “Drug X will reduce your chance of death by 34%”. Essentially, it is true that one’s chance of death is reduced by 34% at any point in time. Doesn’t it sound good? Unfortunately, relative estimates of risk are more often than not deceiving. Suppose I was able to reduce a 1% risk of toxicity to a 0.5% risk, this would represent a relative 50% risk reduction, even though in absolute terms, the risk reduction is 0.5%. Most drug evangelists and pharmaceutical companies have made it a fine art to trumpet the relative risk, and it is rare that pharma literature mentions anything other than the relative risk. The layperson who does not understand these finer points may think that his treatment is far more effective than it actually is.

Numbers needed to treat for one positive outcome

We move on to absolute risk reduction, which may most closely approximate what is truly useful to doctors and patients. Curiously enough, it is usually the proverbial needle in the haystack in journal articles! Suppose we choose a time point of 2 years. The same data indicates that the chance of living to 2 years is about 45% for drug X, and 35% without drug X. This means corresponding to an absolute risk reduction of approximately 10%. Therefore, for every 10 people I treat, 1 additional person will live to 2 years. (The irrational me will obviously hope that the additional person successfully treated is my own patient!) A 10% relative risk reduction corresponds to a numbers needed to treat (NNT) of 10 people. Likewise, a 5% absolute risk reduction corresponds to the reciprocal, an NNT of 20 people. To place modern cancer research in sobering perspective, few advances in the last 10 years have exceeded an absolute risk reduction of 10% in terms of 5 year-survival from diagnosis.

This is probably the closest way for a patient to understand the true benefit of a treatment option. Once again, it teases, it falls short, it disappoints.

Weaknesses of the NNT

Given enough time and choosing a distant enough time point, with an outcome based on survival (the oncologic gold-standard), the absolute risk reduction in terms of survival is zero. That’s quite easily understood, since all of us have to go at some point or another. Hence, the magnitude of that benefit also depends to an extent on the choice of time-point. Indeed, for drug X, the benefit is likely to reduce as time wears on beyond 2 years. So are we choosing the most convenient time points to maximize our own perception of benefit, really?

Unfortunately, to make things worse, the NNT is commonly misinterpreted by evidence-based zealots and clinicians who claim some knowledge of quality and cost-effectiveness. In the example above, with an NNT of 10 (with an additional 1 in 10 people surviving to 2 years with Drug X), some would suggest that 9 of 10 did not derive any benefit. This fallacy is commonly seen in the scientific literature. Might not an individual person Mr. B, living to eighteen months have benefited from Drug X, if his survival was likely to have been 1 year untreated? Such an outcome for Mr. B is unfortunately not classified as benefit under the mantra of absolute risk reduction at a time point of two years. Perhaps all patients benefited, with life prolonged several months for all, but not achieving the two year cutoff. In this, the use of median survival more readily captures the benefit of palliative treatment, but in terms that are hardly understandable to the individual patient.

Perhaps the wise oncologists who only nod and smile to reassure their patients know more about informed consent than one thinks.

For the record, drug X is a monoclonal antibody bevacizumab, also known as Avastin. The data referred to is derived from a major study in metastatic colorectal cancer, where patients who received a cocktail of palliative chemotherapy were divided randomly into two groups – one group additionally receiving bevacizumab, and the other placebo.