“What if it were your own mother, what would you tell her?”
Patients may use this or similar language, invoking both the merits of familial loyalty and candid advice when facing different – and difficult – options for their care. This question can anchor their own decision making in a more humane and empathetic context, and sidestep overly scientific or statistical discussions. Whether the physician’s response literally or figuratively considers their own mother, patients may understandably find solace to navigate their situation more confidently in times of fear and vulnerability through their doctor’s personal perspective.
Yet this “mother-statistic”, like p-values and other statistics, can be misunderstood and inadvertently steer the decision ship into uncertain waters. Ultimately, navigating uncertainty in medical decision making involves two types of information: evidence from past experience and research about the medical condition in question, and the risk tolerance of those choosing between options that each have some uncertainty.
Even if we prefer to use qualitative thinking to assess the information of these two categories, it is useful to at least consider how statistics can provide a framework for decision making under uncertain circumstances. Let’s think about two examples where the translation of a statistical concept into “normal” language could result in muddy waters.
First, if we think of p-values as a measure of how likely an effect is to be real (versus chance), even putting aside the critiques of p-values in general [1], it is unknown whether a patient would apply the standard 5% threshold if judging their own treatment and chances of success. In fact, more liberal and more strict thresholds would both alter the way in which medical evidence would be considered, but such discussions rarely if ever occur it seems.
Second, let’s consider how the patient perspective on treatment choice might be swayed by the way data is presented. Take the example of carotid endarterectomy, a surgery to open up clogged neck artery to reduce stroke risk. Extensive evidence suggests that, for patients at high risk of having a stroke, there is a substantial reduction in risk of future strokes by this endarterectomy procedure. We may counsel patients that this approach makes sense: since the cholesterol plaque likely causes strokes, we remove the offending clog, and thus prevent future stroke events. A simple plumbing problem, solved by highly specialized plumbing procedure. The medical evidence is among the most clear and convincing in neurology, and the narrative of why this surgery works is indeed clear and compelling. But now let’s phrase the same problem a different way. Line up 8 men, each with high stroke risk and a clogged carotid. Then say to them: “We need to operate on all 8 of you in order to prevent a stroke in one of you”. This might be interpreted in a different light than the more qualitative narrative of why the surgery works.
Stating the benefit in this way is known as the number needed to treat, an index of how effective a treatment is that takes into account the baseline risk of some event before versus the risk after some intervention. Here, the stroke risk is reduced from approximately 20% down to 8%, a 12% absolute benefit. The inverse of this fraction (1 / 0.12) is about 8. This risk reduction is quite favorable compared to many other medical treatments. On a “population” level, performing endarterectomies makes perfect sense and prevents thousands of strokes every year. Phrasing it in the second way, however, carries a less favorable view compared to the earlier plumbing narrative. The exact same data are used to create each narrative. In neither case is the “real” semantic – that of probability – expressly stated. In the second narrative, though, one is confronted with the implication of probability: it is possible (perhaps even likely) that the operation will not actually prevent the stroke as intended. But with an understanding of probability, a patient might still rationally opt for surgery to improve the probability of stroke.
In the end, whether we know or care about statistics and decision theory, we all have some heuristic for risk tolerance. When patients request that I calculate the proverbial mother-statistic for their case, I phrase my answer by sharing the wisdom of both of my parents: my father is relatively risk tolerant while my mother is relatively risk averse. This reminds me to describe how people at each end of the risk tolerance continuum might reason through a medical decision. Whatever method a person actually uses to reason through decisions (including the enteric route of pure gut reaction), recognizing that complicated decisions may not have a clear mother-statistic answer may be as important to the decision process as the final decision itself.
Contributed by: Matt Bianchi, MD PhD
A version of this blog was originally posted on HorseAndZebra (no longer active), in 2011
References
[1] www.ncbi.nlm.nih.gov/pmc/articles/PMC3058025/
Patients may use this or similar language, invoking both the merits of familial loyalty and candid advice when facing different – and difficult – options for their care. This question can anchor their own decision making in a more humane and empathetic context, and sidestep overly scientific or statistical discussions. Whether the physician’s response literally or figuratively considers their own mother, patients may understandably find solace to navigate their situation more confidently in times of fear and vulnerability through their doctor’s personal perspective.
Yet this “mother-statistic”, like p-values and other statistics, can be misunderstood and inadvertently steer the decision ship into uncertain waters. Ultimately, navigating uncertainty in medical decision making involves two types of information: evidence from past experience and research about the medical condition in question, and the risk tolerance of those choosing between options that each have some uncertainty.
Even if we prefer to use qualitative thinking to assess the information of these two categories, it is useful to at least consider how statistics can provide a framework for decision making under uncertain circumstances. Let’s think about two examples where the translation of a statistical concept into “normal” language could result in muddy waters.
First, if we think of p-values as a measure of how likely an effect is to be real (versus chance), even putting aside the critiques of p-values in general [1], it is unknown whether a patient would apply the standard 5% threshold if judging their own treatment and chances of success. In fact, more liberal and more strict thresholds would both alter the way in which medical evidence would be considered, but such discussions rarely if ever occur it seems.
Second, let’s consider how the patient perspective on treatment choice might be swayed by the way data is presented. Take the example of carotid endarterectomy, a surgery to open up clogged neck artery to reduce stroke risk. Extensive evidence suggests that, for patients at high risk of having a stroke, there is a substantial reduction in risk of future strokes by this endarterectomy procedure. We may counsel patients that this approach makes sense: since the cholesterol plaque likely causes strokes, we remove the offending clog, and thus prevent future stroke events. A simple plumbing problem, solved by highly specialized plumbing procedure. The medical evidence is among the most clear and convincing in neurology, and the narrative of why this surgery works is indeed clear and compelling. But now let’s phrase the same problem a different way. Line up 8 men, each with high stroke risk and a clogged carotid. Then say to them: “We need to operate on all 8 of you in order to prevent a stroke in one of you”. This might be interpreted in a different light than the more qualitative narrative of why the surgery works.
Stating the benefit in this way is known as the number needed to treat, an index of how effective a treatment is that takes into account the baseline risk of some event before versus the risk after some intervention. Here, the stroke risk is reduced from approximately 20% down to 8%, a 12% absolute benefit. The inverse of this fraction (1 / 0.12) is about 8. This risk reduction is quite favorable compared to many other medical treatments. On a “population” level, performing endarterectomies makes perfect sense and prevents thousands of strokes every year. Phrasing it in the second way, however, carries a less favorable view compared to the earlier plumbing narrative. The exact same data are used to create each narrative. In neither case is the “real” semantic – that of probability – expressly stated. In the second narrative, though, one is confronted with the implication of probability: it is possible (perhaps even likely) that the operation will not actually prevent the stroke as intended. But with an understanding of probability, a patient might still rationally opt for surgery to improve the probability of stroke.
In the end, whether we know or care about statistics and decision theory, we all have some heuristic for risk tolerance. When patients request that I calculate the proverbial mother-statistic for their case, I phrase my answer by sharing the wisdom of both of my parents: my father is relatively risk tolerant while my mother is relatively risk averse. This reminds me to describe how people at each end of the risk tolerance continuum might reason through a medical decision. Whatever method a person actually uses to reason through decisions (including the enteric route of pure gut reaction), recognizing that complicated decisions may not have a clear mother-statistic answer may be as important to the decision process as the final decision itself.
Contributed by: Matt Bianchi, MD PhD
A version of this blog was originally posted on HorseAndZebra (no longer active), in 2011
References
[1] www.ncbi.nlm.nih.gov/pmc/articles/PMC3058025/