What’s your blood count? Blood pressure? Bone density? PSA? If it is abnormal, odds are that you will want it to be normal. In doing so, you are making the same leap of faith that your doctor is making when he commences treatment: that treating the numbers will improve your health. Like much of what we do, treating the numbers is often naïve, and sometimes harmful, no matter how well intentioned. Read these short examples and tell me if you still want your numbers normalised.
- Tight control of blood glucose levels in a medical intensive care unit does not improve outcomes, and it has been shown to be harmful in another study.
- Attempts to prop up the haemoglobin level of patients with chronic kidney disease raised the haemoglobin level, but caused overall harm.
- Attempts to prop up the blood pressure in trauma patients caused more bleeding, leading to the current policy of ‘permissive hypotension’ (low blood pressure).
- Propping up the haemoglobin in critically ill people makes them slightly more likely to die.
- This study showed that giving patients beta-blockers to control their heart rate and blood pressure around surgery lowered their chance of dying from a heart attack, but increased their chance of dying (from anything).
- Giving oxygen to newborn babies who can’t breath properly (and have low oxygen levels) increases their chances of dying.
The problem here is the use of “surrogate outcomes”. This is where we control a surrogate value, as a short cut for fixing the real thing. In theory, this sounds fine but it is only valid when the connection between the surrogate and the real thing is pretty tight. Sometimes, using a surrogate is valid, but there are many reasons why control of your surrogate measure might not change the clinically important outcome. For example:
- The two may not be causally related
- Treatment of extreme values may be helpful, but generalisation to milder cases doesn’t work
- The connection may be valid but the treatment might not alter the surrogate value
- The treatment might alter the surrogate, but have unintended harmful consequences via another pathway.
Sometimes it works, for example if you control the blood glucose levels of diabetics, they are less likely to have diabetic-related complications in the future. But so many of the surrogate outcomes we use have either not been validated, or have been shown to be invalid measures of the real, clinically important outcome.
The famous case
A famous case of the failure of surrogate outcomes is that of flecainide (and related antiarrhythmic drugs). In the period after a heart attack, some patients die suddenly of an arrhythmia (abnormal heart rhythm). In order to prevent these sudden deaths, doctors gave antiarrhythmics. This sounds completely logical, but as many of you know, biological explanations don’t carry much weight with me. The effectiveness of these drugs was proven in clinical trials where they gave some patients antiarrhythmics and some patients placebo and then followed them around with continuous heart monitors: the patients who took the antiarrhythmics definitely had less arrythmias.
Based on these trials, the drug was approved by the FDA (thanks to the flawed accelerated approval process) and used all across the USA on hundreds of thousands of patients in the 1980s. Although an association with lower mortality was not established, doctors felt that a randomised trial to test this would be unethical, as the drugs clearly worked (by now, regular readers will understand why I flinch whenever I hear that claim).
My argument is that instead of reading reams of printouts from heart monitors and counting the regular and irregular beats (the surrogate outcome), why didn’t they measure the mortality (the relevant outcome)? It would have been easier. Well that’s what some doctors did, in a large trail called CAST (Cardiac Arrhythmic Suppression Trial), which showed that you were much more likely to die if you were given the antiarrhythmic drug. I guess the only consolation was that you definitely had less arrhythmias before you died. The rate of death with the drugs was so high that they had to stop the trial early. Overall, the introduction of those drugs (based on the surrogate outcome tests) was responsible for ten of thousands of deaths in the US alone.
The normalisation heuristic
Nowhere in medicine is the ‘strive for normal’ greater than in the intensive care unit. The basis for most treatment decisions in intensive care is to make a number normal. It might be the pulse, oxygen, temperature, blood clotting, blood glucose, haemoglobin, salts (Na, K, Ca, Mg, etc.), urine output, blood urea and creatinine, or the blood pressure, pulmonary artery pressure, respiratory pressure, intra-cranial pressure, abdominal pressure, or compartment pressure (once the pressure transducer was invented, it got used, everywhere). Prescribing the drugs gets tricky though, particularly when each drug raises one number but lowers another, and other drugs need to be given to counter the effects of the previous drug. Intensive care patients often end up on a finely balanced regimen of multiple drugs, and have tests done repeatedly each day just to get the fine-tuning right.
This obsession with the ‘normal’ has been termed the “Normalization Heuristic” and is explained extremely well by the authors (follow the link).
Outside of the intensive care setting, you will regularly see doctors treating numbers and test results, always hoping that by doing so, some measure of health improvement may result. In my field, this is referred to as “treating the X-ray”. But even effective surrogate treatments like cholesterol lowering drugs and blood pressure lowering drugs only reduce the absolute chance of death by 1.5%, and sometimes only when treatment is pursued aggressively, as many studies barely even control the surrogate, let alone the main outcome. The effectiveness of blood pressure and cholesterol control are particularly called into question when we aim to treat milder and milder cases.
The bottom line
Be skeptical next time a doctor wants to treat a number (PSA, hormone level, cholesterol, blood pressure, arterial stenosis, or even an X-ray or MRI finding) instead of your health. Ask about the evidence of the treatment making a difference to the outcome that is important to you (whether that be function, quality of life, survival or something else).