Medicine Without Answers

Embracing uncertainty in healthcare

Millions of Americans are walking around taking blood thinners. They fall and hit their heads. Or a blood vessel in the brain ruptures spontaneously. Regardless of the cause, the question becomes the same: when should the blood thinner be restarted?

Wait too long, and the patient risks a complication from a blood clot. They were on anticoagulation for a reason. A stroke or pulmonary embolism can be catastrophic or fatal. Restart it too soon, and the brain hemorrhage may expand, also potentially fatal.

Often, a neurosurgeon is consulted simply to answer the question of “is it safe to restart a blood thinner?”

Which is why this tweet resonates so strongly.

There is no magic point at which restarting anticoagulation is “safe.” There is no duration for which holding it is “safe” either. What exists instead is a tradeoff between two serious risks, neither of which is well quantified.

There is remarkably little high-quality scientific evidence guiding the optimal timing of anticoagulation restart after intracranial hemorrhage. The injuries themselves are heterogeneous: from tiny specks of blood outside the brain, to contusions, to large hemorrhages within the brain tissue. Meanwhile, the indications for anticoagulation are equally varied: atrial fibrillation, mechanical heart valves, prior strokes, prior pulmonary emboli.

So the decision requires balancing two unknowns: the risk of hemorrhage expansion and the risk of a thrombotic event.

Modern medicine is uncomfortable with this level of uncertainty, so it reaches for algorithms. Risk scores, decision trees, and guideline tables offer the appearance of precision. Tools like CHADS-VASc can give a very rough estimate of stroke risk in one specific pathology. There is no comparable, validated tool that tells us the probability that this specific hemorrhage, in this specific patient, will expand if anticoagulation is restarted on day three instead of day seven.

Many large hospital systems and public health advocates want to rely on algorithms for medical care. Yet these algorithms are based on imperfect data, often imperfectly applied. They often fall short in real world situations. These patients do not resemble one another. Their injuries do not behave predictably. Their competing risks are asymmetric and individualized. Forcing this problem into an algorithm does not resolve uncertainty.

The consult to the neurosurgeon often functions less as a data-driven exercise and more as a risk transfer. Neurosurgeons do not have privileged access to hidden evidence. They are asked to opine not only on brain injury risk, but on vascular risk, cardiology risk, and implicitly, on how much risk a patient should be willing to tolerate. What is really being requested is not certainty, but ownership of uncertainty.

Which is why the most important skill in these situations is not plugging numbers into an actuarial table, but communicating what is known and what is not. The physician’s role is to synthesize the limited data, clinical experience, standards of care, and patient-specific factors, then convey that uncertainty honestly. That includes acknowledging tradeoffs, eliciting patient values, and arriving at a shared decision.

Patients generally understand and appreciate this. Most do not expect perfection. What they want is transparency. They want to know why a recommendation is being made, what the risks are on both sides, and where judgment fills in the gaps left by evidence.

This is a reminder that medicine is not algorithmic at its core. Clinical care cannot always be reduced to a yes-or-no answer, a flowchart, or a guideline checkbox. In the highest-stakes decisions, uncertainty is irreducible. Pretending otherwise may be comforting to clinicians and institutions, but it is dishonest to patients.

The most responsible thing a physician can do in these moments is not to manufacture certainty, but to explain its absence and still help the patient choose.