The Ethics of Uncertainty in Medicine

Every medical choice is made with incomplete knowledge. A test can be falsely reassuring or falsely alarming. A treatment effect estimated in a group may not predict one person's result. A new symptom may have several plausible causes, while the information that would separate them has not yet arrived.

Uncertainty is therefore not an exception to medicine. It is a condition of responsible decision-making. The ethical problem begins when uncertainty is concealed, exaggerated, presented without a plan, or shifted onto a patient without guidance; a better approach states the uncertainty precisely, connects it to the available choices, and establishes what happens next.

Uncertainty is not one thing#

Han and colleagues described uncertainty through its sources, issues, and effects. One source is probability: an outcome may be possible but not guaranteed. Another is ambiguity, when evidence is missing, imprecise, or conflicting. Complexity arises when many interacting factors make outcomes hard to predict. A further category is ignorance, including questions that have not been recognized or studied.

These distinctions matter because each one asks a different response of you. Statistical imprecision may shrink with a larger study. A missing test result may arrive tomorrow. Conflicting guidelines may require examination of their populations and value judgments. Individual applicability may remain uncertain even when a trial's average result is precise. Some unknowns cannot be removed in time for the decision.

It is also useful to separate uncertainty from error. A diagnosis can be revised as a disease evolves without the earlier assessment having been careless. Conversely, saying "medicine is uncertain" cannot excuse failure to gather important information, communicate a serious possibility, or arrange follow-up. Ethical uncertainty is acknowledged and managed; it is not used as a shield against accountability.

Probability is a starting point, not a verdict#

Diagnostic reasoning begins with a pretest probability shaped by symptoms, prevalence, age, history, examination, and context. A test result changes that probability according to the test's performance in relevant settings. It rarely turns a possibility into zero or 100 percent.

This is why the same result can mean different things in different people. If a condition is very uncommon in a low-risk setting, a positive result may include many false positives, and if suspicion is already high, a negative result from an imperfect test may not safely exclude the condition. Sensitivity and specificity alone do not answer the post-test question without the starting probability.

Treatment evidence is also probabilistic. A randomized trial estimates an average effect with uncertainty around it. The confidence interval indicates the range of values compatible with the data and model at a stated level; it does not list every source of error. Selection, outcome definitions, adherence, follow-up, and changes in usual care affect how the estimate travels to a particular decision.

Ethical communication avoids turning these conditional estimates into promises. It also avoids treating uncertainty as total ignorance. "The evidence is imperfect" and "nothing is known" are very different statements.

Testing and treatment thresholds#

Pauker and Kassirer's threshold model formalized a practical idea: testing is most useful in the middle zone where its result can change action. Below a testing threshold, the condition may be unlikely enough that further testing causes more expected harm than benefit. Above a treatment threshold, waiting for another test may add little and could delay appropriate care. Between them, information can move the decision.

Thresholds are not universal constants. They depend on the seriousness and reversibility of the condition, treatment benefit, treatment harm, test risk, alternatives, delay, cost, and a person's preferences. A low-burden treatment for a dangerous condition may justify action at a lower probability than a toxic or irreversible intervention. A risky invasive test requires a different balance from a simple measurement.

The model also exposes why "order every test" is not neutral. Tests can lead to incidental findings, cascades, anxiety, procedures, false reassurance, and opportunity costs; at the same time, withholding a test that could materially change care requires a reason grounded in the person's circumstances, not convenience alone.

A good question is not merely, "Can this test find something?" It is, "What result is plausible, how would each result change the plan, and would that change improve the expected balance of benefit and harm?"

Autonomy needs usable information#

Informed consent is more than a signed form. The WMA International Code of Medical Ethics places respect for autonomy alongside duties to promote health, avoid harm, protect confidentiality, and use health resources responsibly. You need understandable information about the nature of the decision, the meaningful options, the material benefits and harms, and the relevant uncertainty.

Numbers help only when they are interpretable. Absolute risks should accompany relative effects where possible. The time horizon and population should be stated. Using the same denominator supports comparison: for example, outcomes per 1,000 people over five years, rather than mixing percentages, relative reductions, and lifetime risk. Presenting both chances of an outcome and chances of its absence can reduce framing effects.

Precision should match the evidence. Reporting 12.347 percent may imply knowledge that a broad confidence interval and uncertain applicability do not support. A range, an order of magnitude, or a verbal category tied to numbers can be more honest.

Uncertainty disclosure must also be relevant. Listing every imaginable doubt can overwhelm you and obscure the choices that matter. The aim is not to recite all unknowns. It is to explain those that could alter expected outcomes, preferences, urgency, or follow-up.

Shared decisions still include professional guidance#

AHRQ describes shared decision-making as a process in which clinicians and patients work together to make decisions informed by evidence, clinical knowledge, and the patient's values and preferences. NICE guidance similarly emphasizes discussing options, benefits, harms, consequences, and what matters to the person.

This process does not require equal technical roles. Clinicians are responsible for identifying medically reasonable options, interpreting evidence, explaining uncertainty, and making a recommendation when appropriate. Patients contribute knowledge about their goals, daily burdens, prior experiences, family responsibilities, and tolerance for possible outcomes.

Some choices have one clearly preferable path because delay would pose substantial harm and alternatives offer no comparable benefit. Others are preference-sensitive because options have different tradeoffs and no single choice dominates. Ethical communication should say which of the two you are in.

People also vary in how much participation they want. Respect can include a request for a stronger recommendation or for a trusted family member to join, provided consent and capacity are addressed. Shared decision-making should not become a ritual that forces you to perform expertise or carry responsibility that belongs with the care team.

A practical structure for communicating uncertainty#

Communication can begin with the current best explanation: what is most likely and why. Next comes calibrated confidence, such as "the pattern strongly supports," "several explanations remain plausible," or "the evidence cannot yet distinguish these options." The explanation should identify the major alternative that matters, not only the leading diagnosis.

The next layer is what is known from evidence and how well it applies. Were the relevant studies randomized or observational? Did participants resemble the person facing the decision? Are outcomes important to patients, or mainly laboratory measures? How wide are the uncertainty intervals? Are recommendations consistent across credible guidelines?

Then come the reasonable choices. Each should include expected benefits, important harms, burdens, timing, and the option of watchful waiting when it is medically reasonable. If one option is preferred, the reason should be explicit. A recommendation can remain conditional: "Given the current risk and your goal to avoid delay, this option has the best expected balance. If the pending result changes the probability, the plan changes too."

Finally, uncertainty needs ownership. Who reviews the result? How will the person receive it? What symptoms require earlier contact? When will the diagnosis or treatment be reconsidered? What is the fallback if the expected course does not occur? A documented safety net converts uncertainty from an open-ended burden into a monitored process.

"No evidence" and "evidence of no effect"#

These phrases are often confused. "No evidence" can mean that no adequate study exists, that available studies are too small, or that an analysis did not cross a statistical threshold. "Evidence of no meaningful effect" requires data precise enough to rule out effects that would matter.

A nonsignificant result does not automatically support equivalence. A wide confidence interval may include substantial benefit and substantial harm. Conversely, a very precise estimate near no difference can make an important effect unlikely. The ethical message depends on precision, design quality, and the decision threshold, not a p-value label alone.

The same care applies to uncertain diagnosis. "Tests are normal" does not mean symptoms are unreal. It means specified findings were not detected under the conditions of those tests. The next question is what explanations remain, what risk has been reduced, and what follow-up is appropriate.

Trust cannot be guaranteed by a script#

Studies of diagnostic-uncertainty communication show varied language, outcomes, and contexts. Disclosure can be received as honesty, but poorly framed uncertainty can also create confusion or seem evasive. It is not defensible to claim that saying "I am uncertain" always raises trust.

Context and structure matter. Uncertainty paired with reasoning, empathy, a recommendation, and a follow-up plan is different from uncertainty offered without support. Communication should neither perform confidence nor dramatize doubt. Its aim is an accurate shared understanding of the decision.

Trust also depends on whether the plan is carried out. A clear promise to review a result must be supported by a reliable result-management system. Ethical communication and system design are connected: words cannot compensate for lost referrals or unreviewed tests.

Fairness and stewardship#

Decisions consume time, tests, medicines, procedures, and attention. Stewardship asks whether those resources produce enough expected benefit and whether access is fair. It does not justify denying useful care through hidden rationing. Relevant limits and criteria should be transparent and consistently applied.

Uncertainty can amplify inequity when evidence underrepresents certain populations, language access is poor, follow-up is harder to obtain, or risk tools encode past patterns of unequal care. Saying that an estimate may not apply equally is only the beginning. The plan may need better interpretation, a different follow-up route, or a lower threshold for reassessment when system barriers make delay more dangerous.

Ethical practice therefore treats uncertainty as both an evidence problem and a responsibility problem. The questions are not only "What is the probability?" but also "Who bears the risk if the estimate is wrong, and what safeguards are available?"

References#

1. Han, Klein, and Arora, varieties of uncertainty in health care
2. WMA International Code of Medical Ethics
3. NICE shared decision making recommendations
4. AHRQ shared decision making overview
5. Pauker and Kassirer, therapeutic decision thresholds
6. Systematic review of diagnostic-uncertainty communication
7. National Academies, Improving Diagnosis in Health Care