Why the Repeat Decision Is Clinical, Not Automatic
A concentration result does not have to be extreme to matter. A value close to a treatment boundary may change how a result is classified, while an apparently extreme value from a poorly timed or contaminated sample may not represent the patient’s concentration at all. The decision to repeat therefore depends on more than whether the number looks unusual: it depends on what might change if the result is wrong, how credible the sample is, and how much uncertainty could affect the decision.
A repeat is not always the same procedure. Reanalysis of the original specimen can help investigate analytical error, but it cannot fix an incorrect collection time or contamination during collection. A new, appropriately collected specimen can address some collection problems and estimate the concentration at a later time, but that later result may differ because the patient’s exposure has changed. Choose the kind of repeat to match the uncertainty being investigated.
The standard names for analytical uncertainty, sample contamination, minimum metadata, within-patient variability, concentration-to-target interpretation, and unexpected-result triage describe important inputs to this decision. Here the task is to combine those inputs into a defensible choice about whether to repeat, what to repeat, and how quickly to do so.
Start With the Consequence and the Sample
First identify the decision that depends on the result and the consequence of acting on a mistaken classification. If a value could prompt a large dose change, withholding treatment, or an urgent safety response, confirmation may be especially valuable when the result is uncertain. But a potentially serious toxicity signal is not a reason to wait passively for a repeat: urgent clinical assessment or protective action may be needed while confirmation is arranged.
Next ask whether the measurement can be interpreted at all. Check the collection time relative to the dose and intended sampling time, the dose history, the sampling site and method, and relevant handling or assay concerns. A result without the metadata needed to interpret it may warrant a properly timed new sample. A plausible analytical uncertainty calculation does not rescue a sample that may have a substantial, unknown collection bias.
- Reanalyse the same specimen when a laboratory or analytical issue is suspected and the specimen is still suitable. This does not test whether the collection time was correct.
- Collect a new specimen when collection, timing, or specimen integrity is in question, using a documented and clinically appropriate sampling procedure.
- Do not delay necessary care when symptoms, toxicity concerns, or other clinical evidence demand action. A repeat can inform ongoing management without replacing immediate assessment.
How Close Is the Result to a Decision Boundary?
A decision boundary is a concentration value used in a particular clinical context to distinguish categories or support a choice. It is not automatically a universal therapeutic threshold. The relevant boundary and its interpretation must be appropriate to the drug, indication, sampling time, and patient.
A simple uncertainty calculation can help show whether a repeat might change the category. Let \(C\) be the observed concentration and \(B\) an upper decision boundary. For a rough local approximation, represent analytical error by a coefficient of variation \(CV_a\), and expected within-patient variation by \(CV_w\). If both are applicable, approximately independent, and expressed on the same concentration scale, the standard deviation for a new measurement around a stable expected concentration near \(C\) is:
Here \(s_R\) is an approximate prediction spread for a new measurement, not a confidence interval for the current result and not a direct estimate of the patient’s “true” concentration. The calculation assumes that the relative uncertainties are reasonably described by the stated coefficients near \(C\), that their errors are independent, and that the patient’s dosing and clinical state remain sufficiently stable. It also uses a normal approximation on the concentration scale. For large relative variability, strong skew, or substantial change in the patient’s state, this approximation may be unsuitable.
Under those assumptions, if the expected concentration for the repeat is approximated by \(C\), the probability that the repeat falls on the other side of an upper boundary is:
In this expression, \(R\) is the new measurement and \(\Phi\) is the cumulative distribution function of a standard normal random variable. This is a model-based chance that a repeat would be classified differently, conditional on the stated approximation. It is not the probability that the current result is wrong, nor the probability that the patient’s concentration is below the boundary. For a lower boundary, the corresponding calculation uses the probability that \(R\) falls above that boundary when \(C\) is below it.
For reanalysis of the same specimen, expected within-patient variation is generally not the uncertainty being tested: the specimen has already been collected. If two analytical errors are independent and have the same standard deviation \(s_a\), the standard deviation of the difference between two assay results is \(\sqrt{2}s_a\), since the variance of their difference is \(s_a^2+s_a^2\). That comparison can help assess whether two assay results differ more than expected analytically; it does not address sampling bias or biological change.
Worked Example: A High Result With a Meaningful Chance of Reclassification
A hypothetical patient has a measured concentration of \(18.0\) mg/L. For this clinical decision, the team is using an upper boundary of \(15.0\) mg/L. Assume the sample is correctly timed, the patient’s state is stable, \(CV_a=5\%\), and expected within-patient variation is \(CV_w=12\%\). These values are stipulated for this example, not estimates from an actual study.
The result is \(3.0\) mg/L above the boundary. The approximate standard deviation for a new measurement is:
The approximate chance that a repeat falls at or below \(15.0\) mg/L is:
Under the stated assumptions, about 10.0% of repeats would be expected to fall on the other side of this boundary. That does not establish that the initial result is unreliable. The team should weigh this chance against the consequences of classification, the cost and delay of another measurement, and whether the result is clinically consistent. A repeat may be useful if the classification would materially change management and confirmation can be obtained safely.
Work Through the Decision in Context
Identify the action that depends on the result. Consider the harm of acting on a misleading result and the harm of delaying a necessary action.
Verify timing, dosing history, collection method, and specimen integrity. If a material problem is suspected, decide whether reanalysis or a new sample can actually address it.
Compare the result with the decision boundary that applies in this context. A result far from the boundary is less likely to be reclassified by modest measurement variation.
Use available analytical and within-patient variability information only when its assumptions are plausible. State whether the proposed repeat is intended to check the assay, the specimen, or a later concentration.
Arrange the most informative repeat without delaying urgent care. Document how the result will be interpreted and what action would follow either outcome.
Worked Example: A Result Just Above a Boundary
A hypothetical concentration is \(15.6\) mg/L, just above an upper decision boundary of \(15.0\) mg/L. The sample is correctly timed and has no known collection problem. Suppose \(CV_a=6\%\) and \(CV_w=10\%\), and the patient’s state is stable enough for the approximation.
The repeat-measurement spread is:
The probability that a repeat is at or below the boundary is approximately:
So, under these assumptions, a repeat could fall into the other category in roughly 37.1% of cases. That makes confirmation potentially informative if crossing the boundary would change management. It does not mean the patient has a 37.1% probability of being below the boundary now. If the decision is not sensitive to this small difference, or the repeat would not change management, repeating may add little value.
Worked Example: A Suspected Collection Problem
A hypothetical result is \(32.0\) mg/L, while the relevant upper boundary in this example is \(20.0\) mg/L. The sample may have been drawn from a line used to administer the drug, so contamination is suspected. The assay’s stated \(CV_a\) is \(5\%\).
The analytical standard deviation near the observed result would be approximately:
That calculation describes analytical imprecision under the stated assumption; it does not quantify possible contamination. Because the potential bias is unknown, the normal-variation calculation cannot establish how likely a valid repeat is to fall below \(20.0\) mg/L. Repeating the assay on the same potentially contaminated specimen could reproduce the misleading result. A new, appropriately collected specimen is the more relevant confirmation. If the patient has signs or symptoms that require urgent assessment, those should be addressed while the new measurement is arranged rather than deferred until it returns.
Comparing the Main Options
| Option | Best suited to | What it cannot resolve |
|---|---|---|
| Reanalyse the same specimen | A suspected analytical or reporting issue when the original specimen remains suitable | Incorrect timing, contamination during collection, or a change in the patient’s concentration |
| Collect a new specimen | A suspected collection or specimen problem, or a result near a consequential boundary when confirmation could change management | It is not a measurement of the exact earlier concentration; the patient or dosing state may have changed |
| Act without waiting for a repeat | A time-sensitive clinical situation in which delaying assessment or protective action could be harmful | It does not remove uncertainty in the result; additional measurement may still be needed for subsequent decisions |
The best option is not necessarily the one that produces the most measurements. A repeat is useful when it can address a specific uncertainty and could reasonably alter a decision. If the result is well away from a decision boundary, the specimen and metadata are credible, and a repeat would not change management, confirmation may not be worthwhile. Conversely, high stakes, a borderline result, or a correctable sample concern can make a repeat valuable.
Common Pitfalls and Limits
- “Repeat every unexpected value.” Repeating without identifying the uncertainty can add delay without improving the decision. Specify whether the concern is analytical, collection-related, or biological.
- “A small assay CV means the result is certain.” Assay precision does not rule out sampling errors, contamination, incorrect timing, or genuine within-patient variation.
- “A repeat crossing the boundary proves the first result was wrong.” A later concentration can differ because of assay error, biological variability, or a changed clinical state. Interpret both results in light of their sampling circumstances.
- “The calculated crossing probability is the chance the patient is on one side of the boundary.” The calculation predicts a repeat under a specified model; it is not a probability distribution for the patient’s current concentration.
- “The numerical boundary alone determines the action.” Clinical response, toxicity, dosing context, and the consequence of delay remain relevant. A concentration is one input to care, not a substitute for clinical assessment.
These calculations are aids to judgment, not universal repeat-testing rules. Their value depends on realistic uncertainty estimates, stable conditions, and an appropriate approximation to the distribution of repeat results. When uncertainty is dominated by an unknown bias, when relative variation is large, or when the clinical state is changing, describe the limitation rather than reporting a precise-looking crossing probability.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A measured concentration is close to a decision boundary, but the sample’s timing is uncertain. Why might reanalysis of the same specimen be insufficient?
- For a hypothetical concentration of \(24.0\) mg/L, an upper boundary of \(20.0\) mg/L, \(CV_a=4\%\), and \(CV_w=8\%\), calculate \(s_R\) and the approximate probability that a repeat is at or below the boundary.
- What does the probability calculated in Question 2 describe, and what does it not describe?
- A patient may have a serious adverse effect, and the concentration is unexpectedly high. Should assessment wait until a repeat result is available? Explain.
- Why does a suspected contaminated specimen call for a different response from ordinary analytical imprecision?
Solutions. (1) Reanalysis can check assay performance on that specimen, but it cannot correct an unknown collection time or establish that the specimen represents the intended sampling point. A new, properly timed specimen may be needed. (2) \(s_R\approx24.0\sqrt{0.04^2+0.08^2}=24.0\sqrt{0.008}=24.0(0.08944)\approx2.15\) mg/L. Then \(P(R\leq20.0)\approx\Phi((20.0-24.0)/2.15)=\Phi(-1.86)\approx0.0314\), or about 3.14%, under the stated assumptions. (3) It approximates the chance that a new measurement falls at or below the boundary, assuming a stable state and the specified normal and independence approximations. It is not the probability that the current concentration is below the boundary or that the original result is wrong. (4) No. Urgent clinical assessment and any necessary protective action should not be postponed solely to await confirmation; a repeat may still help inform subsequent decisions. (5) Analytical imprecision is represented by a bounded error model such as a CV, while contamination may introduce an unknown bias not captured by that model. Reassaying the same specimen can reproduce the bias, so a new, appropriately collected specimen is generally more informative.