Why a diagnostic test needs both analytical and clinical performance: bench, population, ethics committee, samples, MDCG 2022-2, IVDR, RDC 830/2023 and ISO 20916.

Before reaching the market, every diagnostic test has to answer two different questions. The first is whether it measures well what it sets out to measure. The second is whether that result says something true about the person being tested. They look like the same question, and it is precisely this similarity that leads many projects to treat the second as a natural consequence of the first.
The two answers come from distinct stages of performance evaluation: analytical performance and clinical performance. The European guidance on clinical evidence for in vitro diagnostics, MDCG 2022-2, treats each of them as a pillar in its own right, alongside scientific validity, which is the proven association between what the test detects and the condition it is meant to indicate. Understanding what each stage evaluates helps show why neither can do without the other.
At the bench, the test finds the target
Analytical performance evaluates the test under controlled conditions. That is why it can use materials that do not come directly from patients: reference materials, controls, panels and contrived samples, prepared with known amounts of the target spiked into a negative matrix. The samples are well characterised, the concentrations are known and the goal is to find out whether the assay detects what it claims to detect. This stage covers questions such as the smallest amount of the target the test reliably finds, whether it reacts to similar substances or microorganisms, how much the result varies when the assay is repeated on different days, by different operators and with different lots, and whether components of the sample itself, such as excess fat or medications in use, interfere with the reading.
These answers are indispensable. Without them, there is no way to guarantee that the observed signal corresponds to the target. All of them, however, describe the test in front of samples chosen to test it, not in front of the people who will use it.
In the population, the result has to mean something
Clinical performance changes the question. Instead of checking whether the test finds the target, it evaluates whether the result actually relates to the clinical condition, in the population the product is intended for. This is the stage where clinical sensitivity and specificity appear, calculated against a reference method, along with predictive values, which indicate the chance that a positive or negative result is correct. Here the rule is to work with real samples, collected from people in the population the test is intended for.
The real population brings variables the bench does not reproduce. Samples arrive collected by different professionals, stored for different lengths of time, from people at different stages of the disease and with other conditions that resemble the one being looked for. The frequency of the disease also changes from one place to another, and that alters the meaning of each result. A test with 98% sensitivity and 99% specificity gets about 92% of positive results right where the disease affects one person in ten. Applied where it affects one person in two hundred, the same test gets only about one positive in three right. The assay is still the same. What changed was the population.
What the clinical study requires
Because it involves people and human material, clinical performance depends on a structure the analytical stage does not require. The first layer is ethical. In Brazil, research with human beings follows Law 14.874/2024 (Lei 14.874/2024), and the study protocol is reviewed by a research ethics committee before it begins, including when it works with samples already collected. The committee assesses how samples will be obtained, how participants will be informed and how their data will be protected, and this timeframe needs to be in the schedule from the start.
The second layer is the biological sample. The study can work with prospective samples, collected specifically for it, or retrospective ones, already stored and reused, and each choice has consequences for the representativeness, traceability and quality of the material. The number of samples, the ratio of positives to negatives and the origin of each one are defined in the study design, because they determine the degree of confidence the result will have.
The prospective study is the design closest to real use, but it is expensive and slow. When the disease is rare, it may be unfeasible: if it affects one person in a thousand, gathering 50 positives would require testing around 50,000 people. In these cases, the study turns to retrospective samples. When the frequency of the disease in the country does not allow enough real positives to be gathered, regulatory guidelines accept completing the set with contrived samples, provided the choice is justified in the study plan. Real samples remain the foundation.
The third layer is regulatory. Clinical performance is part of the evidence that supports certification and registration of the product. In the European Union, the in vitro diagnostic regulation (IVDR) requires, for most risk classes, that a notified body assess this evidence before certifying the product. In Brazil, Anvisa's RDC 830/2023 defines the risk classification and the notification and registration regimes for in vitro diagnostic devices. The ISO 20916 standard brings together good practices for planning, conducting and reporting clinical performance studies, so that the data hold value before a regulator.
One kit, many populations
This point carries more weight when the product comes from abroad. A kit may have demonstrated clinical performance in its country of origin and behave differently in Brazil. The frequency of diseases is different, and with it the predictive values change. The infectious agents in circulation also vary: viral subtypes, genotypes and bacterial lineages predominant in one region may be rare in another, and a test designed from one variant may poorly recognise another. In molecular tests, a single mutation in the region where the primer or probe binds is enough for the kit to stop detecting a variant that circulates only in a given country. The result is a false negative that an assay run with reference material from another variant would not have shown. There are also diseases that share the same territory. Where dengue and Zika circulate together, serological tests must prove they distinguish one infection from the other, because antibodies against viruses of the same family can cross-react.
Placing on the market a kit without verified performance in the local population transfers to the lab report a risk the study should have found. A false negative leaves an infection untreated and without transmission control. A false positive leads to additional tests, unnecessary treatment and worry for the person receiving the result. For those who manufacture, import or distribute, the cost appears later, in technical complaints, loss of trust from laboratories and physicians and the need to redo steps that could have been resolved before launch.
Two answers for the same product
Analytical performance shows that the test measures well. Clinical performance shows that this measurement means something for the person being tested, in the population where the product will be used. One stage prepares the other, and the product is only ready when both answers are complete.
Projects that think about clinical performance from the design stage, with the clinical question, the population and the samples defined early, reach submission with fewer surprises and with evidence that holds up beyond the bench.
