Predictive Value Explorer – what a test result really means
How well a diagnostic test performs is described by its sensitivity and specificity, and these stay the same wherever the test is used. How much a single result can be trusted is a different matter: the positive and negative predictive values (PPV and NPV) depend on how common the condition is in the tested population. The same test can be highly informative in symptomatic patients and produce mostly false positives in population screening. Our interactive Predictive Value Explorer makes this relationship visible and easy to explore.
Why predictive values matter
Clinicians and patients rarely ask how sensitive a test is. They ask what a result means: if the test is positive, how likely is the disease? If it is negative, can the disease be ruled out? PPV and NPV answer these questions. Unlike sensitivity and specificity, they change with prevalence, the proportion of people in the tested population who actually have the condition. The rarer the condition, the more false positives a test produces relative to true positives, even when its performance stays the same.
What the tool shows
Enter a test’s sensitivity and specificity, set the prevalence (pre-test probability), and the Explorer calculates:
- PPV and NPV for the chosen setting
- Likelihood ratios (LR+ and LR−), which describe how much a result shifts the probability of disease, independent of prevalence
- Expected results per 10,000 people tested, split into true and false positives and negatives
- PPV and NPV across the full prevalence range, on a log or linear scale, with the current setting marked
- A Fagan nomogram, a graphical form of Bayes’ theorem that converts pre-test into post-test probability with a straight line
The prevalence slider uses a logit scale, so low prevalences typical of screening get enough resolution. Presets for screening (1 %), symptomatic (10 %) and case–control (50 %) settings let you compare typical scenarios with one click.
An example: one test, three settings
Take a test with 90 % sensitivity and 90 % specificity (LR+ 9, LR− 0.11):
| Setting | Prevalence | True positives | False positives | PPV | NPV |
|---|---|---|---|---|---|
| Screening | 1 % | 90 | 990 | 8.3 % | 99.9 % |
| Symptomatic patients | 10 % | 900 | 900 | 50 % | 98.8 % |
| Case–control study | 50 % | 4,500 | 500 | 90 % | 90 % |
Per 10,000 people tested.
In screening, more than 9 out of 10 positive results are false, while a negative result almost certainly rules out the disease. In a case–control study with equal numbers of cases and controls, the same test looks far better. This is why diagnostic accuracy from case–control studies often cannot be transferred directly to clinical practice.
How to read likelihood ratios
As a rule of thumb, an LR+ above 10 or an LR− below 0.1 usually changes clinical decisions. Likelihood ratios between 0.5 and 2 barely move the probability of disease. The Fagan nomogram shows this directly: a strong test draws a steep line from pre-test to post-test probability, while a weak test draws an almost flat one.
Relevance for biomarker and IVD development
Predictive values are a central part of diagnostic test development. Under the In Vitro Diagnostic Regulation (IVDR, EU 2017/746), PPV, NPV and likelihood ratios are among the clinical performance characteristics a manufacturer has to establish, where applicable, for the intended purpose of a device. The expected prevalence in the target population therefore influences how a test is positioned: as a rule-in test, a rule-out test, or a tool for risk stratification.
At TAmiRNA, we consider these questions from the start of every biomarker program, from choosing the intended use population to designing clinical performance studies. The Explorer is useful for:
- planning biomarker validation and clinical performance studies
- interpreting diagnostic accuracy data from publications
- defining the intended purpose and target population of a new IVD test
- teaching and training in diagnostic statistics
Limitations
The calculations assume that sensitivity and specificity stay the same across settings. In practice, spectrum effects can change them: a test validated in patients with advanced disease may detect early or mild cases less reliably. Predictive values should therefore always be interpreted together with the characteristics of the study population in which the test was evaluated.
