A glossary for interpreting BVCT results

Reading a BVCT result means knowing its comparator, endpoint and effect measure. BVCT reports a hazard ratio for time-to-event endpoints, or an odds ratio or Relative Ratio for binary endpoints, against a named standard of care, to one decimal place. The terms below define each measure and how the track record is counted.

Last updated 25 September 2026 · Maintained by BioinvestGPT ApS

Standard of care (SoC): The comparator selected for the research question: the established therapy for the disease, population, line of therapy, geography and period concerned. Name the actual comparator rather than relying on the abbreviation alone.
Hazard ratio (HR): A comparison of event hazards over time between two arms. For a harmful event such as death, a value below 1 generally favors the intervention; the endpoint definition remains essential.
Odds ratio (OR): The odds of an outcome in one arm divided by the odds in the other. It is not generally numerically equal to a Relative Ratio.
Relative Ratio (RR): BVCT's term for the proportion of participants with a defined binary outcome in the intervention arm divided by that proportion in the comparator arm over a defined period; numerically, the risk ratio (relative risk). Whether a value above or below 1 favors the intervention depends on whether the event is beneficial, such as response, or harmful, such as relapse.
0.1 precision: BVCT reports effect sizes to one decimal place (for example, an illustrative HR 0.6) rather than as qualitative labels. This is display resolution; measured prediction error is a separate quantity, discussed on the methodology page.
Overall survival (OS): A time-to-event endpoint concerning death from any cause, with the starting point and analysis defined by the study.
Progression-free survival (PFS): A time-to-event endpoint concerning progression or death under the trial's definitions. An improvement in PFS should not be assumed to establish an improvement in OS.
Prospective (ex-ante) prediction: A prediction made before the relevant outcome becomes known. Assess its timing against the first relevant public disclosure.
Scored readout: A reported trial result against which a BVCT prediction was adjudicated. Some predictions have more than one scored readout; BVCT's published track record counts each scored readout.
Not adjudicable: A readout that could not be scored as correct or incorrect under the published rule. It is listed on the dashboard but excluded from the accuracy, PPV and NPV denominators.
Accuracy, PPV and NPV: Classification measures with different denominators: correct calls among all adjudicated calls, correct calls among positive calls, and correct calls among negative calls. The evaluation page defines each one.
Patient-data-free: BVCT does not require patient records as input. It uses public scientific knowledge, such as knowledge graphs, literature and trial registries, which is distinct from private patient data.
Virtual clinical trial: In BVCT's product context, a computational comparison of modeled drug effects. Elsewhere the term can also mean remotely conducted human studies; identify which meaning is intended.