How results are checked
Numbers you can check.
Every statistical test, formula and score in PreclinIQ is compared with established reference software or with the published equations before a release. Here is what is checked and how.
t-tests, one- and two-way ANOVA, repeated measures with Greenhouse–Geisser correction, mixed-effects models, post-hoc tests, non-parametric tests, survival, regression, ROC, kappa and Bland–Altman are compared with SciPy, statsmodels, pingouin and lifelines.
41 checks against SciPy and statsmodels: four-parameter logistic fits and IC50, Z′, colony counts pooled over dilutions, time-kill activity and synergy, MIC50 and MIC90, plaque titres, and TCID50 by Spearman–Kärber and Reed–Muench.
Speciality scores are recalculated by an independent Python implementation: 3,600 values for gastroenterology and pancreas and 3,432 for oncology, with no differences. Cardiovascular risk equations are checked against the published worked examples.
The protocol planner's sample sizes matched statsmodels exactly in 16 scenarios (continuous, yes/no, crossover, non-inferiority and equivalence) and the Schoenfeld formula for time-to-event outcomes.
27 checks of Spearman and Pearson correlations, partial correlations, false-discovery-rate correction, Cronbach's alpha and questionnaire totals (reverse items and proration).
34 automated test suites run on every build, on both the readable and the protected version of the program, before a release goes out.
What this means, and what it does not
- Each result shows the test used and its assumption checks, so you, a co-author or a reviewer can follow it.
- You remain responsible for choosing appropriate methods and checking results before publication.
- The clinical module is for academic research with pseudonymised data. It is not validated for regulatory submissions (GCP / 21 CFR Part 11).
Found a number that looks wrong? Please tell us with the study file or the values, and we will check it.