Outcome imbalance in clinical trials: Mitigating loss of statistical power
Bibliographic record
Abstract
Abstract Background Unequal rates of decline between placebo and treatment arms reduce a trial’s power. We use machine learning to differentiate decliners from non‐decliners to mitigate the impact of outcome imbalance. Method We trained a model to classify decliners (i.e. individuals with higher CDR‐SB at 24 months of follow‐up) and non‐decliners on 1329 individuals with MCI or Alzheimer’s dementia from ADNI (adni.loni.usc.edu) and NACC (naccdata.org). Input features included baseline age, sex, APOE4 status, gray matter volumes of brain regions from MRI, MMSE, and CDR‐SB. We 1) simulated 100,000 trials by randomizing individuals to placebo and treatment (n = 250 per arm) and measured the proportions of decliners between the arms to assess the likelihood of imbalance, 2) measured the power to detect a 25% treatment effect across simulated trials with varying levels of imbalance (0‐5% more decliners in treatment, 1000 simulations per imbalance level), and 3) studied whether covariate adjustment and enrichment with predicted decliners alleviate imbalance‐related power losses. Result 22.4% of our simulated trials had inter‐arm imbalances of 5% or more, which translated into reduced power (‐15%) and effect sizes (‐0.11) compared to balanced trials (mean ± sem 89.2 ± 0.6% power, 0.39 ± 0.006 Cohen’s d for balanced trials; 73.8 ± 0.9% power, 0.28 ± 0.004 Cohen’s d for 5%‐imbalanced trials). Covariate adjustment on prognostic factors (e.g. APOE4, diagnosis, baseline CDR‐SB) increased power, but imbalance still reduced power by 9% (balanced: 97.2 ± 0.3%; 5%‐imbalanced: 87.8 ± 0.7%). Subgroups analyses of the predicted decliners (excluding predicted non‐decliners) increased power, despite smaller sample sizes (balanced: 97.4 ± 0.3%; 5%‐imbalanced: 92.2 ± 0.5%). Using enriched samples with 250 predicted decliners per arm obtained the greatest power and constrained the power loss to only 3% (balanced: 99.1 ± 0.1%; 5%‐imbalanced: 96.2 ± 0.4%) (Figure). Conclusion To improve a trial’s likelihood of success, covariate adjustment and enrichment with likely decliners should be used to mitigate power loss related to outcome imbalance.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.489 | 0.710 |
| Meta-epidemiology (narrow) | 0.003 | 0.002 |
| Meta-epidemiology (broad) | 0.004 | 0.005 |
| Bibliometrics | 0.005 | 0.004 |
| Science and technology studies | 0.001 | 0.009 |
| Scholarly communication | 0.005 | 0.006 |
| Open science | 0.005 | 0.007 |
| Research integrity | 0.005 | 0.007 |
| Insufficient payload (model declined to judge) | 0.005 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; the direct Gemma label and the distilled Codex classifier agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".