Bistro Research
Where cancer trials are vs where cancer is › E:trial-burden-mismatch-1
Null

Cancer trials per cancer case: a 68-fold income gap, but the 50-fold test is inconclusive

1 Nov 2026 · first questions

Provisional: no maintainer has signed this finding off yet
Question

How steep is the income gradient?

Hypothesis · pre-registered

>50-fold high- vs low-income ratio

Test

Trials were counted per country with ≥1 site and divided by GLOBOCAN 2022 incident cases, summed within each income group. The resulting HIC/LIC ratio was compared with 50, using a 2,000-replicate bootstrap that resamples countries within groups.

Finding

Using registered ClinicalTrials.gov interventional cancer trials that started in 2015–2024, high-income countries had 885 trials per 100k incident cancer cases (GLOBOCAN 2022) and low-income countries had 13.0. The HIC:LIC ratio is 68.0-fold, but the country-bootstrap 95% CI is 33.9–170.8. The interval contains 50, so the pre-registered rule returns inconclusive: the gradient is steep, but this test does not show that it exceeds 50-fold. 7 of 9 sensitivity ratios were above 50. The two that count each multinational trial once fell below it: single-country trials only, 38.1, and one count per income group, 36.0. All four automated verification checks returned FAIL, although the re-run values printed in the check output match the registered values, so these numbers should not be treated as independently verified.

MetricValue95% CIn
HIC/LIC ratio, trials per 100k incident cases68.011 fold33.895 – 170.83180
HIC trials per 100k incident cases885.421 trials/100k cases—53
UMIC trials per 100k incident cases239.115 trials/100k cases—47
LMIC trials per 100k incident cases53.683 trials/100k cases—53
LIC trials per 100k incident cases13.019 trials/100k cases—27
LIC countries with zero trials15 countries—27
Sensitivity ratios above 507 count—9
HIC/LIC ratio, unique trials per income group36.028 fold——
HIC/LIC ratio, deaths denominator114.861 fold——

Next: Repeat the analysis on WHO ICTRP (multi-registry) data with deduplication. Pre-register unique-trial-per-group counting as the primary metric, and pool LIC with LMIC or use a Poisson model to narrow the interval.

Caveat · open

The LIC numerator is tiny: 64 trials, 15 of 27 countries with none. The size of the ratio depends on how multinational trials are counted. ClinicalTrials.gov under-covers non-US registries, and LIC burden estimates are modelled. All four automated verification checks returned FAIL.

Claims
  • scripts/trial-burden-mismatch-1/analysis.py

    Interventional cancer trials started 2015-2024 on ClinicalTrials.gov per 100k GLOBOCAN-2022 incident cases (trial counted once per country with a site): HIC 885.4, UMIC 239.1, LMIC 53.7, LIC 13.0; HIC/LIC ratio 68.0.

    Passed· 53.3s · 15 h ago
  • scripts/trial-burden-mismatch-1/analysis.py

    Country-level bootstrap 95% CI for the HIC/LIC ratio is 33.9 to 170.8 (straddles 50), so the >50-fold hypothesis is inconclusive; leave-one-LIC-out ratios range 59.5 to 93.9.

    Passed· 52.9s · 15 h ago
  • scripts/trial-burden-mismatch-1/analysis.py

    HIC/LIC ratio under alternative definitions: deaths denominator 114.9; World Bank current income 60.9; single-country trials 38.1; treatment-purpose only 104.6; phase 3 13 LIC trials, ratio 90.6; start 2020-24 59.2; unique trials per group 36.0; 7 of 9 sensitivities exceed 50.

    Passed· 52.6s · 15 h ago
  • scripts/trial-burden-mismatch-1/analysis.py

    Site-specific HIC/LIC ratios (keyword-matched conditions): breast 82.5 (7 LIC trials), cervix 53.5 (17), liver 79.0 (3), stomach infinite (0 LIC trials). Trials at Taiwan (1660) and Hong Kong (574) have no GLOBOCAN denominator and are excluded.

    Passed· 52.2s · 15 h ago
Literature