Observed CTIT onset timing is associated with platelet surveillance schedules in a real-world cohort

Abstract

Chemotherapy-induced thrombocytopenia (CTIT) is detected through platelet testing, so routine-care estimates of onset and recovery may depend on surveillance schedules. We characterized within-cycle timing and assessed whether apparent differences among drug categories persisted after accounting for measurement-time uncertainty, repeated cycles, and measured confounding. This retrospective single-center study used 538 CTIT cycles from 368 patients (June 2015–January 2025), representing 18.8% of 2,860 screened cycles. Eligible cycles had a platelet nadir < 75 × 10⁹/L and documented decline and recovery, yielding a restricted cohort. Primary analyses used interval-censored Weibull AFT models for onset and recovery, adjusted for prespecified patient, treatment, and tumor covariates with patient-clustered robust covariance. Sensitivity analyses included observed-time Kruskal-Wallis tests, patient-random-intercept mixed models, restriction to detection intervals ≤ 7 days, first-cycle-only analyses, and recovery to ≥ 100 × 10⁹/L. The cohort contained 3,462 platelet measurements (median 6 per cycle; median interval 3 days). In the primary interval-censored analysis, the drug-category term was nonsignificant for onset ( P = 0.671) or recovery ( P = 0.984); dense-sampling results were compatible ( P = 0.343 and 0.493). Median observed onset was 11 days (IQR 7–20), recovery from the first low count 9 days (IQR 6–15), and recovery from nadir 7 days (IQR 4–12). Unadjusted observed onset differed by drug category ( P = 1.72 × 10⁻⁹; medians 8 days for gemcitabine and 17 days for oxaliplatin and T-DM1), but recovery did not ( P = 0.871). Drug categories differed in first post-treatment check timing and onset detection-interval width ( P = 7.50 × 10⁻²³ and P = 4.46 × 10⁻¹³). Recovery was longer with deeper nadir (7.5, 10, and 12 days; P = 5.07 × 10⁻⁷). Apparent drug-category differences in observed onset were not retained after accounting for measurement-time uncertainty and were closely aligned with platelet surveillance patterns. These findings are compatible with surveillance-related bias but do not exclude treatment-related biological differences. Recovery was associated with nadir severity, and no drug-category difference was detected in the primary recovery analysis. Complete-trajectory selection limits generalizability. Prospective multicenter studies with protocolized sampling and complete follow-up are needed before regimen-specific timing can be inferred.