KEY POINTS
- The study combined A549 NSCLC cell-survival data with clinical data from 34 early-stage NSCLC studies comprising 41 three-year local-control datapoints.
- Three radiobiological models were tested: saturable repair and repairability, universal survival curve, and linear-quadratic models. Unlike conventional one-dimensional fitting, the new maximum-likelihood approach explicitly modeled uncertainty on both the dose and local-control axes.
- Estimated TCD50 values were 57.1 Gy EQD2 with SRR, 51.6 Gy with USC, and 50.2 Gy with LQ. Corresponding TCD95 values were 81.4, 71.8, and 60.5 Gy EQD2, showing substantial dependence on radiobiological formalism.
- Effective dose-axis uncertainty was 16.6%, 16.9%, and 10.5% for the three models, compared with mean local-control-axis uncertainties of only 5.0%, 5.0%, and 5.8%. Ignoring dose uncertainty can therefore artificially flatten fitted dose-response curves.
- Clonogenic cell-number heterogeneity explained only 3–15% of modeled variance in TCP steepness. Inter-patient radiosensitivity accounted for the remaining 85–97%.
- The LQ model produced the lowest cell-survival fitting AIC but only by estimating an unusually high α/β of 31.1 Gy, illustrating how a good statistical fit can conceal biologically questionable parameter values.
- The framework depended on a single A549 cell-survival dataset and literature-derived clinical cohorts. Assumptions regarding repopulation, EQD2 conversion, tumour volume, and independence between biological variables remain model dependent.
CLINICAL TAKEAWAY
The paper argues that uncertainty in dose and differences in individual tumour radiosensitivity matter more to population TCP curves than variation in clonogen number. This is relevant for future biologically individualized prescription models, but it does not establish new clinical dose thresholds for NSCLC.