Thermal runaway assessment in automotive battery development is still largely driven by isolated abuse tests, while design decisions require quantitative insight into how cell geometry, material thresholds, and thermal boundary conditions influence thermal runaway onset and severity. This paper presents a systematic sensitivity study using a coupled electrochemical and thermal model augmented with Arrhenius-based decomposition reactions to represent the dominant exothermic pathways. Thermal runaway onset is defined using a temperature rise-rate criterion to distinguish gradual heating from runaway acceleration. Two trigger modes are considered: an internal short circuit initiated by nail penetration and an external heating trigger. Four parameter groups are investigated: cell length scaling, separator decomposition temperature, external heating power, and the convective heat transfer coefficient to the environment. For the nail-triggered internal short circuit, larger cells exhibit lower peak temperatures but longer times to reach the maximum, indicating a geometry-driven shift from rapid escalation to a slower, more moderated evolution. In the external heating case, increasing cell size significantly delays onset, while peak temperature shows a nonlinear trend and approaches saturation rather than scaling inversely with size. Increasing the separator decomposition temperature also shows a saturation effect because alternative reactions can dominate the triggering sequence. External heating power exhibits a threshold: below a critical level, convective losses balance the input and prevent runaway. Even when external heating is stopped at an intermediate temperature, higher preheating power can still lead to higher peak temperatures due to a larger remaining reactive inventory when the internal short circuit occurs. Improved heat rejection consistently delays onset, reduces peak temperature, and accelerates cool-down. Overall, the study extends prior trigger-specific analyses by providing a unified reduced-order sensitivity view across two abuse pathways and by identifying threshold and saturation behaviors that translate directly into design-relevant robustness levers.