Background
Gastrointestinal (GI) cancer outcomes in low- and middle-income countries (LMICs) remain poorly characterized due to data limitations and inadequate analytic frameworks. Multistate models (MSMs) can capture the dynamic nature of treatment and outcome trajectories but are rarely implemented in LMIC datasets. Understanding transition-specific risks across key treatment states is critical to optimizing cancer care in resource-constrained settings.
Objectives
We aimed to demonstrate the feasibility of implementing multistate survival models using routine data from the African Research Group for Oncology (ARGO) and to estimate transition-specific effects of key covariates on GI cancer patient pathways in a Nigerian cancer cohort. Compared to standard single-outcome survival models, the multistate approach enables clinical interpretation of care stages: e.g., separating delays in surgical uptake from post-treatment mortality, which would otherwise be aggregated.
Methods
We analyzed longitudinal data from 241 patients with GI cancers enrolled in the ARGO cohort between 2009 and 2025. Data were converted into long format (547 rows) to enable transition-specific modelling. The multistate framework defined four transitions: (1→2) diagnosis to surgery, (1→4) diagnosis to right-censoring, (2→3) surgery to death, and (2→4) surgery to right-censoring. Transition 1→3 (diagnosis to death without surgery) was excluded due to sparse or missing data. A flexible parametric Weibull model was implemented using merlin in Stata. Covariates included age, sex, and chemotherapy. This work was co-led by early-career Nigerian researchers at the Obafemi Awolowo University, with mentorship and joint supervision from collaborators at Memorial Sloan Kettering Cancer Center (MSKCC).
Results
Chemotherapy was associated with a lower hazard of transition across states (β = –0.267, p = 0.030). Transition 1→2 (diagnosis to surgery) showed a significantly elevated baseline hazard (β = 1.619, p < 0.001), while transition 1→4 (diagnosis to censoring) had a lower hazard (β = –2.126, p < 0.001). The hazard for surgery-to-death (2→3) was also elevated (β = 0.607, p = 0.031), suggesting the importance of post-surgical monitoring. Age and sex were not statistically significant predictors across transitions.
Conclusions/Implications
This multistate survival analysis provides more granular insight into the GI cancer care continuum, distinguishing survival differences by care stage and enabling targeted interventions (e.g., improving time to surgery). Unlike traditional survival analysis, this method reveals which transitions are bottlenecks versus inflection points. Transition-specific risks offer nuanced insights into patient trajectories, highlighting critical periods for intervention such as peri-surgical care and chemotherapy adherence. Future analyses should incorporate time-varying covariates and explore recurrence-related transitions to refine outcome prediction and guide resource allocation.
 

Survival probabilities by transition state in GI cancer care among a Nigerian cohort