The recurrence reality after curative surgery. Lung adenocarcinoma is treated with surgical resection as the primary curative intent approach, yet up to 68 percent of patients develop recurrence after surgery. These recurrences are not a single event but can take fundamentally different forms: local recurrence within the same lung and chest region, distant metastasis to other organs, or both occurring simultaneously or sequentially. Each pathway carries distinct implications for prognosis and treatment selection.
The limitation of binary endpoints. Standard lung cancer clinical trials measure two endpoints: overall survival (the time from treatment to death) and disease-free survival (the time from treatment to any recurrence or death). These endpoints treat all recurrences as equivalent events and cannot distinguish between a patient who develops a curable local recurrence from one who immediately develops widespread metastatic disease. This oversimplification obscures the diverse natural histories that actually occur in lung adenocarcinoma.
The multi-state model as a solution. Multi-state models (MSMs) are statistical frameworks designed to analyze patient movement between defined disease states over time. Unlike the Cox proportional hazards model which evaluates the effect of risk factors on a single endpoint, MSMs simultaneously model multiple transitions between states, revealing how prognostic factors differentially influence each pathway of disease progression.
Study goal. This study applied MSM analysis to 335 lung adenocarcinoma patients followed for over 10 years post-surgery, defining five disease states and eight transition pathways to investigate how clinical and pathological factors specifically and differentially drive each type of recurrence progression and death.
Patient cohort and follow-up. Data were retrospectively collected from 340 patients who underwent complete surgical resection for invasive lung adenocarcinoma at Samsung Medical Center between October 2003 and August 2011. After excluding 5 patients with incomplete pathological analysis, the final cohort included 335 patients (186 men, 149 women) with a mean age of 61 years. Follow-up continued until death or June 2023, resulting in over 10 years of longitudinal data. Most patients (94.6 percent) underwent lobectomy or pneumonectomy.
Five states and eight transitions. The MSM defined one initial state (operation, OP), three intermediate states (local recurrence LR, distant metastasis Meta, and concurrent LR plus Meta denoted LR+Meta), and one terminal state (death). Eight transitions connect these states: four from OP (to LR, Meta, LR+Meta, or death), two from LR and Meta (each to LR+Meta), and two from Meta and LR+Meta (each to death). This structure allows the model to capture both primary transitions from surgery and secondary transitions after an initial recurrence event.
Proportional transition intensity model. Each transition was modeled using a proportional transition intensity function analogous to the Cox hazard function. This model estimates the regression coefficient for each covariate on each specific transition, allowing the researcher to determine which clinical and pathological variables influence, for example, the risk of distant metastasis independently from the risk of local recurrence.
Clinical and pathological variables analyzed. Primary clinical variables included age, sex, smoking history, adjuvant chemotherapy receipt, and operation type. Pathological variables included TNM stage, T status, lymph node (N) status, tumor size, tumor differentiation, predominant histologic pattern (lepidic, acinar, papillary, micropapillary, or solid), and the presence of pleural, vascular, lymphatic, or perineural invasion. All analyses used R version 4.3.1 with the mstate and survival packages.
Overall recurrence rates by pathway. Of 335 patients, 142 (42.4 percent) had no recurrence events and survived to the end of the study. Among those who experienced events, distant metastasis was the most common pathway with 142 patients (42.4 percent of all patients), followed by concurrent LR plus Meta in 54 patients (16.1 percent), and isolated local recurrence in only 7 patients (2.1 percent). All seven patients with isolated LR subsequently developed LR+Meta, making the LR-only state a transient condition rather than a stable endpoint.
Death outcomes by pathway. Of 149 patients (44.5 percent) who died by study end, 67 (45.0 percent of all deaths) died having experienced distant metastasis, 53 (35.6 percent) died from LR+Meta, and 29 (19.5 percent) died without any intermediate recurrence event. Almost all LR+Meta patients (53 of 54, representing 98.1 percent) died, making concurrent recurrence the highest-risk disease state identified in this study.
Timing of transitions after surgery. The median time from surgery to distant metastasis was 24.2 months. Concurrent LR+Meta occurred earlier at a median of 16.8 months. Isolated LR occurred fastest at a median of 10.2 months. Death in patients without intermediate events occurred at a median of 65.7 months, far later than in patients who developed any recurrence state. In the lobectomy subgroup, Meta occurred at a median of 2.05 years, LR+Meta at 1.38 years, and death at 6.28 years.
Secondary transition risk from metastasis. Among the 142 patients who developed distant metastasis as a primary event, 32 (22.5 percent) subsequently transitioned to LR+Meta, and 67 (47.2 percent) died without developing LR. Transition probability analysis showed that the risk of moving from Meta to LR+Meta peaks at 2 to 4 years after surgery: at 2 years the transition probability to LR+Meta was 0.23 versus 0.36 to death, but by 6 years the LR+Meta probability dropped to 0.06 while death probability rose to 0.73.
Lymph node status: the most consistent risk factor. Pathological lymph node status (pN) was a significant risk factor for three separate transitions. After surgery, positive pN independently increased the risk of developing distant metastasis (transition OP to Meta, p=0.001) and concurrent LR plus Meta (transition OP to LR+Meta, p=0.001). Even after metastasis had already occurred, pN remained a risk factor for subsequent progression to LR+Meta (transition Meta to LR+Meta, p=0.005). This persistent effect of lymph node involvement across multiple disease stages suggests that pN reflects systemic tumor biology rather than only local disease burden.
T status drives early metastasis risk. Pathological T status (pT) was a significant risk factor specifically for the transition from surgery to distant metastasis (OP to Meta, p=0.024), but was not significant for transitions to local recurrence or death. This specificity indicates that primary tumor size and local invasion independently predict systemic spread risk, consistent with the biological principle that larger primary tumors have had more time and opportunity to seed distant sites before resection.
Age affects death but not recurrence type. Age was a significant risk factor for direct mortality without recurrence (OP to death, p less than 0.001) and for death after metastasis (Meta to death, p less than 0.001), but was not associated with the type or timing of recurrence itself. This finding suggests that age-related mortality in lung adenocarcinoma primarily reflects competing comorbidities and physiologic decline rather than more aggressive cancer biology, as tumor aggressiveness driving recurrence type appeared to be independent of patient age.
Operation type and sex effects. Sublobar resection was associated with significantly higher local recurrence risk compared to lobectomy or pneumonectomy (OP to LR, p=0.015), providing statistical support for established surgical guidelines favoring lobectomy as the standard of care. Female sex was protective against progression from isolated metastasis to concurrent LR+Meta (Meta to LR+Meta, p less than 0.001), suggesting a sex-related difference in metastatic evolution that warrants further biological investigation.
The pathological invasion analysis. An independent subanalysis was conducted in 307 patients with complete pathological data, adding detailed histologic variables including tumor differentiation, predominant histologic pattern, pleural invasion, and the presence of any vascular, lymphatic, or perineural invasion. The LR to LR+Meta transition was excluded from this analysis because all five LR patients in this subset also developed LR+Meta, making regression modeling for that transition impossible.
Vascular, lymphatic, and perineural invasion as a composite predictor. The presence of any vascular, lymphatic, or perineural invasion was a significant independent risk factor for the transition from surgery to distant metastasis (OP to Meta, p less than 0.001) and from surgery to concurrent LR+Meta (OP to LR+Meta, p=0.014). This composite invasion variable was the most powerful pathological predictor of systemic spread in the model, consistent with the biological mechanism by which tumor cells use vascular and lymphatic channels to disseminate to distant sites.
Why pleural invasion and histologic subtype were not significant. Factors including tumor differentiation, predominant histologic pattern (lepidic, acinar, papillary versus solid and micropapillary), and pleural invasion were not statistically significant predictors of any transition in the pathological analysis. The authors propose that this may reflect appropriate incorporation of these factors into surgical and adjuvant treatment decisions at the time of initial care: patients with high-risk histologic features likely received more aggressive treatment, which may have attenuated the prognostic signal in observational data.
MSM's advantage over standard survival modeling. The MSM enabled separation of prognostic factors that a conventional single-endpoint Cox model would conflate. For example, age specifically predicted mortality transitions but not recurrence type, while invasion markers specifically predicted metastasis onset but not subsequent mortality. This granular separation is clinically actionable: it allows treatment decisions to be targeted based on the specific risk profile of each individual patient across the disease trajectory.
A roadmap for post-surgical surveillance. The MSM findings provide a data-driven framework for surveillance scheduling. The peak risk of developing LR+Meta from an existing metastatic state occurs in the 2 to 4 year post-surgery window, making this the highest-yield period for intensive local disease surveillance in metastatic patients. Earlier transition timing for LR+Meta (median 16.8 months) compared to isolated metastasis (median 24.2 months) suggests the most aggressive disease trajectories manifest within the first two years.
Targeted intervention based on recurrence type. Because risk factors differ across pathways, clinical interventions can be better matched to patient profiles. Patients with positive lymph nodes at surgery are at risk not only for initial metastasis but for complex concurrent recurrence patterns, suggesting they may benefit from more intensive systemic therapy. Patients undergoing sublobar resection have elevated local recurrence risk, supporting the clinical recommendation to reserve this approach for confirmed stage I disease with tumors 2 cm or smaller.
Limitations of the study. The retrospective, single-institution design limits generalizability, and findings require validation in larger multicenter cohorts. The patient cohort predates the widespread availability of tyrosine kinase inhibitors and immunotherapy agents that are now standard in lung adenocarcinoma treatment, meaning the transition patterns observed may differ in contemporary patients. Imaging surveillance was not performed at standardized intervals, potentially introducing lead-time bias. Circulating tumor DNA data were not available to correlate molecular recurrence signals with clinical disease states.
Broader application of MSM in oncology. This study demonstrates that multi-state modeling can provide clinically superior insight compared to conventional survival analysis by separating the effects of prognostic factors across distinct disease states. Applying this methodology prospectively in a larger cohort that includes molecular biomarkers, circulating tumor DNA, and current treatment options would substantially extend its clinical utility and enable dynamic risk stratification throughout the post-surgical disease course.