Conditional survival analysis of patients with resected non–small cell lung cancer

Background Conditional survival (CS) analyses provide an estimate of survival accounting for years already survived after treatment. We aim to evaluate the difference between actuarial and conditional survival in patients following lung resection for non–small cell lung cancer (NSCLC). In addition, CS analyses are used to examine whether prognosticators of survival change over time following surgery. Methods Patients who underwent anatomic lung resection at a single institution for pathologic stage I-IIIA NSCLC between 2010 and 2021 were identified; those who underwent wedge resection for node-negative tumors ≤2 cm were also included. CS estimates were calculated as the probability of remaining disease-free after x years of nonrecurrence (CSx). Kaplan–Meier, log-rank, and Cox proportional hazard methods for examining CS were used for subgroup comparisons and assessing associations with baseline covariates. Results Overall, 863 patients met the study inclusion criteria, with a median follow-up of 44.1 months. Conditional overall survival (OS) and disease-free survival (DFS) were greater than actuarial rates at all time points after surgery. At the time of resection, male sex (hazard ratio [HR], 1.33; 95% confidence interval [CI], 1.03 to 1.72; P = .032), tumor size >3 cm (HR, 1.17; 95% CI, 1.11-1.23; P < .001), node positivity (HR, 3.31; 95% CI, 2.52-4.33; P < .001), and American Joint Committee on Cancer stage (P < .001) were associated with DFS. However, if a patient lived 3 years without recurrence (CS3), these factors were no longer prognostic of DFS. Conclusions Conditional survival analyses provide dynamic assessments of OS and DFS after NSCLC resection. After 3 years without recurrence, certain characteristics associated with DFS at the time of surgery no longer prognosticate recurrence.

Disease free survival and 3-year conditional disease free survival of patients with resected non-small cell lung cancer.CS3, Conditional 3-year survival.

CENTRAL MESSAGE
As time progresses without recurrence, conditional survival analyses may better prognosticate survival than actuarial methods in patients with resected non-small cell lung cancer.

PERSPECTIVE
Patients who continue to survive after cancer treatment experience dynamic changes to their risk for death and recurrence.Conditional survival analyses may allow for more accurate prognostication of these risks than actuarial estimates.The current data suggest an association between patient characteristics and overall and disease-free survival changes as patients live longer after treatment.
5][6] Prognosis varies across subtypes and staging, with better survival in patients with early-stage non-small cell lung cancer (NSCLC) than in those with more advanced stages.Treatment is primarily based on staging at the time of diagnosis with prognostication data available from that initial point in time.However, as patients survive beyond this, with dynamic changes in health, it becomes increasingly difficult to prognosticate overall survival (OS) and disease-free survival (DFS). 7,8][10] Importantly, CS analyses account for years already survived after diagnosis, which can alter predictions of future survival. 11By understanding the CS of patients with lung cancer over time, providers may be able to individualize surveillance schedules and provide more accurate long-term prognostication. 112][13][14][15][16][17][18][19] In this study, we aimed to investigate the difference between actuarial and CS in patients following lung resection for NSCLC.In addition, we used CS analyses to examine whether prognosticators of survival change over time.

Patients who underwent lung resection at Rush University Medical
Center between 2010 and 2021 were identified from an institutional database.Inclusion criteria included patients with stage I-IIIA pathologically confirmed primary lung adenocarcinoma or squamous cell carcinoma who underwent anatomic lung resection with an R0 resection.Patients who underwent a wedge resection for a pathologic tumor 2 cm with N0 status were also included; the results of the CALGB14053 trial prompted this inclusion. 20Exclusion criteria included age <18 years, M1 disease, induction therapy, resection of multiple tumors, other nonlung primary malignancy, and <6 months of clinical follow-up from the date of surgery.The Institutional Review Board or an equivalent Ethics Committee of Rush University Medical Center approved the study protocol and publication of data (ORA 18070602, August 14, 2018, and ORA 18021402, June 14, 2018).The requirement for patient written consent for publication of the study data was waived owing to the study's retrospective nature.
Clinical data were collected on sex, race, ethnicity, Eastern Cooperative Oncology Group Performance Status, smoking status (current, ever, never), diabetes, coronary artery disease (CAD), congestive heart failure (CHF), histologic subtype, tumor size (largest measurement of invasive component; categorized as 3 cm vs >3 cm), node status, and American Joint Committee on Cancer (AJCC) eighth edition stage.Operative data included surgical procedure (wedge resection, segmentectomy, lobectomy, bilobectomy, or pneumonectomy), and surgical procedure date.
Postoperative data collected included adjuvant therapy, recurrence, date of recurrence, last known date alive, and date of death.OS was defined as time from surgery to the date of death or the last known date alive, and DFS was defined as time from surgery to first recurrence.

STATISTICAL ANALYSIS
Conditional survival, CS x (t), was calculated as the probability of survival, where x is the time (in years) that the patient has already survived and t is the additional time beyond x that they may survive.Kaplan-Meier, log-rank, and Cox proportional hazard methods for CS were used for subgroup comparisons and associations with baseline covariates. 21Hazard ratios (HRs) were calculated to express the prognostic power of each covariate.Variables considered included age, sex, race, ethnicity, Eastern Cooperative Oncology Group Performance Status, smoking status, diabetes, CAD, CHF, histologic subtype, tumor size, node status, and AJCC eighth edition stage.

Actuarial Survival Versus CS
The median follow-up was 44.1 months (IQR, 26.7-62 months) for OS and 38.2 months (IQR, 20.9-55.9months) for DFS.After resection, actuarial OS was 99% at 1 year, 96% at 2 years, 92% at 3 years, and 83% at 5 years, and actuarial DFS was 92% at 1 year, 84% at 2 years, 78% at 3 years, and 69% at 5 years.A total of 132 patients, or 58% of our cohort, experienced a recurrence event within the first 2 years after surgery; 96 patients, or 42% of our cohort, experienced a recurrence event after the initial 2 years (Table E1).To fully evaluate the changes of survival estimates that conditional analyses would provide, we calculated the likelihood of surviving for up to 5 years after surgery; these were termed 5Y-CS 0 consecutively through 1Y-CS 4 .For example, 2Y-CS 3 corresponds to the probability of 3-year survival after already surviving 2 years.As patients survived longer after surgery, both their OS and DFS increased (Figures E1 and E2).
Conditional OS after 3 years of survival (CS 3 OS) and conditional DFS after 3 years of nonrecurrence (CS 3 DFS) were greater than their actuarial equivalents across all time points (Figure 1, A and B) Despite similar rates of change for both actuarial survival and CS, their absolute values for both CS 3 DFS and OS were greater than their actuarial DFS and OS counterparts.For example, the CS 3 OS at 2 years was 90%, whereas the actuarial OS at 5 years was 83%, and the CS 3 DFS at 2 years was 89%, whereas the actuarial DFS at 5 years was 69%.Likewise, the CS 3 OS at 5 years was 79%, whereas the actuarial OS at 8 years was 72% and the CS 3 DFS at 5 years was 79% while the actuarial DFS at 8 years was 61% (Figure 2, A and B) The figures report the y-axis as 3-year survival; for example, at 24 months in the figure, the 3-year survival correlates with the 5-year survival, as discussed above.
In certain subgroups, patients with high-risk features at diagnosis had better CS estimates than the actuarial survival of patients without those features.For example, nodepositive patients had a 2-year CS 3 DFS of 82%,whereas node-negative patients had a 5-year actuarial DFS of 76%.Similarly, patients with a tumor size >3 cm had a 2-year CS 3 DFS of 90%, whereas patients with a tumor size 3 cm had a 5-year actuarial DFS of 72%.This trend was observed across most time points following surgery (Figures 3 and 4). Figure 5 provides a graphical abstract of the study.

DISCUSSION
Traditional actuarial survival estimates take into account both early and late deaths, resulting in an overly pessimistic prognosis for patients who survive beyond the early postoperative period. 14,18The ability to prognosticate recurrence and survival from the current point in time rather than from estimates at the time of diagnosis can be helpful for both clinicians and patients.CS analyses provide information about how prognosis changes over time, as opposed to the static predictions provided by traditional survival estimates. 11,22Importantly, CS estimates allow for future survival estimates to be calculated Comparison of the actuarial disease-free survival of patients with a tumor 0 to 3 cm versus the 3-year conditional disease-free survival of patients with a tumor >3 cm as time after surgery progresses.DFS, Disease-free survival; CS, conditional survival.
Conditional analyses better estimate long term disease-free survival as patients survive beyond diagnosis.Node positivity and tumor size are no longer risk factors for recurrence after 3 years of disease-free survival.

Tumor Size
Actuarial DFS -HR 1.17; 95% Cl, 1.  cumulative impact of events and the evolving nature of survival probabilities.
To our knowledge, this is the first North American study to compare actuarial and conditional OS and recurrence rates for patients who underwent curative intent lung resection for NSCLC.Our current data demonstrate that CS 3 OS and CS 3 DFS were greater than the actuarial rates at all time points after surgery.Patients with characteristics associated with mortality and recurrence at the time of resection exhibited the greatest differences between actuarial survival and CS.In addition, although male sex, tumor size >3 cm, node positivity, and AJCC stage were associated with DFS at the time of resection, after 3 years without recurrence, these factors no longer prognosticated DFS.
There is a relative paucity of CS analyses examining NSCLC in the literature.The current results are supported by a smaller, international study that found no association between sex, T stage, node positivity, or AJCC stage with recurrence after 3 years. 8However, this study did not include squamous cell carcinomas, was performed using the AJCC seventh edition staging system, and did not provide data on types of lung resections included.Similar to the current study, investigators examining other malignancy types have reported that high-risk features at the time of treatment are associated with greater differences between CS and actuarial survival estimates.For example, the calculated CS exceeded the actuarial survival for patients with resected cholangiocarcinoma and high-risk features, including larger tumor size and lymph node metastasis. 117][28] Accordingly, the cohort of patients that remain alive and disease free through the early postoperative years, when recurrence is most common, have a different prognosis than seen the initial cohort at the time of treatment.This is reflected in the difference between actuarial survival and CS estimates.CS analyses account for time survived after surgery and ultimately provide more accurate estimates of prognosis for patients from that time point onward.
The differential between actuarial and conditional OS was smaller than that between actuarial and conditional DFS.There may be several explanations for this observation.Despite not experiencing recurrence, patients with larger tumors or greater AJCC stage carried a higher comorbidity burden, including diabetes, CAD, and CHF.These comorbidities have the potential to affect OS without changing the risk of DFS.
As a patient accumulates years of survival, the relevance of certain prognostic factors present at the time of surgery changes.CS analysis is a powerful tool for estimating how prognosis changes in patients over time.These findings have important implications for clinical decision making and patient counseling.They may allow for individualized surveillance schedules and help alleviate patient anxiety as they survive for longer periods after surgery.
This study has several limitations.Its single institution retrospective nature decreases the external validity of our findings.Despite the large number of subjects and longterm follow-up, survival estimates beyond 5 years became limited owing to a smaller total study population.Validation within both North American and international populations would allow for greater generalization of our results.

CONCLUSIONS
In patients with resected stage I-IIIA NSCLC, CS estimates remain greater than actuarial survival rates for both OS and DFS.Patients with characteristics most strongly associated with mortality and recurrence demonstrate the greatest difference between CS and actuarial survival.In addition, sex, tumor size, node positivity, and AJCC stage may have a diminishing prognostic effect as time passes.These findings suggest that CS analyses provide a quantitative representation of changes in patient prognosis after treatment that may be valuable to both patients and providers.

FIGURE 5 .
FIGURE 5. Graphical abstract representing key findings of this study.
AJCC, American Joint Committee on Cancer.P

TABLE 3 .
Difference between conditional survival and actuarial disease-free survival

after surgery, mo Disease-Free Survival Difference 3 Year Survival, %
FIGURE 3. Comparison of the actuarial disease-free survival of node-negative patients versus the 3-year conditional disease-free survival of node-positive patients as time after surgery progresses.DFS, Disease-free survival; CS, conditional survival.

TABLE E1 .
Recurrences by year after surgery