Imaging Surveillance for Surgically Resected Stage I Non-Small Cell Lung Cancer: Is More Always Better?

Abstract

Objective:

Routine surveillance imaging for patients with resected non-small cell lung cancer is standard for the detection of disease recurrence and new primary lung cancers. However, surveillance intensity varies widely in practice, and its impact on long-term outcomes is poorly understood. We hypothesized that surveillance intensity was not associated with 5-year overall survival in patients with resected stage I non-small cell lung cancer. Additionally, we examined patterns of recurrence and new primary lung cancer development.

Methods:

Cancer registrars at Commission on Cancer accredited institutions re-abstracted records to augment National Cancer Database patient data with information on comorbidities, imaging surveillance including intent and result of imaging, and recurrence (2007–2012). Pathologic stage I non-small cell lung cancer patients undergoing computed-tomography surveillance were placed into three imaging surveillance groups based on clinical practice guidelines: high intensity (3 month), moderate intensity (6 month), and low intensity (annual). Kaplan Meier analysis and Cox regression were used to compare overall survival among the three surveillance groups.

Results:

2442 patients were identified, with 805 (33%), 1216 (50%), and 421 (17%) patients in the high, moderate, and low surveillance intensity groups, respectively. Five-year overall survival was similar between intensity groups (p=0.547). Surveillance on asymptomatic patients detected 210 (63%) cases of locoregional recurrences and 128 (72%) cases of new primary lung cancer.

Conclusions:

In a unique national dataset of long-term outcomes for stage I non-small cell lung cancer, surveillance intensity was not associated with 5-year overall survival.

Introduction

Surveillance imaging after lung resection remains a critical component of oncologic care for patients with stage I non-small cell lung cancer (NSCLC)^{1, 2}. Survival following resection is dependent on multiple patient and tumor-related factors^2–4. Specifically, risk of lung cancer recurrence or development of a new primary lung cancer (NPLC) can influence long-term survival ^{3, 5}. Surveillance imaging for the early detection of recurrence or screening for NPLC remains an important component of survivorship care. For patients whose recurrence or NPLC is detected early, there is a possibility that additional curative intent therapy can be performed⁵.

Recommendations regarding surveillance imaging intensity after NSCLC resection vary widely in clinical practice. Existing current guidelines recommend visits as frequent as every 3 months during the first two years, while others recommend annual visits^8–10. The impact of surveillance imaging on long-term survival is poorly understood. The lack of consensus is due to limited quality longitudinal follow-up data on post-resection patients^{5, 6}.

We performed a retrospective cohort study using supplemented data from the National Cancer Database (NCDB) to compare imaging surveillance in patients with pathologic stage I NSCLC. Our primary aim was to determine if the intensity of surveillance with computed-tomography (CT) was associated with 5-year overall survival (OS). We hypothesized that surveillance intensity was not associated with OS.

Methods

We performed a retrospective cohort study to compare OS in pathologic stage I NSCLC patients undergoing varying CT surveillance intensities. Additionally, we performed descriptive analysis of trends in disease recurrence (locoregional and distant), and development of NPLC.

Study Population and Cohort Selection

Using the Special Study mechanism (described in the next section), we identified patients who underwent resection for pathologic stage I NSCLC. Patients were required to have undergone postoperative imaging surveillance. We only analyzed CT scans (as opposed to chest x-rays) based on published level I evidence advocating for CT imaging as standard of care in lung cancer screening¹¹. The National Lung Screening Trial randomized high-risk patients to receive low-dose CTs or standard chest x-ray^{11, 12}. The trial found that CT scanning was associated with significantly higher detection of NSCLC and lower mortality. Given this level I evidence suggesting the superiority of CT imaging, only CT scans were considered for analyses. We a priori constructed surveillance intensity cohorts (high intensity, moderate intensity, and low intensity) based on current guidelines. We created visit windows to assign patients into these cohorts based on days from surgery to first surveillance CT. These visit windows included 60–150 days (3 months), 151–300 days (6 months), and 301–450 days (1 year) which corresponded to the high, moderate, and low intensity cohorts, respectively.

Data Abstraction

To obtain the detailed level of information required for this study, registrars at Commission on Cancer (CoC) accredited institutions abstracted comorbidity, surveillance, and cancer recurrence/NPLC information utilizing the special study mechanism of the CoC, which supplemented data that already reside in the NCDB.

Up to 10 NSCLC patients were randomly selected from each institution for further abstraction. These patients underwent surgery for clinical stage I-III NSCLC (1/2006–12/2007), were alive 90 days post-surgery, and had available medical records. Patients with unknown recurrence status were excluded without replacement from the same institution. Patients were followed for five years post-diagnosis or until death.

Registrars obtained information on comorbidities, locoregional and distant recurrence, and development of NPLC for five years following surgery. Differentiation between disease recurrence and NPLC was based on original documentation by the treating clinician in the medical record. Locoregional recurrence was defined as tumor recurrence in the ipsilateral and/or regional lymph nodes, while distant recurrence was defined as tumor recurrence in the contralateral lung/lymph nodes or non-lung site.

Detailed information on postoperative imaging, date, and indication for imaging were recorded. Indications included (1) surveillance in absence of a new sign/symptom; (2) follow-up for new sign/symptom; (3) follow-up for suspicious finding on other imaging; (4) imaging performed in response to newly detected malignancy; (5) not cancer related; (6) or unable to determine. Registrars were instructed to use radiology reports as the primary source to obtain symptom status. They also had access to outpatient records including clinic and consult notes from visits that preceded the scan and made mention of intent to order imaging as a consequence of a new sign or symptom. These notes were from primary care providers, medical oncologists, radiation oncologists, surgeons, or other relevant providers. When available, information was collected on whether or not a patient underwent treatment (surgery, chemotherapy, radiation, or combination therapy) for recurrent disease. Registrars were trained by weekly webinars to standardize the abstraction process. Study data were merged with existing NCDB data by the Commission on Cancer, de-identified, and provided to the study team. Data collection was completed in 2015. Data were deemed not human subjects research and were exempted from IRB review.

Inclusion and Exclusion Criteria

We restricted our study inclusion to pathologic stage I patients who underwent surgical resection and had their first surveillance imaging CT scan between 60–450 days after surgical resection. Additionally, patients must have been asymptomatic at the time of the first postoperative CT. We excluded patients who underwent chemoradiation therapy, had positive surgical margins, or for whom the indication for first surveillance CT was unknown (Supplemental Figure 1).

Descriptive and Inferential Statistics

Descriptive statistics were reported using mean ± standard deviation for those variables that were approximately normally distributed, median (Q1, Q3) for highly skewed variables, and count (percent) for categorical variables. Continuous variables were compared between the three cohorts using Kruskal-Wallis tests, while categorical data were assessed with Chi-square or Fisher’s exact tests where appropriate. All p-values were two-tailed, and values <0.05 were considered significant. For each treatment cohort, the median number of asymptomatic surveillance images per person-year were calculated. Person-years were defined as follow-up time until death, loss to follow-up, or first onset of recurrence/NPLC. Median numbers of surveillance images were reported for 2 person-years, as the majority of existing practice guidelines are framed within the first two years of patient follow-up^8–10.

Cumulative incidences of locoregional/distant recurrence were estimated using competing risks analysis, with death or other recurrence/NPLC treated as competing risks. Interested covariates included in the disease recurrence competing risk models were determined a priori based on clinical experience and literature review, and included: pathologic stage, lymph nodes sampled, tumor size, tumor histology, tumor grade, and surgical procedure. The final model included the variables with p value of less than 0.10, where group, pathologic stage, and number of lymph nodes were forced into the model. Overall survival for all three cohorts was modeled using univariate and multivariable Cox proportional hazards regression analysis, and covariate effects are presented as hazard ratios with 95% confidence ratios. The proportionality assumption was tested by adding a time-dependent covariate for each variable. When the test indicated non-proportional hazards over time, models were divided into 2 time periods, and the maximized partial likelihood method was used to find the most appropriate breakpoint and the proportionality assumptions were further tested. Interested covariates in Cox proportional hazard modeling were also determined a priori based on literature and clinical significance, and included: age, gender, race, surveillance intensity group, chronic obstructive pulmonary disease (COPD), congestive heart failure (CHF), coronary artery disease (CAD), diabetes, psychiatric disease, pathologic stage, tumor size, tumor grade, lymph nodes sampled, tumor histology, and surgical procedure. The final model was built through the stepwise selection, where group, pathologic stage, and number of lymph nodes were forced into the model. Kaplan-Meier survival curves were also constructed to compare OS among surveillance cohorts.

Missing covariate data were less than or equal to 1% in all demographic variables: four patients were missing tumor histology, and one patient was missing gender status. Given the low number of patients with missing data, missing covariate information was reported in descriptive statistics, but was not included in regression analyses. Data that were reported as unknown or indeterminate were handled as separate categorical variables for analyses. All analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC).

Results

We identified 4340 patients with pathologic stage I NSCLC who underwent postoperative CT imaging. When examining indication for the first CT scan, 3170 (73%) patients were asymptomatic and eligible for inclusion as true surveillance imaging. 2565 patients received their first scan within 60–450 days from surgery. We excluded 123 patients who underwent chemoradiation therapy or had positive surgical margins, leaving 2442 patients in our study cohort that met inclusion criteria. Of included patients, 805 (33%), 1216 (50%), and 421 (17%) patients were placed in the high intensity, moderate intensity, and low intensity cohorts, respectively (Supplemental Figure 1). When calculating the median number of asymptomatic surveillance images, high intensity, moderate intensity, and low intensity groups received 4.67 (2.43, 15.02), 3.65 (1.97, 12.54), and 2.96 (1.71, 10.27) images per 2 person-years, respectively. This confirmed that our pre-defined surveillance intensity cohorts, which were based on time from surgery to time to first surveillance scan, corresponded to differing frequencies of imaging. The cohorts demonstrated similarity across most patient, tumor, and treatment-related variables (Table 1).

Table 1.

Patient and tumor-related factors, All Patients

	All patients (n=2442)	High Intensity (n=805)	Moderate Intensity (n=1216)	Low Intensity (n=421)	p-value
Patient-Related Factors
Age (years)	66.2 (9.7)	65.7 (9.9)	66.4 (9.7)	66.6 (9.6)	0.171
Sex (Male)	1106 (45.3%)	340 (42.3%)	569 (46.8%)	197 (46.8%)	0.110
Race					0.698
Caucasian	2198 (90.0%)	729 (90.6%)	1096 (90.1%)	373 (88.6%)
African-American	176 (7.2%)	56 (7.0%)	88 (7.3%)	32 (7.6%)
Other	68 (2.8%)	20 (2.4%)	32 (2.6%)	16 (3.8%)
Charlson/Deyo Score					0.381
0	1236 (50.6%)	414 (51.4%)	622 (51.2%)	200 (47.5%)
1	890 (36.4%)	278 (34.5%)	445 (36.6%)	167 (39.7%)
2+	316 (13.0%)	113 (14.1%)	149 (12.2%)	54 (12.8%)
Comorbidity
Chronic obstructive pulmonary disease	998 (40.9%)	324 (40.3%)	485 (39.9%)	189 (44.9%)	0.179
Congestive Heart Failure	131 (5.4%)	50 (6.2%)	60 (4.9%)	21 (5.0%)	0.428
Coronary Artery Disease	531 (21.7%)	168 (20.9%)	260 (21.4%)	103 (24.5%)	0.319
Diabetes	357 (14.6%)	129 (16.0%)	170 (14.0%)	58 (13.8%)	0.385
Peripheral Vascular Disease	224 (9.2%)	64 (8.0%)	109 (9.0%)	51 (12.1%)	0.053
Psychiatric History	197 (8.1%)	76 (9.4%)	88 (7.2%)	33 (7.8%)	0.201
Substance Abuse	124 (5.1%)	46 (5.7%)	56 (4.6%)	22 (5.2%)	0.533
Tumor-Related Factors
Surgical Resection					0.478
Wedge Resection	353 (14.4%)	105 (13.0%)	183 (15.1%)	65 (15.5%)
Segmentectomy	65 (2.7%)	28 (3.5%)	26 (2.1%)	11 (2.6%)
Lobectomy/Bilobectomy	1980 (81.1%)	659 (81.9%)	985 (81.0%)	336 (79.8%)
Pneumonectomy	44 (1.8%)	13 (1.6%)	22 (1.8%)	9 (2.1%)
Tumor Size, mm (IQR)	22 (15, 30)	22 (15, 30)	22 (15, 30)	21 (15,30)	0.819
Number of lymph nodes sampled (IQR)	1 (1, 1)	1 (1, 1)	1 (1, 1)	1 (1, 1)	0.278
Pathologic Stage Stage					0.904
Stage 1A	1497 (61.3%)	485 (60.2%)	754 (62.0%)	258 (61.3%)
Stage 1B	866 (35.5%)	292 (36.3%)	426 (35.0%)	148 (35.1%)
Stage 1 (A/B unknown)	79 (3.2%)	28 (3.5%)	36 (3.0%)	15 (3.6%)
Histology					0.935
Adenocarcinoma	1161 (47.6%)	391 (48.6%)	572 (47.2%)	198 (47.0%)
Squamous	647 (26.5%)	209 (26.0%)	324 (26.7%)	114 (27.1%)
Other	489 (20.1%)	159 (19.7%)	241 (19.9%)	89 (21.1%)
Unknown	141 (5.8%)	46 (5.7%)	75 (6.2%)	20 (4.8%)
Grade					0.567
Well differentiated	402 (16.5%)	136 (16.9%)	195 (16.0%)	71 (16.9%)
Moderately differentiated	1071 (43.9%)	345 (42.9%)	540 (44.4%)	186 (44.1%)
Poorly differentiated	765 (31.3%)	251 (31.2%)	379 (31.2%)	135 (32.1%)
Undifferentiated	18 (0.7%)	8 (1.0%)	5 (0.4%)	5 (1.2%)
Indeterminate	186 (7.6%)	65 (8.0%)	97 (8.0%)	24 (5.7%)

Surveillance Intensity and Survival Analysis

Cox proportional hazard modeling showed that surveillance intensity was not associated with OS (p=0.302) (Table 2). Factors associated with worse overall survival included age, pathologic stage, male gender, COPD, CHF, psychiatric disease, histologic grade, and non-lobectomy surgical resection (p<0.05). Kaplan Meier analysis also showed similar 5-year OS probability among our surveillance intensity cohorts (70.7% high intensity vs. 70.9% moderate intensity vs. 73.2% low intensity, p=0.547) (Figure 1).

Table 2.

Cox Proportional Hazard Model of 5-year Overall Survival

Covariate	Hazard Ratio	P Value
Age	1.03 (1.02–1.04)	<0.001
Gender
Male	(reference)	0.036
Female	0.85 (0.73–0.99)
Comorbidity
Chronic obstructive pulmonary disease	1.27 (1.09–1.48)	0.003
Congestive heart failure	1.57 (1.20–2.05)	0.001
Psychiatric History	1.54 (1.20–1.98)	0.001
Pathologic Stage
Stage 1A	(reference)
Stage 1B	1.25 (1.06–1.46)	0.024
Stage 1	1.11 (0.73–1.70)
Histologic Grade
Well differentiated	(reference)	<0.001
Moderately differentiated	1.65 (1.26–2.16)
Poorly differentiated	1.76 (1.32–2.35)
Undifferentiated	4.02 (2.03–7.97)
Unknown	1.62 (1.11–2.36)
Resection type
Lobectomy/bilobectomy	(reference)	0.015
Wedge resection	1.32 (1.07–1.63)
Segmentectomy	1.39 (0.93–2.09)
Pneumonectomy	1.48 (0.89–2.46)
Number of lymph nodes sampled	0.97 (0.84–1.12)	0.689
Surveillance group		0.302
Low intensity	(reference)
Moderate intensity	1.13 (0.91–1.39)
High intensity	1.20 (0.95–1.50)

Figure 1.

Overall Survival by Surveillance Intensity Group

Locoregional Disease Recurrence

Five-year cumulative incidence of locoregional recurrence was similar across surveillance intensity cohorts (high intensity: 0.13 95% CI: 0.11–0.15; moderate intensity: 0.13 95% CI: 0.11–0.15; low intensity: 0.14 95% CI: 0.11–0.17; p=0.967) (Figure 2). Variables associated with increased risk of locoregional disease recurrence included tumor size, histology, and non-lobectomy resection type (p<0.05) (Table 3). Of those with locoregional disease recurrence, 166 (49.9%), 96 (28.8%), and 66 (19.8%) had evidence of locoregional disease in the same lung, regional lymph nodes, or both lung and lymph nodes, respectively. For five patients (1.5%), the location of local recurrence was unable to be determined. When determining how local recurrences were first detected, 210 (63.1%) patients had disease first detected by routine surveillance CT scans and were asymptomatic at the time of imaging (Supplemental Table 1).

Figure 2.

Risk of locoregional disease recurrence by surveillance intensity

Table 3:

Competing Risks Model for Time to Locoregional Cancer Recurrence

	HR (95% CI)	P-value
Tumor size (mm)	1.02 (1.00–1.03)	0.014
Substance abuse	1.67 (1.08–2.58)	0.022
Resection type
Wedge resection	(reference)	0.005
Segmentectomy	1.30 (0.70–2.42)
Lobectomy/bilobectomy	0.55 (0.35–0.89)
Pneumonectomy	0.77 (0.31–1.91)
Pathologic
Stage 1A	(reference)	0.901
Stage 1B	0.93 (0.68–1.27)
Stage 1	1.02 (0.56–1.85)
Histology:
Adenocarcinoma	(reference)	0.004
Squamous	0.91 (0.70–1.19)
Other histology	0.51 (0.36–0.74)
Unknown	0.82 (0.49–1.37)
Number of lymph nodes sampled	0.68 (0.33–1.42)	0.305
Surveillance Group
Low intensity	(reference)	0.944
Moderate intensity	0.98 (0.72–1.35)
High intensity	1.03 (0.74–1.43)

Of those who had locoregional recurrence, 243 (73.0%) of patients underwent subsequent treatment after detection of locoregional recurrence (Supplemental Table 2). Of those that did not undergo subsequent treatment, 24 patients (21.8%), 30 patients (18.4%), and 16 patients (26.7%) belonged to the high intensity, moderate intensity, and low intensity groups, respectively.

Distant Recurrence

Five-year cumulative incidence of distant recurrence was significantly difference across surveillance intensity groups (high intensity: 0.15 95% CI: 0.13–0.18; moderate intensity: 0.13 95% CI: 0.11–0.15; low intensity: 0.10 95% CI: 0.07–0.13; p=0.024) (Figure 3). Variables associated with distant recurrence included surveillance intensity, tumor size, tumor grade, and histology (p<0.05) (Table 4). Bone metastases were the most common site of distant recurrence in 89 patients (26.6%), followed by contralateral lung and brain metastases in 87 (26.0%) and 76 (22.8%) patients, respectively. Routine surveillance imaging detected distant recurrence in 122 patients (36.6%), while 135 patients (40.5%) had recurrence detected by imaging ordered after new signs or symptoms were documented by a healthcare provider (Supplemental Table 1). Of those with distant recurrence, 83 patients (24.9%) did not undergo any treatment (Supplemental Table 2), with 24 patients (19.4%), 43 patients (26.4%), and 16 patients (34.8%) belonging to the high intensity, moderate intensity, and low intensity groups, respectively.

Figure 3.

Risk of distant recurrence by surveillance intensity

Table 4:

Competing risks model for time to distant cancer recurrence

	HR (95% CI)	P-value
Tumor size (mm)	1.02 (1.01–1.04)	0.001
Resection type
Wedge resection	(reference)	0.121
Segmentectomy	1.28 (0.63–2.58)
Lobectomy/bilobectomy	0.91 (0.63–1.30)
Pneumonectomy	1.90 (0.90–4.03)
Pathologic Stage
Stage 1A	(reference)	0.504
Stage 1B	0.88 (0.64–1.21)
Stage 1	1.26 (0.68–2.32)
Histology:		0.006
Adenocarcinoma	(reference)
Squamous	0.62 (0.46–0.83)
Other histology	0.71 (0.49–1.03)
Unknown	1.01 (0.64–1.61)
Histologic Grade
Well differentiated	(reference)	0.018
Moderately differentiated	1.72 (1.13–2.60)
Poorly differentiated	2.01 (1.29–3.13)
Undifferentiated	2.73 (0.84–8.81)
Unknown	2.18 (1.28–3.71)
Number of lymph nodes sampled	1.02 (0.79–1.31)	0.886
Surveillance Group
Low intensity	(reference)	0.032
Moderate intensity	1.31 (0.93–1.85)
High intensity	1.59 (1.12–2.27)

Development of New Primary Lung Cancer (NPLC)

Five-year cumulative incidence of NPLC was similar across surveillance intensity groups (high intensity: 0.05 95% CI: 0.04–0.08; moderate intensity: 0.08 95% CI: 0.06–0.11; low intensity: 0.07 95% CI: 0.06–0.09; p=0.393). Ninety-seven patients (54.5%) developed a NPLC with lung cancer histology different from their initial tumor. Routine surveillance imaging detected NPLC in 128 cases (71.9%) (Supplemental Table 1).

Conclusion

There are more than 400,000 lung cancer survivors in the United States¹³. These individuals are at high risk for development of recurrence or NPLC^{1, 13}. Despite the need for evidence-based guidelines to inform optimal surveillance strategies, existing guidelines are not uniform and are based on small retrospective studies and expert opinion. Organizations including the American College of Chest Physicians (ACCP), the National Comprehensive Cancer Network (NCCN), and the International Association for the Study of Lung Cancer offer differing recommendations for intervals and modalities for follow-up imaging^8–10. Variability of surveillance in clinical practice is even greater, with adherence to guidelines often being quite poor. In a study of adherence to NCCN and ACCP guidelines, Erb et al. utilized the Surveillance, Epidemiology, and End Results (SEER)-Medicare database and found that only 61.4% of stage I NSCLC patients received guideline-adherent surveillance during the initial 2 years after treatment¹⁴. Poor adherence can be partially explained by the paucity of quality longitudinal data to inform best practice.

The presumed role for surveillance imaging is three-fold: (1) the early detection of recurrence; (2) screening for NPLC; and (3) to offer timely subsequent treatment when possible. Using a large, nationally representative database containing complete 5-year follow-up data, we found that surveillance intensity was not associated with overall survival in patients with stage I surgically resected NSCLC. Additionally, we did not find surveillance intensity to be associated with significant differences in time to detection of locoregional recurrence, which represents a group that could potentially benefit from subsequent curative therapy.

Our findings are similar to studies that have previously reported on intensity of follow-up. Calman et al. published a meta-analysis of 1669 stage I-III NSCLC patients comparing intensity of postoperative follow-up programs⁵. The majority of included studies were small and retrospective, with only one randomized controlled trial and one prospective study. The authors found no difference in OS for NSCLC patients treated with curative intent who received more intensive follow-up. They too did not find a correlation between intensive follow-up and reduced time to local recurrence detection. However, their analysis used wide inclusion criteria, resulting in significant heterogeneity of included studies, patient populations, and surveillance interventions. This limits the direct conclusions that can be inferred to the impact of surveillance imaging intensity. We chose to focus on stage I NSCLC. This allowed for comparison within a more homogeneous population, and thus allowed for less biased estimates of disease recurrence and survival.

Theoretically, detection of asymptomatic recurrence or NPLC could offer a survival advantage as these diseases are more likely to be early stage and amenable to therapy. Calman et al. analyzed survival by asymptomatic vs. symptomatic presentation in their meta-analysis and found asymptomatic recurrence to be associated with longer survival time (HR: 0.61, CI: 0.60–0.74)⁵. However, asymptomatic presentation is likely a function of the patient’s initial stage of disease. Although our results cannot support the aggressive use of CT surveillance, this does not preclude the use of CT surveillance for stage I patients. Our data have shown that recurrence and NPLC are often detected at the asymptomatic period, and many go on to receive subsequent therapy. Sixty-three percent of locoregional recurrences and 72% of NPLC were detected via routine surveillance. Other stage-specific analyses have shown similar findings. Lou et al. retrospectively reviewed patients who had undergone resection for stage I-IIIA disease and subsequently underwent CT surveillance every 6–12 months³. They identified 1294 stage I /II patients, of which 257 (20%) had recurrence. Similar to our study, surveillance CT detected 61% of asymptomatic recurrences in these patients. Approximately 78% of the recurrences in early stage patients reported by Lou et al. underwent subsequent therapy, with 12% being offered surgery with or without radiation. Our study identified that 73% of locoregional recurrences underwent subsequent therapy, with at least 7% undergoing surgical treatment.

The role of surveillance imaging has previously been debated given that the efficacy of post-recurrence therapies has not been consistently demonstrated. However, in a large retrospective study of 9001 stage I-III NSCLC patients with post-resection recurrence, subsequent treatment has been shown to be highly beneficial. Using the same special study, Wong et al. compared post-recurrence survival between those who received active therapy vs. supportive care¹⁵. Similar to our study, a high percentage (79.5%) of patients with locoregional recurrence received some form of treatment. Median survival for these patients was 19.9 months compared to 4.4 months in patients who received supportive care only. Interestingly, this same pattern was observed in distant recurrence. In our study, 69.4% of patients with distant recurrence underwent some form of therapy, and more frequent imaging was associated with higher incidence of detection of distant recurrence. Wong et al. reported that 77.3% of patients with distant metastases underwent treatment, which translated to longer median survival (11.6 months vs. 3.0 months).

While treatment of distant recurrence was traditionally associated with poor outcomes, innovations in immunotherapy may change future management. Ghandi et al. published a randomized trial in patients with stage IV NSCLC that compared standard platinum-based chemotherapy with pembrolizumab to standard chemotherapy with placebo. The results were promising, showing an almost 20% increase in estimated 12-month OS, with a benefit seen across all levels of programmed death ligand 1 (PD-L1) expression. These innovations may increase the future relevance of surveillance imaging for distant disease.

There are important limitations to note. Our study consisted of patients that were diagnosed more than ten years ago. These patients were used so additional data abstraction on 5-year follow-up data could be performed. Thus, there was a trade-off required to capture complete 5-year follow-up information including recurrence and surveillance, which are not routinely captured in the NCDB. An additional and important limitation includes our categorization of surveillance intensity cohorts. In the real world, patients and physicians often deviate from interval-based regimens (especially as time from primary treatment increases). Previous studies have documented remarkably low adherence to existing guideline-recommended surveillance regimens, approximating 60%¹⁴. Additionally, one must contend with several biases when quantifying surveillance in observational research. First, there is healthy survivor bias—patients who remain on their intended surveillance frequency and do not deviate from their regimens are more likely to be healthy. This would be due to the fact that patients who receive more regimented surveillance scans over time would have to be alive, without recurrence, and without new symptoms or findings on surveillance imaging that prompted a change to more frequent imaging. Alternatively, there is a possibility for an “unhealthy survivor bias.” Patients who have a recurrence, NPLC, or new signs or symptoms could then be subject to more frequent imaging, resulting in a higher number of scans. Due to these challenges, there is no perfect method to measuring surveillance intensity in observational studies. Thus, we created our cohorts a priori based on how surveillance initially happened, using time from surgery to first surveillance scan as a proxy. As expected, we found that patients in the dataset did not adhere to strict regimens of surveillance imaging throughout their follow-up. However, the median numbers of scans per person year in each cohort were significantly different between the cohorts. An additional limitation is that our analysis is confined to the variables captured in the NCDB. We observed a trend towards lower intensity surveillance in patients who did not undergo any therapy for locoregional/distant recurrence. It is possible that clinicians who survey less aggressively tend to not pursue treatment for patients who recur, representing a treatment selection bias that could lead to worse OS. However, our analysis found no difference in OS between the surveillance intensity cohorts. Alternatively, it is possible that these patients had concerning factors that were not captured in our dataset. Factors including functional status, pulmonary function, margin distance, and presence of lymphovascular invasion are are not included in the Special Study dataset.

Despite these limitations, our study exhibits several strengths. Our study utilized high quality, longitudinal data from re-abstracted NCDB records that contain complete 5-year recurrence and survival information. The dataset also includes surveillance imaging indications, which are critical to distinguish true surveillance from imaging ordered to investigate a new sign or symptom. Additionally, as the NCDB includes 70% of new lung cancers in the United States, these data are broadly representative and our conclusions are generalizable¹⁶.

We observed that surveillance intensity was not associated with overall survival, suggesting that imaging frequency is not a one-size-fits-all approach. Necessary considerations when determining the optimal utilization of any surveillance test include additional factors such as accuracy, financial cost and reimbursement for diagnostic studies, the invasiveness of diagnostic procedures, and the emotional stress from fear of recurrence^{1, 5}. An optimal, patient-centered approach to surveillance should be driven by knowledge of tumor biology, patient history, patient preference, and candidacy for subsequent treatment. Additional high quality longitudinal studies are needed to determine how surveillance intensity best fits into this comprehensive approach.

Supplementary Material

Supplementary Figure 1

Supplemental Figure 1. Inclusion/Exclusion Criteria and Selection of Study Cohort

Supplementary Figure 2

Supplemental Figure 2. Video summary of study design and findings

Central Message

Surveillance imaging intensity after resection for pathologic stage I non-small cell lung cancer is not associated with long-term overall survival.

Perspective Statement

There are limited quality long-term data on postoperative imaging surveillance for patients with resected non-small cell lung cancer. Using a large, nationally representative database containing longitudinal data on surgically resected pathologic stage I non-small cell lung cancer patients, we compared 5-year overall survival among patients undergoing different imaging surveillance intensities.

Glossary of Abbreviations

ACCP

American College of Chest Physicians

CoC

Commission on Cancer

COPD

Chronic Obstructive Pulmonary Disease

CHF

Congestive Heart Failure

CT

Computed Tomography

NCCN

National Comprehensive Cancer Network

NCDB

National Cancer Database

NPLC

New Primary Lung Cancer

NSCLC

Non-Small Cell Lung Cancer

OS

5-year Overall Survival

PD-L1

Programmed Death-Ligand 1

SEER

Surveillance, Epidemiology, and End Results Program