Real-world Decision-making Process for Stereotactic Body Radiotherapy Versus Minimally Invasive Surgery in Early-stage Lung Cancer Patients.

Introduction
The local treatment options for early-stage non-small cell lung cancer (NSCLC) include surgical resection, stereotactic body radiotherapy (SBRT), or computed tomography-guided ablation.1,2 Despite efforts to directly compare surgical resection to SBRT in randomized clinical trials, these trials were closed early due to poor accrual.3 A pooled analysis of these randomized trials involving 58 patients showed comparable recurrence-free survival but improved overall survival (OS) with SBRT in operable stage I lung cancer. Notably, in the surgical cohort, the majority of patients (74%) underwent open surgery and 7% had disease progression at time of surgery.3,4 Studies in the real-life clinical setting, although often confounded by imbalances in age and performance status, find SBRT to provide excellent short-term morbidity, mortality and favorable tumor control.5–7 However, although SBRT and MIS are well tolerated by the majority of patients, radiation-induced pneumonitis and esophagitis or postoperative complications can affect the quality of life after treatment and need to be taken into consideration in the decision-making process.8–11 In addition, while recurrence-free survival and OS after SBRT appeared worse compared to surgery in the overall cohorts,5–7 when adjusted for confounding variables treatment modality did not appear to be associated with survival.4,5 This observation may suggest that when patients are appropriately selected for the optimal local treatment modality, very comparable outcomes can be achieved with either surgery or SBRT.
Consequently, surgical resection, and in current thoracic practices, especially minimally invasive surgery (MIS),1 remains the standard of care for operable patients, with SBRT more frequently recommended for those deemed high-risk for surgery or declining surgery.12,13 In our practice, patients are first evaluated by thoracic surgeons for technical resectability and medical operability. Technical resectability usually does not pose a problem in early-stage NSCLC. Various factors, such as the use of home oxygen, severe underlying pulmonary disease with poor pulmonary function, increased frailty and age, have been recognized to increase the risk for postoperative short- and long-term morbidity and mortality and inform medical operability.6,13–16 Subsequently, when these factors are significant enough to be deemed medically inoperable, patients are referred to SBRT as the definitive treatment. However, the extent to which these factors influence decision-making in real-life clinical practice, and consequently, who should be selected for surgery vs SBRT, remains unclear and challenging.
Therefore, the aim of our study was to identify clinical factors associated with selection for SBRT by thoracic surgeons and develop a decision-making model to predict these referrals. Secondly, we assessed recurrence and OS for patients who underwent MIS and those who received SBRT in order to validate the performance of the model.

Patients and Methods

Study population and data collection
This study included all consecutive patients who underwent intended MIS surgery or SBRT for clinical stage I NSCLC at a tertiary reference cancer center from January 1, 2020 to July 31, 2023. All patients had cytologic or pathologic confirmation of NSCLC (adenocarcinoma, squamous cell carcinoma or NSCLC not otherwise specified). Patients who received neoadjuvant therapy, open surgery, had preoperatively diagnosed in situ carcinoma or were surgical candidates who opted for SBRT were excluded from the analysis (Supplemental Figure 1, Supplemental Digital Content 1). The study received approval by the Institutional Review Board (#18–391), and all procedures were conducted in accordance with the Declaration of Helsinki. This study adhered to the STROBE reporting guidelines.17
Data regarding the surgical patients was obtained from a prospectively maintained database managed by the Thoracic Service. Patients referred to SBRT by Thoracic Service surgeons were identified by Dataline, our institutional database collection service operated by the technology department. Data on demographics, patient characteristics, staging, pathology, post-procedural outcomes, survival and recurrence were collected from the patient’s hospital stay and outpatient visits at our institution. For patients receiving additional treatment and/or follow-up outside our institution, information regarding their status was obtained by telehealth, telephone follow-up or outside correspondence received by caregivers.
The MSK-frailty score was used to evaluate the patient’s clinical frailty and is composed of the following clinical parameters each contributing to 1 point on a scale from 0 to 11: chronic obstructive pulmonary disease, acute myocardial infarction, congestive heart failure, hypertension, peripheral artery disease, coronary artery disease, stroke, transient ischemic attack, diabetes, cognitive impairment, reduced activities of daily living.16 Post-procedural outcomes included 90-day mortality, and overall complication rate for surgical patients, while for SBRT patients this included new-onset or worsened treatment-induced toxicity (cough, dyspnea, chest wall pain, pneumonitis, pulmonary fibrosis, bronchial stricture, esophagitis, dysphagia, myocardial infarction, pericarditis and radiation dermatitis).9 All complications were graded according to the Common Terminology Criteria for Adverse Events (version 5) on a scale from 1 to 5, based on the intervention needed to treat the complication.
Recurrence was differentiated from metachronous lesions through pathological comparison, assessment of clonal similarity through genomic analysis (when available)18 or the Martini and Melamed criteria.19 Information regarding the date and cause of death were obtained from various sources, including in-house deaths, Social Security Death Index updates, Medicare database searches, and death notifications from caregivers via physician offices to the Cancer Registry’s Death Notification mailbox. Follow-up was closed on January 31, 2024.

Statistical analysis
Categorical variables are described as number (percentage) and continuous variables as median (interquartile range, IQR). The patient characteristics by treatment modality were compared using the chi-squared test or Fisher’s exact test for categorical variables and Mann-Whitney U or t test for continuous variables.
Our primary objective was to identify factors associated with referring patients to SBRT. A multivariable logistic regression model, including significant patient demographics and clinical factors identified through univariable analysis, was used to determine factors independently associated with receiving SBRT. Significant factors from the multivariable model were utilized to develop a nomogram to calculate the probability of selection for SBRT. The performance of the prediction model was evaluated using the receiver operating characteristic (ROC) curve analysis, with optimism correction, as well as a calibration curve analysis. We calculated two cutoff values based on ROC curve and distribution of patients, stratifying the patients in 3 categories on the basis of their perioperative risk, including postoperative complications. The first cutoff identified patients who were most likely to be referred to MIS, maximizing sensitivity while maintaining a minimum specificity of 90%, which is considered indicative of “good to perfect” predictive performance according to Plante and Vance, defining the low-risk category which would select patients for surgery.20 The second cutoff was identified patients who were unlikely to be considered from MIS, maximizing sensitivity while maintaining a minimum specificity of 99%, defining the high-risk category which would select patients for SBRT. Patients with probability scores falling between these two cutoff values were considered intermediate-risk. Additionally, the Youden index was calculated for exploratory purposes. The calculation of the cutoff values was performed using the cutpointr package.
Our secondary objective was to validate the performance of the model by assessing in post-procedural outcomes, recurrence and OS between patients who underwent MIS and those who received SBRT. Cumulative incidence of recurrence (CIR) was calculated from date of surgery or end of SBRT to the time of recurrence; treating death without recurrence as a competing event. OS was measured from date of surgery or end of SBRT to date of death from any cause. CIR was estimated using the Fine & Gray method and compared between the MIS and SBRT groups using Gray’s test. OS was estimated using the Kaplan-Meier method and compared between the MIS and SBRT groups using log-rank tests. A Cox proportional hazard analysis was performed to identify factors associated with OS, with significant factors identified through univariable analysis included in the multivariable analysis.
All statistical tests were 2-tailed, and p<0.05 was considered to indicate statistical significance. Statistical analyses were performed using R (version 4.3.3, R Foundation for Statistical Computing, Vienna, Austria).

Results

Patient cohort and treatment
In total, 1291 patients were treated for stage I adenocarcinoma or squamous cell carcinoma NSCLC, of which 1116 (86%) patients underwent intended MIS and 175 (14%) received SBRT. Patients treated with SBRT were typically older (77 years versus 70 years, p<0.001) and exhibited worse performance status (p<0.001), had lower pulmonary function test values (p<0.001) and a higher MSK-Frailty score (p<0.001) (Table 1 and Supplemental Table 1, Supplemental Digital Content 1).
Among the 1116 patients who underwent surgical resection, 582 (52%) underwent lobectomy or bilobectomy, 260 (23%) underwent segmentectomy and 274 (25%) underwent wedge resection. Additionally, 52 patients (5%) required conversion to open surgery to complete the surgical resection and 13 (1%) had an R1 resection upon final histopathological evaluation. The median (IQR) total dose of SBRT was 50 Gy (48 – 54 Gy), most commonly delivered as 10 Gy × 5 fractions. All patients successfully completed their entire SBRT course.

Factors associated with selection for SBRT and construction of a decision-making model
On univariable analysis, the factors associated with selection for SBRT included age, performance status of 2–3, having a smoking history, chronic renal insufficiency, previous pulmonary resection, MSK-frailty score, forced expiratory volume in one second (FEV1), and diffusion capacity of the lung for carbon monoxide (DLCO) (Table 2). Upon inclusion of these factors in a multivariable analysis, age (odds ratio [OR], 1.09, 95% confidence interval [CI]: 1.05–1.13), performance status of 2 or 3 (OR, 7.40, 95%CI: 3.21–17.08), previous pulmonary resection (OR, 3.31, 95%CI: 1.98–5.54), MSK-Frailty score (OR, 1.50, 95%CI: 1.27–1.77), FEV1 (OR, 0.97, 95%CI: 0.96–0.98) and DLCO (OR, 0.96, 95%CI: 0.94–0.97) remained independently associated with selection for SBRT. Figure 1 displays the nomogram constructed from these independent variables.
The prediction model, constructed from these significant variables, demonstrated an area-under-the-curve of 0.908 (95%CI: 0.876–0.938), with an optimism-corrected area-under-the-curve of 0.902. Figure 2A illustrates the distribution of the probabilities for both MIS and SBRT. Figure 2B demonstrates the calibration curve analysis, indicating a close to ideal estimation of selection for SBRT. The bias-corrected curve showed minimal overestimation of selection of SBRT for predicted probabilities above 50%. Based on the ROC analysis for patients with probability scores calculable (n=1197; MIS: n=1076 and SBRT: n=121), a cutoff probability of 17% included 90% of all patients treated with MIS, corresponding to a sensitivity of 77%. Additionally, the cutoff probability of 60% identified 99% of all patients treated with MIS, corresponding to a sensitivity of 33%. Consequently, the low-risk category included 998 patients (MIS, n=970 and SBRT, n=28), the intermediate-risk category included 149 patients (MIS, n=96 and SBRT, n=53) and the high-risk category included 50 patients (MIS, n=10 and SBRT, n=40). The intermediate-risk category was assessed for factors associated with selection for SBRT. On multivariable analysis, hypertension (OR, 0.39 [95% CI, 0.16–0.97]; p=0.042) and COPD (OR, 3.09 [95% CI, 1.31–7.85]; p=0.009) were associated with a referral to SBRT (Supplemental Table 2, Supplemental Digital Content 1). Finally, the intermediate-risk category was assessed for factors associated with survival via a Cox regression analysis; only age (hazard ratio [HR], 1.12 [95% CI, 1.02–1.23]; p=0.021) was identified as a predictor of OS on univariable analysis. The HR of receiving SBRT was 1.35 (95% CI, 0.43–4.20) (p=0.60) (Supplemental Table 3, Supplemental Digital Content 1). For completeness, using the Youden method, an optimal cutoff probability of 7.6% was calculated for the prediction model, reflecting a sensitivity of 90% and specificity of 79%. The changes in sensitivity and specificity in relation to different cutoff probabilities is presented in Supplemental Figure 2, Supplemental Digital Content 1, and Supplemental Table 4, Supplemental Digital Content 1.

Outcome after MIS and SBRT
The median follow-up of the total cohort was 1.75y (1–2.6), including 1.73y (1–2.5) for the SBRT group and 1.80 (1–2.6) for the MIS group. Regarding the overall cohort, it is important to recognize the distinct characters of the patient populations. The rates of postoperative overall complication and radiation-induced toxicity were 28% (n=310) for MIS and 29% (n=51) for SBRT, respectively (p=0.71). The incidence of grade ≥3 complications was 4% (n=48) for MIS and 3% (n=5) for SBRT (p=0.37). In particular, radiation-induced pneumonitis could be observed in 7% (n=13) of the patients who received SBRT, including 5 grade 1 (3%) and 8 grade 2 (4.6%) pneumonitis. The 90-day mortality rates were 0.4% (n=5) in the MIS group and 0.6% (n=1) in the SBRT group (p=0.58). Patients who underwent MIS (12%, n=139) received adjuvant therapy more frequently compared to SBRT (1%, n=2; p<0.001).
When the oncological outcomes were compared throughout the entire study population, the CIR for any recurrence (3-years: SBRT, 19% [95%CI: 12%−25%] vs MIS, 9% [95%CI:6%−12%]; Gray’s p<0.001), locoregional recurrence (3-years: SBRT, 9% [95%CI: 4%−14%] vs MIS, 3% [95%CI: 2%−5%]; Gray’s p<0.001) and distant recurrence (3-years: SBRT, 9% [95%CI: 4%−14%] vs MIS, 6% [95%CI: 4%−8%]; Gray’s p=0.004) was significantly different (Supplemental Figure 3A–C, Supplemental Digital Content 1). In addition, OS was significantly different (Table 3 and Supplemental Figure 3D, Supplemental Digital Content 1). However, the majority of patients in both the SBRT (74%, 23/31) and MIS group (71%, 30/42) died from causes unrelated to lung cancer. In the intermediate-risk group, which included patients eligible for either treatment modality, OS was comparable between SBRT (3-years: 83% [95%CI: 70%−99%]) and MIS (3-years: 91% [95%CI: 85%−98%]; log-rank p=0.60) (Figure 3). This finding underscores the importance of multidisciplinary discussion and decision-making for these patients in order to select the appropriate treatment modality for each patient. The clinical factors identified within
In support of the validity of the clinical decision-making process, the Cox multivariable proportional regression analysis for OS only identified age (HR, 1.05, 95%CI: 1.01–1.09; p=0.019) and increasing tumor stage (HR of stage IB, 4.11, 95%CI: 1.18–14.25; p=0.026) as independent predictors of OS, with MSK-frailty score (HR, 1.19 [95%CI: 0.99–1.42]; p=0.058) and reduced DLCO (HR, 1.02, 95%CI: 1.00–1.03; p=0.060) being other important risk factors. Treatment modality was not associated with OS (HR of SBRT, 1.67, 95%CI: 0.80–3.48; p=0.20) (Supplemental Table 5, Supplemental Digital Content 1).

Discussion
The existing literature demonstrates excellent morbidity and short-term mortality in patients undergoing MIS or SBRT.3,6,21–23 However, long-term outcomes of SBRT compared to MIS remain ambiguous, even after propensity score matching.5,6,21 Consequently, although randomized controlled trials (e.g. STABLE-MATES; NCT02468024)24 and comparative multicenter prospective trials (e.g. SORT; NCT05183932) are still ongoing, the question as to whether SBRT should be considered as an alternative for patients who can tolerate surgery seems more and more irrelevant, as direct comparison of the treatment modalities remains challenging. In fact, at present, surgery and SBRT are conducted in distinct patient populations, evidenced by the different clinical profiles of patients undergoing these treatments.5,7,25 This difference in patient demographic and the distinct patient selection for each approach is corroborated by the differences in age, comorbidities and functional status between surgical and SBRT patients in our study. In this context, SBRT remains an invaluable treatment option for patients with NSCLC who are deemed inoperable or who decline surgery.12,26 Consequently, the pertinent question then becomes: Which patients should be considered as inoperable or high risk for surgery, even MIS, and subsequently be referred to SBRT? Our study, including nearly 1300 patients with clinical stage I NSCLC, is the first to identify a comprehensive array of clinical factors, including age, reduced performance status, previous pulmonary resection, MSK-Frailty score, FEV1 and DLCO, independently associated with the referral to SBRT. Additionally, we were able to develop a robust prediction model based on these variables, offering valuable insights into the decision-making process surrounding SBRT referral.
The factors influencing selection for SBRT are not surprisingly factors known to be associated with increased postoperative morbidity and long-term mortality.15,16,27,28 As highlighted in a consensus statement of the American Association of Thoracic Surgeons, patient frailty, DLCO, performance status and FEV1 rank among the top ten predisposing patients with stage I lung cancer to increased perioperative risk when undergoing lobectomy.13 Consequently, severe impairments or combinations of these factors may prompt clinicians to consider referring patients to SBRT.27 The factors identified in our study are thus highly relevant and likely already routinely collected by a significant proportion of thoracic surgeons or pulmonologist world-wide. This suggests a straightforward implementation of our model into clinical practice.
To facilitate the practical application, we categorized the patient cohort into 3 risk categories based on the mathematical analysis of the ROC curve. The majority of patients in the low-risk category were referred to MIS. It can be reasoned that patients in this category who underwent SBRT would also likely have tolerated and experienced favorable outcomes after surgery. Conversely, the majority of patients in the high-risk category underwent SBRT, suggesting that patients who underwent MIS in this category may have been exposed to increased perioperative risks. Finally, the intermediate-risk patients were more evenly distributed, emphasizing the need for a nuanced and multidisciplinary decision-making. For these patients, alternative treatment options, such as sublobar resections, SBRT or, as a last resort, computed tomography-guided ablation, could also be considered.25,29,30 Therefore, patients in the intermediate-risk category are the ones who are most likely to benefit from a multidisciplinary discussion at tumor boards or evaluation at a multidisciplinary outpatient clinic, corresponding to approximately 60 patients per year at our institution. Our study could not identify clinically relevant variables associated with survival or selection for SBRT in the intermediate category. This could, in part, result from the reduced power of the analysis related to the smaller sample size and number of events, as well as the presence of unknown clinical, psychological, or socioeconomic variables. Consequently, we are currently prospectively collecting data on decision-making and survival in these intermediate patients to address this remaining knowledge gap.
Recently, a prospective cohort study from the International Early Lung Cancer Action Program and Initiative for Early Lung Cancer Research on Treatment, as well as a large retrospective cohort study in more than 30.000 patients from the National Cancer Database compared the long-term OS between SBRT and surgery, demonstrating an association between SBRT and reduced OS on Cox regression analysis.5,6 Although our results confirmed that OS in patients receiving SBRT was lower compared to MIS, our multivariable Cox regression model did not identify SBRT as a risk factor for survival. However, it is important to emphasize that the Cox regression analyses in the two large cohort studies did not correct for some important variables influencing OS, such as patient frailty and reduced pulmonary function.11,15,16 These unaccounted factors may explain the difference with our analysis. In this context, our multivariable analysis for referring patients to SBRT specifically identified age, patient frailty and reduced pulmonary function as factors favoring SBRT over MIS. Consequently, in the two large cohort studies, SBRT may have ultimately served as a surrogate for these confounding variables. Our finding of comorbidities,10,11,16,28,31 such as age, patient frailty and pulmonary function, driving OS after SBRT or surgery is supported by a study from Yun at al.7 This study also demonstrated that treatment modality (sublobar resection or SBRT) was not associated with OS when adjusted for age, sex and pulmonary function.7 Supporting this observation, the majority of patients in the SBRT cohorts in our study and other studies died from causes unrelated to recurrence.5,32 The lack of significant prognostic effect of the treatment modality on OS in our study further supports the performance and validity of the prediction model, mimicking the decision-making of the surgeons in our group.
Similarly to findings in the literature,32 our study demonstrated an increased recurrence rate in patients who underwent SBRT. This observation can be related to the lesser extent of oncologic treatment in this patient cohort (i.e., no lymph node dissection), precluding the detection of unexpected lymph node metastasis and subsequent administration of adjuvant therapy.5,33,34 In current literature, patients who underwent SBRT experienced locoregional recurrence in 10% to 13% of the cases and distant recurrence in approximately 14%,3,32,35,36 findings which align with those of our study.
Our study also demonstrated comparable short-term outcomes between patients receiving MIS and SBRT in terms of overall and grade ≥3 complications, as well as 90-day mortality. These findings suggest that the clinical selection process accurately identifies patients suitable for either treatment modality, as the expected postoperative morbidity and mortality after surgery in high-risk patients would have been significantly higher.37–39 These observations further emphasize the precision and efficacy of the clinical judgement and the prediction model. The short-term outcome of the SBRT patients in our study align with other reports, where radiation-induced toxicity ranged from 20%−78% and grade ≥3 complications ranged from 0%−10%.3,40–43 Although our patient population was older, with 10% having larger tumors—known risk factor for development of radiation-induced pneumonitis44,45—the observed incidence of grade 2 or higher radiation-induced pneumonitis of 4.6% in our cohort was consistent with the literature, where incidences ranged from 1%−9%.8,9,41,44 However, it remains crucial to acknowledge that the risk of radiation-induced pneumonitis can increase up to as high as 27%, e.g. with repeated course of SBRT.46 In such cases, stereotactic proton therapy or computed tomography-guided ablation could be potential alternative approaches.30,45 The 90-day mortality for SBRT was also consistent with results observed in the literature, ranging from 0%−1.7%.3,6,46
We must acknowledge that this study is retrospective in nature, which introduces the possibility of selection and reporting biases. However, by utilizing a prospectively maintained surgical database and rigorously reviewing the referred SBRT patients adhering to similar standards, we mitigated these biases to the best of our abilities. Furthermore, the results from this study reflect the patient demographic characteristics and experience of a high-volume referral cancer center, which should be taken into consideration when interpreting the results. Although this distribution may not reflect the distribution of the general population, we sought to develop an unbiased decision model based on clinical variables, rather than racial, ethnic, or religious background. This model is intended to assist, and not replace, decision-making. Ultimately, the final decision should rest with the treating physician or surgeon and, above all, the patient. Additionally, our median follow-up period of 1.75 years may be perceived as relatively short for accurately assessing OS. However, given that 18% of the SBRT patients died over a 3-year period, we can reasonably conclude that a sufficient number of events have occurred to draw meaningful conclusions regarding OS. Furthermore, given that 80% of recurrences typically manifest within 2 years of treatment, we expect minimal additional recurrent cases in both groups and thus closely reflecting the current incidences in our population. Since our study aimed to identify factors associated with the selection for SBRT, and the outcomes were primarily assessed to evaluate the model’s performance, a matching approach was deemed inappropriate as it could confound the identification of risk factors. In the context of treatment outcomes, the comparison of treatment outcomes is currently underway in the STABLE-MATES and SORT trials. While we did not conduct an external validation, primarily due to the lack of comprehensive external databases including both surgery and SBRT patients, we are planning a prospective multicenter study to assess the model’s performance in clinical practice.
The major strength of our study is that this is the first study to construct an accurate prediction model for referring patients with stage I NSCLC to SBRT, materializing the clinical decision-making process and judgement of 12 expert thoracic surgeons. Moreover, the model leverages data from a large patient cohort with a comprehensive list of collected variables. As a result, we were able to identify 3 distinct risk categories, offering practical guidance for decision-making in clinical settings and facilitating the selection of patients for multidisciplinary evaluation. While we explored the Youden index, we found that this cutoff probability primarily offered high sensitivity. However, since clinical practice aims to offer surgery to as many patients as can tolerate it, we focused on increasing specificity to reduce false positive rates. Nonetheless, this adaptability highlights that the continuous probability scale of the nomogram enables other institutions to adjust the cutoffs for low- and high-risk categories according to their practices and outcomes, facilitating a flexible institutional-oriented implementation of the prediction model.

Conclusion
Our results demonstrated the feasibility of translating clinical expertise into a robust predictive model for guiding the referral of clinical stage I NSCLC patients to SBRT, effectively categorizing them into three distinct risk groups. Patients in the intermediate category could benefit from a multidisciplinary decision-making process. Despite the higher recurrence and lower survival rates observed following SBRT, the treatment modality was not significantly associated with overall survival, which supports the precision and efficacy of the clinical judgement and subsequently the prediction model.

Supplementary Material
SupplementSupplemental Digital Content 1
Supplemental Table 1
Supplemental Table 2
Supplemental Table 3
Supplemental Table 4
Supplemental Table 5
Supplemental Figure 1
Supplemental Figure 2
Supplemental Figure 3