Two Interventions on Pathologic Nodal Staging in a Population-Based Lung Cancer Resection Cohort

Abstract

BACKGROUND:

Despite its prognostic importance, poor pathologic nodal staging of lung cancer prevails. We evaluated the impact of two interventions to improve pathologic nodal staging.

METHODS:

We implemented a lymph node specimen collection kit to improve intraoperative lymph node collection (surgical intervention) and a novel gross dissection method for intrapulmonary node retrieval (pathology intervention) in non-randomized stepped-wedge fashion, involving 12 hospitals and seven pathology groups. We used standard statistical methods to compare surgical quality and survival of patients who had: neither (Group 1), pathology only (Group 2), surgical only (Group 3), and both (Group 4) interventions.

RESULTS:

Of 4,019 patients from 2009-2021, 50%, 5%, 21% and 24%, respectively, were in Groups 1-4. Rates of non-examination of lymph nodes and non-examination of mediastinal lymph nodes were: 11%, 9%, 0% and 0%; 29%, 35%, 2% and 2% respectively in Groups 1-4 (p<0.0001). Rates of attainment of American College of Surgeons Operative Standard 5.8 were: 19%, 22%, 70%, 83%; and International Association for the Study of Lung Cancer definition of ‘complete resection’, 14%, 21%, 53%, 61% (p<0.0001).

Compared to Group 1, adjusted hazard ratios for death were: Group 2, 0.93 (95% confidence interval: 0.76-1.15); Group 3, 0.91 (0.78-1.03); Group 4, 0.75 (0.64-0.87). Compared to Group 2, Group 4 adjusted hazard ratio was 0.72 (0.57-0.91); compared to Group 3, it was 0.83 (0.69-0.99). These relationships remained after excluding wedge resections.

CONCLUSIONS:

Combining a lymph node collection kit with a novel gross dissection method significantly improved pathologic nodal evaluation and survival.

For patients who undergo curative-intent surgery for non-small cell lung cancer (NSCLC), the pathologic nodal stage is a major determinant of subsequent management and prognosis.^1–3 Lymph node involvement connotes a worse prognosis, and eligibility for adjuvant therapy.^2–5 Accurate pathologic nodal staging is increasingly important as more effective adjuvant therapy options become available. It depends on the combination of retrieval of hilar, mediastinal and intrapulmonary nodes, thorough examination of all specimens, and complete, accurate reporting of the final pathologic findings.⁶

Deficits in these processes have been widely reported, with adverse impact on patient survival. For example, 11-18% of ‘pathologic node-negative resections have no examined lymph nodes;^7–9 30-60% of ‘mediastinal node-negative’ resections have no examined mediastinal lymph nodes;^10,11 80% of lobectomy specimens have discarded intrapulmonary nodes, 30% of which have metastasis, including 12% of ‘pathologic node-negative’ resections.¹² Each of these quality deficits is associated with poor survival.^7–13 Examination of surgical and pathologic quality in recent clinical trials reveals similar problems.^14,15

Lung cancer surgery with a lymph node specimen collection kit markedly improves patient survival, but may not redress the incomplete retrieval of intrapulmonary lymph nodes from resection specimens.^16,17 A novel gross dissection protocol significantly improved intrapulmonary node retrieval in pilot studies, but the survival impact is unknown.¹⁸ To an ongoing implementation of a lymph node kit by surgery teams, we introduced a novel gross dissection method for implementation by pathology teams.¹⁹ Hypothesizing that the combination of process interventions would be more effective than either alone, we compared the outcomes of surgery in a population-based observational cohort exposed or not exposed to each.

MATERIAL AND METHODS

The Mid-South Quality of Surgical Resection cohort.

With the permission of the Institutional Review Boards of all participating institutions, including a waiver of the informed consent requirement for this low-risk quality improvement study, we abstracted details from the medical records of lung cancer surgery patients at all eligible hospitals within five contiguous Dartmouth Hospital Referral Regions in Arkansas, Mississippi and Tennessee from 2009 onward. Eligible hospitals had five or more annual curative-intent resections and were structurally diverse in teaching status, case-volume, rurality and surgeon characteristics.^16,17,20

Surgical intervention.

We encouraged surgery teams to use a specimen collection kit for all curative-intent resections. These pre-labelled kits were designed to improve the intraoperative retrieval and labelling of lymph nodes, and the security of specimen transfer. Between 2010 and 2014, we pilot-tested the kit at three Metropolitan Memphis hospitals.²¹ From January 1, 2015 onward, we sequentially introduced it at all participating institutions using a non-randomized stepped-wedge design in which institutions were assigned into three relatively homogeneous implementation sub-groups based on case-volume, teaching status and rurality (Supplementary Figure 1).¹⁶

PATHOLOGY INTERVENTION.

We developed a novel gross dissection protocol to improve the retrieval of intrapulmonary lymph nodes. It involves a series of blunt dissections from the hilum radially outwards, from the central to the peripheral bronchial tree, with particular attention to abluminal areas of bronchial bifurcation where lymph nodes aggregate.¹⁸ We pilot-tested this novel gross dissection method in one pathology group from 2012 to 2016 and continued the approach within this pathology group to date, but began the staggered implementation across the remaining six groups in December 2021. For this, we developed a training video (Supplementary Video) and line diagrams of the process (Supplementary Figure 2). We had planned for hands-on training workshops to teach the novel gross dissection method, but relied solely on the videos and line diagrams during the COVID-19 pandemic. As of this analysis two of seven pathology groups, servicing three of the 12 institutions, had implemented the novel dissection method.

PATIENT SELECTION AND INTERVENTION GROUPS.

For this analysis, we excluded patients with clinical M1 disease, whose prognosis is greatly influenced by distant metastasis; recipients of neoadjuvant therapy, which can confound lymph node yield; and included only the first operation for patients who had more than one episode of lung cancer surgery. We assigned the remaining patients into four sub-groups according to exposure to the interventions: Group 1 (neither); Group 2 (pathology intervention only); Group 3 (surgical intervention only); Group 4 (both).

SURGICAL QUALITY BENCHMARKS.

We evaluated markers of poor and good quality in each of the four intervention groups. Poor quality markers included the rates of pathologic NX,^7,9 non-examination of mediastinal nodes,^10,11 non-examination of nodes from specific stations- 11 – 14, 10, 7 and 4.^22,23 Markers of good quality included the rates of attainment of American College of Surgeons Operative Standard 5.8,²⁴ the International Association for the Study of Lung Cancer (IASLC) definition of complete resection,²⁵ and the National Comprehensive Cancer Network (NCCN) guidelines for lung cancer surgery quality.³

Operative Standard 5.8 requires sampling of at least one ‘hilar’ node (actually, stations 10 – 14) and nodes from at least three named or numbered mediastinal stations.²⁴ The IASLC defines ‘complete resection’ as the combination of negative resection margins, systematic or lobe-specific nodal dissection, no extracapsular extension of nodal metastasis, and no metastasis in the highest mediastinal station. Incomplete resections are defined as resections with any positive margins or positive pleural or pericardial effusions. ‘Uncertain resections’ have negative margins, but fail to meet one or more of the other criteria for complete resection.²⁵ The NCCN defines quality as anatomic resection, negative margins, examination of at least one N1 lymph node, and a minimum of three mediastinal stations.³

STATISTICAL METHODS.

We summarized patient-level demographic, clinical, and treatment characteristics with median, range and interquartile range, frequencies and proportions. We compared summary statistics across all four intervention groups with chi-squared (Fisher’s exact if sample sizes were small) and Kruskal-Wallis tests. We initially examined survival with Kaplan-Meier plot and Log-rank test, and further quantified it with hazard ratios and 95% confidence intervals (CI) from Cox proportional hazard models. We calculated crude and adjusted hazard ratios. Adjustment variables were age, sex, race, histology, clinical stage, technique of surgery, and extent of resection.

In secondary analyses, we excluded wedge resections because of limited opportunities for intrapulmonary lymph node retrieval and a greater likelihood of poor overall health and higher risk for competing causes of mortality. For the evaluation of clinical to pathologic nodal upstaging, we counted pNX as pN0, given the typical management of such patients. We performed all analyses using SAS 9.4 (2013, SAS Institute Inc., Cary NC) and set significance levels to alpha=0.05. Our report followed Strengthening the Reporting of Observational Studies guidelines.

RESULTS

COHORT CHARACTERISTICS.

From January 1, 2009 to August 31, 2021, there were 4,019 eligible resections from the 12 eligible hospitals (CONSORT, Figure 1) of which 1,991 (50%) were in Group 1, 212 (5%) in Group 2, 837 (21%) in Group 3, and 979 (24%) in Group 4. The cohort included work by 59 surgeons and 117 pathologists (Supplementary Table 1). Differences in patient characteristics between the groups mostly seemed of little significance (Table 1). Adenocarcinoma was most common in Group 4 (59%), the other groups ranged from 51% (Group 2) to 56% (Group 3); clinical T-category was advanced (T3 or T4) in 16%, 17%, 12% and 13% of patients from Group 1 to Group 4 in order; and clinical N0 was highest in Groups 3 (92%) and 4 (93%) compared to 87% in Group 1 and 88% in Group 2; however, poorly or undifferentiated histology ranged from 24% in Group 3 to 42% in Group 2 (Table 1; Supplementary Table 2).

Figure 1.

Cohort selection.

Table 1.

Patient demographic, disease and treatment characteristics.

Variablesa	Groups
	1: N=1991b	2: N=212c	3: N=837d	4: N=979e
Race *
White	1527(77)	170(80)	692(83)	751(77)
Black	435(22)	41(19)	138(16)	214(22)
Asian	11(1)	1(0)	2(0)	8(1)
Other	18(1)	0(0)	5(1)	6(1)
Age ** f	67(61-74)	69(63-75)	67(61-73)	69(64-75)
Sex **
Male	1151(58)	108(51)	450(54)	484(49)
Female	840(42)	104(49)	387(46)	495(51)
Insurance ****
Medicare	974(49)	118(56)	366(44)	500(51)
Medicaid	251(13)	28(13)	148(18)	156(16)
Commercial	689(35)	64(30)	292(35)	302(31)
Self-Insured	77(4)	2(1)	31(4)	21(2)
Number of Co-morbidities ** f	2(1-3)	2(2-3),(0-7)	2(1-3),(0-8)	2(2-3),(0-7)
Histology ****
Adenocarcinoma	1030(52)	108(51)	469(56)	580(59)
Squamous	697(35)	68(32)	283(34)	298(30)
Other	264(13)	36(17)	85(10)	101(10)
Grade ****
Well	233(12)	26(12)	112(13)	115(12)
Moderately	839(42)	80(38)	340(41)	402(41)
Poorly/Undifferentiated	610(31)	88(42)	199(24)	387(40)
Not Reported	309(16)	18(8)	186(22)	75(8)
Aggregate Clinical Stage ****
Stage I	1260(63)	142(67)	618(74)	741(76)
Stage II	344(17)	32(15)	132(16)	138(14)
Stage III	275(14)	28(13)	57(7)	83(8)
Unknown	112(6)	10(5)	30(4)	17(2)
Pathologic Aggregate Stage **
Stage 1	1236(62)	130(61)	505(60)	659(67)
Stage II	431(22)	41(19)	196(23)	189(19)
Stage III	315(16)	38(18)	134(16)	130(13)
Stage IV	9(0)	3(1)	2(0)	1(0)
Change in nodal staging *****
Upstaging**	278(14)	30(14)	142(17)	139(14)
cN0 → pN1	159(57)	14(47)	78(55)	78(56)
cN0 → pN2/3	101(36)	14(47)	54(38)	57(41)
cN1 → pN2/3	17(6)	2(7)	10(7)	4(3)
cN2 → pN3	1(<1)	0(0)	0(0)	0(0)
No change	1547(78)	165(78)	665(79)	803(82)
Down staging	157(8)	15(7)	18(2)	33(3)
Unknown	9(<1)	2(1)	12(1)	4(<1)
Extent of resection *****
Pneumonectomy	119(6)	11(5)	24(3)	23(2)
Bilobectomy	102(5)	3(1)	41(5)	23(2)
Lobectomy	1391(70)	173(82)	726(87)	867(89)
Segmentectomy	105(5)	9(4)	22(3)	42(4)
Wedge	274(14)	16(8)	24(3)	24(2)

^*p>0.05;

^**p<0.05;

^***p<0.01;

^****p<0.001;

^*****p<0.0001;

^anumbers (%) unless otherwise stated;

^bno intervention;

^cpathology intervention only;

^dsurgical intervention only;

^eboth interventions;

^fmedian (interquartile range);

^**Upstaging frequency is the denominator for breakdown of the different combinations of upstaging.

The pattern of care differed between groups: preoperative positron emission tomography-computer tomography (PET-CT) scans were performed in 79% of Group 1 patients, compared to 88 – 91% of patients in the other groups (p<0.001, Figure 2a); invasive staging was performed in only 13% and 14% of patients in Groups 1 and 3, and 24% and 22% of patients in Groups 2 and 4, respectively (p<0.001, Figure 2b). The technique of surgery showed the greatest difference: 69% of patients in Group 1 had open thoracotomy, compared to 32% of Group 4; whereas Group 4 had the highest proportion of robotically-assisted thoracoscopic resection (60%), compared to other groups that ranged from 14% in Group 1, to 44% in Group 2 (p<0.0001, Figure 2c–e).

Figure 2.

Use of positron emission tomography-computer tomography (PET-CT), invasive staging, and surgical technique. Groups: 1-no intervention; 2-pathology intervention only; 3-surgical intervention only; 4-both interventions. Bars indicate significantly different rates across intervention groups after Tukey adjustment for multiple comparisons.

SURGICAL ONCOLOGIC QUALITY.

The quality of nodal evaluation differed significantly across intervention groups by every metric. Poor quality measures were significantly more common in Groups 1 and 2 than in Groups 3 and 4 (Figure 3). For example, the pNX rates were 11%, 9%, 0% and 0% in order from Groups 1-4; rates of non-examination of mediastinal lymph nodes were 29%, 35%, 2% and 2%, respectively. However, rates of non-examination of stations 11-14 were 33% versus 24% versus 19% versus 20%; station 10, 48%, 45%, 17% and 8%; station 7, 66%, 72%, 26% and 11%; and non-examination of station 4, 77%, 79%, 58% and 53% (p<0.0001 for all group-wide comparisons; Supplementary Table 3).

Figure 3.

Rates of non-examination across all intervention groups. Groups: 1-no intervention; 2-pathology intervention only; 3-surgical intervention only; 4-both interventions. Bars indicate significantly different rates across intervention groups after Tukey adjustment for multiple comparisons.

The aggregate metrics of good quality also varied significantly: Operative Standard 5.8 was achieved in 19%, 22%, 70% and 83% of resections from Group 1-4, respectively; IASLC complete resection rates were 14%, 21%, 53% and 61%; and attainment of the aggregate NCCN quality criteria were 28%, 32%, 73% and 86%, respectively; p<0.0001 for all group-wide comparisons (Figure 4; Supplementary Table 3). All groups had a 14% rate of nodal upstaging except Group 3, which had a rate of 17% (p<0.0001). In the upstaged subsets, the rates of upstaging from cN0 to pN1 were 57%, 47%, 55%, and 56%, respectively; cN0 to pN2 or pN3 was 36%, 47%, 38%, and 41%; and from cN1 to pN2 or pN3 was 6%, 7%, 7%, and 3% (p<0.0001).

Figure 4.

Quality benchmark attainment rates. Groups: 1-no intervention; 2-pathology intervention only; 3-surgical intervention only; 4-both interventions. NCCN- National Comprehensive Cancer Network; ACS CoC - American College of Surgeons Commission on Cancer; IASLC - International Association for the Study of Lung Cancer (IASLC) ‘R-Factor’ definitions. Bars indicate significantly different rates across intervention groups after Tukey adjustment for multiple comparisons.

Patterns of non-examination and metrics of good quality persisted when wedge resections were excluded. For example, Groups 1 and 2 had higher pNX rates than Groups 3 and 4- 3% versus 5% versus 0% and 0% (p<0.0001); non-examination of mediastinal nodes was 23% versus 33% versus 2% and 2% (p<0.0001), and non-examination of station 10 was 42% versus 41% versus 16% and 7% (p<0.0001; Supplementary Figure 2; Supplementary Table 4). Non-examination of stations 11-14 was significantly more in Group 1 (24%), compared to Groups 2, 3 and 4 (18%, 17%, and 18%, respectively, p<.0001). Groups 3 and 4 had significantly higher rates of attaining aggregate metrics of good quality, such as NCCN, Operative Standard 5.8, and IASLC complete resection, with absolute differences greater than 40% (Supplementary Figure 3 and Table 3). Nodal upstaging rates were 16%, 15%, 17%, and 14% across Groups 1-4 (p<0.0001), despite the greater use of preoperative PET-CT and invasive staging in groups 3 and 4.

SURVIVAL COMPARISONS.

The median follow-up time for the whole cohort was 3.1 years (interquartile range: 1.3-5.8), but median follow up time varied between intervention groups: 4 years (1.6-7.4) for Group 1; 3.3 (1.3-6.4) for Group 2; 2.3 (0.8-3.9) for Group 3, and 2.7 (1.1-4.7) for Group 4. Patients in Group 4 (combined intervention) had significantly better survival than patients in all the other groups (Log-rank p<0.0001; Figure 5). Compared to patients in Group 1 (non-intervention), patients in Groups 3 (kit only) and 4 had significantly lower crude hazards for death, with hazard ratios 0.8 (95% CI: 0.7-0.91) and 0.62 (0.54-0.71), respectively (Table 2). This pattern held for Group 4 compared to Group 1, even after adjusting for confounding variables (0.75 [0.64-0.87], ‘HR¹’ in Table 2). The combination of interventions was associated with lower hazard than the separate interventions (adjusted hazard ratio 0.72 [0.57-0.9] versus Group 2 and 0.83 [0.69-0.99] versus Group 3). The differences remained significant after excluding wedge resections (‘HR²’, Table 2) but not after further adjustment (‘HR³’, Table 2).

Figure 5.

Kaplan-Meier plot across all intervention groups: 1-no intervention; 2-pathology intervention only; 3-surgical intervention only; 4-both interventions.

Table 2.

Crude and adjusted hazard ratios (95% confidence intervals) of comparing each intervention group.

Intervention Groups†	Crude Hazard Ratio	HR1	HR2	HR3
Group 2 vs Group 1 (ref)	0.93(0.76, 1.15)	1.04(0.84, 1.28)	0.93(0.75, 1.15)	0.99(0.8, 1.24)
Group 3 vs Group 1	0.8(0.69, 0.91)	0.91(0.78, 1.03)	0.82(0.71, 0.94)	0.89(0.77, 1.03)
Group 4 vs Group 1	0.62(0.54, 0.71)	0.75(0.64, 0.87)	0.65(0.56, 0.74)	0.76(0.65, 0.89)
Group 3 vs Group 2 (ref)	0.86(0.68, 1.08)	0.86(0.68, 1.09)	0.89(0.7, 1.12)	0.89(0.70, 1.14)
Group 4 vs Group 2	0.67(0.53, 0.84)	0.72(0.57, 0.91)	0.7(0.55, 0.89)	0.76(0.60, 0.97)
Group 4 vs Group 3 (ref)	0.78(0.66, 0.93)	0.83(0.69, 0.99)	0.79(0.67, 0.94)	0.85(0.71, 1.03)

^†Bolded hazard ratios indicate α<0.05 level after Tukey adjustments for pairwise comparisons; HR¹ -adjusted for age, sex, race, histology, clinical stage, technique of surgery, extent of resection; HR² - wedge resections excluded with no adjustments; HR³ -wedge resections excluded and adjusted for age, sex, race, histology, clinical stage, technique of surgery, extent of resection.

COMMENT

The combination of a lymph node specimen collection kit and a novel lung gross dissection method significantly improved pathologic nodal staging and overall survival in this diverse, prospective population-based cohort. The combination of interventions had greater impact than individual interventions and was associated with significantly lower hazard for death. The results were sustained in additional analysis excluding wedge resections.

The prelude to this project was demonstration of discordance between surgeons, pathologists and a blinded objective reviewer in categorizing the lymphadenectomy procedure performed, with concordance solely in cases the surgeon indicated had no lymph node examination.²⁶ We developed the ‘chain of responsibility’ hypothesis, conceptualizing pathologic nodal staging as a team activity, involving surgery and pathology teams and the connection between them. We deconstructed the key activities as surgical retrieval of hilar and mediastinal nodes (directly inaccessible to pathology teams, thus primarily the responsibility of the surgery team); accurate specimen labelling (surgery team’s responsibility); secure specimen transfer (hand-off between teams); retrieval of intrapulmonary lymph nodes (typically a pathology team responsibility, although surgeons sometimes take this on); thorough specimen examination (pathologist’s responsibility); and complete, accurate reporting of all findings (pathologist’s responsibility).²⁷

The lymph node kit targets events in the Operating Room and the hand-off to the pathology team, its use significantly improved concordance between surgeon, pathologist and two blinded independent reviewers.²⁸ We developed the novel gross dissection intervention after finding that approximately 60% of intrapulmonary lymph nodes were discarded in the conventional gross dissection of lobectomy specimens, including 12% of pN0 cases with discarded intrapulmonary lymph node metastasis, with associated adverse survival for patients with discarded lymph node metastasis.^12,13 It improved the retrieval of intrapulmonary lymph nodes in pilot studies.^18,19

The problems we address are prevalent, even in clinical trial cohorts. Only 53% of resections in the contemporary ALCHEMIST adjuvant therapy trial met the NCCN quality standard.¹⁵ For context, although higher than Groups 1 and 2 of our cohort (27% and 30%, respectively), it is much lower than Groups 3 (79%) and 4 (92%), which both received the surgical intervention. A re-examination of the mediastinal lymph node dissection arm of ACOSOG Z0030 revealed a relatively low intrapulmonary lymph node count, and a strong direct association between N1 lymph node count and survival.^14,29

LIMITATIONS.

Our pragmatic non-randomized stepped-wedge implementation design left imbalances between the groups. Unrelated secular changes may have influenced the patterns of care between groups, for example in the use of PET-CT, invasive staging, and surgical techniques. Implementation of the two interventions began at different times, with potentially different exposure to secular changes. Surgeon and team proficiency may have independently improved over time. Nodal upstaging rates did not differ materially between cohorts, indicating that upstaging is a poor indicator of quality in this context. Improvements in adjuvant therapy may have a confounding effect, but these occurred from 2020 onward and probably had little impact.^4,5 The COVID-19 pandemic disrupted our pathology implementation schedule, precluded planned in-person training workshops. Only two of seven pathology groups had implemented at the time of this analysis, reducing the sample size of Groups 2 and 4. We retroactively applied Operative Standard 5.8.²⁴ Finally, we can only hypothesize the causes of survival differences, including greater use of effective adjuvant therapy, and intrinsic oncologic benefit from the resection of oligo-metastatic nodal disease. We plan to examine the relative impact of the adjuvant therapy versus oligo-metastatic disease hypotheses, in future studies measuring circulating tumor DNA.³⁰

Nevertheless, these safe and relatively simple process changes were associated with better quality and survival in diverse, community-based healthcare settings. Limiting the confounding survival effect of variable quality care should, in theory, facilitate research into biologic drivers of outcomes differences.^30,31 Such research might explore the comprehensive biomarker profile of lung cancer in populations with disparate outcomes, biomarkers of minimal residual disease such as circulating tumor DNA; and host immune responses.^32,33 By these means, the field might accelerate toward more accurate characterization of person-level risk-stratification and personalization of adjuvant treatment selection.