Abstract
Treatment response and prognosis estimation in advanced pulmonary
adenocarcinoma are challenged by the significant heterogeneity of the
disease. The current Response Evaluation Criteria in Solid Tumors
(RECIST) criteria, despite providing a basis for solid tumor response
evaluation, do not fully encompass this heterogeneity. To better
represent these nuances, we introduce the intertumoral heterogeneity
response score (THRscore), a measure built upon and expanding the
RECIST criteria. This retrospective study included patients with 3–10
measurable advanced lung adenocarcinoma lesions who underwent
first‐line chemotherapy or targeted therapy. The THRscore, derived from
the coefficient of variation in size for each measurable tumor before
and 4–6 weeks posttreatment, unveiled a correlation with patient
outcomes. Specifically, a high THRscore was associated with shorter
progression‐free survival, lower tumor response rate, and a higher
tumor mutation burden. These associations were further validated in an
external cohort, confirming THRscore's effectiveness in stratifying
patients based on progression risk and treatment response, and
enhancing the utility of RECIST in capturing complex tumor behaviors in
lung adenocarcinoma. These findings affirm the promise of THRscore as
an enhanced tool for tumor response assessment in advanced lung
adenocarcinoma, extending the RECIST criteria's utility.
Keywords: intertumoral heterogeneity, intertumoral heterogeneous
response, lung adenocarcinoma, RECIST criteria
__________________________________________________________________
In the clinical management of advanced pulmonary adenocarcinoma, a
commonly observed phenomenon is the intertumoral heterogeneous response
(THR), which refers to the significant variation in treatment responses
among different tumors within the same patient. We introduce the
intertumoral heterogeneity response score (THRscore), a measure built
upon and expanding the RECIST criteria, to evaluate THR effectively,
stratifying patients by progression risk.
graphic file with name MCO2-5-e493-g004.jpg
1. INTRODUCTION
In the clinical management of advanced pulmonary adenocarcinoma, a
commonly observed phenomenon is the intertumoral heterogeneous response
(THR), which refers to the significant variation in treatment responses
among different tumors within the same patient.[80] ^1 , [81]^2 ,
[82]^3 , [83]^4 , [84]^5 This variability in response can be attributed
to a range of factors, including genetic variations among tumors that
affect drug sensitivity, differences in the tumor microenvironment that
impact drug delivery and efficacy, diverse immune responses within each
lesion, and selective pressures induced by treatment itself that lead
to the survival of drug‐resistant cells.[85] ^6 , [86]^7 , [87]^8 ,
[88]^9 Such THR, adds considerable complexity to treatment
decision‐making and prognosis estimation, posing significant challenges
to oncologists.[89] ^10 , [90]^11 Consequently, a thorough
understanding and assessment of this heterogeneity is pivotal for
developing effective treatment strategies and accurately predicting
disease prognosis.
Response Evaluation Criteria in Solid Tumors (RECIST),[91] ^12 widely
adopted as the standard for assessing treatment response in solid
tumors, focuses on measuring the sum of diameters of selected
measurable lesions. Its modified version, RECIST 1.1,[92] ^13 is
extensively used in clinical practices and trials worldwide. However,
this approach has inherent limitations in advanced pulmonary
adenocarcinoma, particularly in its ability to reflect the complexities
of tumor heterogeneity. RECIST's emphasis on overall tumor burden may
mask the varied responses exhibited by different lesions within the
same patient. Such limitations highlight a critical gap in accurately
assessing the full spectrum of tumor behavior and responses, especially
in cases with multiple distinct lesions. Evidence suggests that even if
the RECIST criteria indicates partial response (PR) or stable disease
(SD), there may still be disease progression and a shortened
progression‐free survival (PFS).[93] ^14 , [94]^15 , [95]^16 Although
efforts have been made to develop improved versions such as iRECIST[96]
^17 and PERCIST[97] ^18 to accommodate advancements in immunotherapy
and emerging imaging technologies, they still primarily focus on the
overall tumor burden but neglect heterogeneity of tumor. This gap
underscores the need for more comprehensive assessment tools that can
capture the diverse nature of tumor response.
To bridge this existing gap, we introduce a novel approach, employing
the coefficient of variation (CV)[98] ^19 of all lesion changes as a
measure of THR. The CV, the standard deviation divided by the
arithmetic mean, is the most widely used measure of the extent of trait
variation.[99] ^19 , [100]^20 Indeed, its applications are already
proven in numerous fields such as finance, meteorology, and engineering
where it is effectively used to measure dispersion of data points.[101]
^21 , [102]^22 , [103]^23 In our study, we hypothesize that by applying
the CV, we can capture the THR within a patient and gain insights into
the heterogeneity of treatment response.
Given this background, we aim to establish a new, clinically convenient
system for assessing THR, which both builds upon and expands the
existing RECIST framework. Central to this is the derivation of an
intertumoral heterogeneity response score (THRscore), projected to
correlate with PFS and tumor response according to RECIST 1.1.
Moreover, we utilize next‐generation sequencing (NGS) to further delve
into potential mechanisms that link THR with clinical outcomes. This
comprehensive and integrated approach promises to enhance our
understanding and management of tumor heterogeneity.
2. RESULTS
2.1. Patient characteristics
In this retrospective study, we initially screened 2082 treatment‐naïve
patients in Cohort One, of which 174 met the inclusion criteria after
excluding those who did not fulfill our study requirements
(Figure [104]1A). The included patients had a median age at diagnosis
of 59 years (range, 36−84 years); 105 (60%) were male and 58 (33%) were
current or former smokers. Based on the time of treatment, patients
were assigned to the discovery set (treated between June 2018 and
December 2020) or the validation set (treated between January 2016 and
May 2018). Among these included patients, 101 (58%) underwent
platinum‐based doublet chemotherapy, 73 (42%) received targeted
therapies, and a total of 57 (33%) patients had epidermal growth factor
receptor (EGFR) mutation and received EGFR tyrosine kinase inhibitor.
Significant differences were observed between the two set regarding
treatment regimens, presence of brain metastases, Eastern Cooperative
Oncology Group Performance Score (ECOG‐PS), number of lesions, and the
presence of driver genes. Table [105]1 provides a detailed overview of
patient characteristics.
FIGURE 1.
FIGURE 1
[106]Open in a new tab
Study scheme. (A) Flowchart illustrating the data collection and sample
allocation process. Cohort One comprises eligible patients assigned to
either the discovery set (June 2018 to December 2020) or the validation
set (January 2016 to May 2018) from Fujian Cancer Hospital (FCH).
Cohort Two includes eligible patients for external validation (January
2010 to December 2020) from Fujian Medical University Union Hospital
(FMUUH). (B) Schematic representation of the method used to measure the
intertumoral heterogeneity response following systemic treatment.
Simulating the heterogeneity of lesion changes in patients during
systemic treatment: blue, enlargement; yellow, reduction; red, no
change. The measurement methods for lesions and lymph nodes were
conducted according to the Response Evaluation Criteria in Solid Tumors
version 1.1 (RECIST) guidelines. Specifically, the maximum diameter was
utilized for assessing lesions, while the shortest diameter (short
axis) was employed for the evaluation of metastatic lymph nodes. (C)
Main analysis workflow.
TABLE 1.
Baseline demographic and clinical characteristics of the patient
sample.
Full analysis set Discovery set Validation set
(N = 174) (N = 85) (N = 89) p
Age (years)
Median [range] 59 [26, 80] 59 [33, 80] 59 [26, 75] 0.597
Sex
Female 69 (39.7%) 30 (35.3%) 39 (43.8%) 0.320
Male 105 (60.3%) 55 (64.7%) 50 (56.2%)
Smoking 0.789
Absence 116 (66.7%) 58 (68.2%) 58 (65.2%)
Presence 58 (33.3%) 27 (31.8%) 31 (34.8%)
ECOG‐PS 0.029
1 162 (93.1%) 75 (88.2%) 87 (97.8%)
2 12 (6.9%) 10 (11.8%) 2 (2.2%)
Brain metastasis 0.017
Absence 128 (73.6%) 70 (82.4%) 58 (65.2%)
Presence 46 (26.4%) 15 (17.6%) 31 (34.8%)
Bone metastasis 0.181
Absence 114 (65.5%) 51 (60.0%) 63 (70.8%)
Presence 60 (34.5%) 34 (40.0%) 26 (29.2%)
Adrenal metastasis 0.086
Absence 149 (85.6%) 69 (81.2%) 80 (89.9%)
Presence 25 (14.4%) 16 (18.8%) 9 (10.1%)
Liver metastasis 0.901
Absence 149 (85.6%) 72 (84.7%) 77 (86.5%)
Presence 25 (14.4%) 13 (15.3%) 12 (13.5%)
Lesions number
3 125 (71.8%) 56 (65.9%) 69 (77.5%) 0.039
4 42 (24.1%) 22 (25.9%) 20 (22.5%)
5 4 (2.3%) 4 (4.7%) 0 (0%)
6 3 (1.7%) 3 (3.5%) 0 (0%)
Targetable driver genes[107] ^a 0.008
Mutation 100 (57.5%) 59 (69.4%) 41 (46.1%)
Wild type 60 (34.5%) 21 (24.7%) 39 (43.8%)
Unknown 14 (8.0%) 5 (5.9%) 9 (10.1%)
Treatment 0.007
Chemotherapy 101 (58.0%) 40 (47.1%) 61 (68.5%)
Targeted therapy 73 (42.0%) 45 (52.9%) 28 (31.5%)
[108]Open in a new tab
Eligible patients were assigned to discovery set (between June 2018 and
December 2020) or validation set (between January 2016 and May 2018).
Abbreviation: ECOG‐PS, Eastern Cooperative Oncology Group Performance
Score.
^^a
Targetable driver genes include EGFR, ALK, ROS1, RET, BRAF, MET, and
HER2.
2.2. THRscore as an independent predictor of treatment efficacy
In the discovery dataset (n = 85), the optimal cutoff for predicting
PFS was determined to be 0.46 by maximally selected rank statistics
(see Methods, Figure [109]S1). Based on this cutoff, 51 (60%) and 34
(40%) were considered THRscore^low and THRscore^high, respectively. The
median PFS of the THRscore^low versus THRscore^high groups was 15.6 and
5.4 months, respectively (hazard ratios [HR], 3.40; 95% confidence
interval [95% CI], 1.76−6.58; p < 0.001; Figure [110]2A). Similar
findings were obtained on the validation set (n = 89) whereby the
median PFS for the THRscore^low (n = 46) and THRscore^high (n = 43)
group were 12.9 months and 5.1 months, respectively (HR, 2.69; 95% CI,
1.59−4.55; p < 0.001; Figure [111]2B).
FIGURE 2.
FIGURE 2
[112]Open in a new tab
Association between intertumoral heterogeneity response score
(THRscore) and prognosis. The Kaplan‐Meier curve for progression‐free
survival (PFS) in the discovery set (A) and validation set (B). (C)
Forest plot depicting the different stratification factors used for the
PFS subgroup analyses. The horizontal line represents the 95%
confidence interval (95% CI) for each group, while the vertical dotted
line represents the hazard ratio (HR) for all patients. The vertical
solid line represents an HR of 1. An HR above 1.0 indicates that a high
THRscore is an unfavorable prognostic biomarker. Targetable driver
genes include EGFR, ALK, ROS, RET, BRAF, MET, and HER2. (D)
Kaplan–Meier curves for PFS stratified according to the THRscore for
the patient with stable disease (SD) or partial response (PR),
respectively. All Kaplan–Meier curves were compared using a Log‐rank
statistical test. ECOG‐PS, Eastern Cooperative Oncology Group
Performance Score; Mut, mutation; n, number of patients indicated; W/U,
wild type/unknown.
Univariate and multivariate analyses were performed on the entire data
set, and the THRscore^high was shown to be an independent predictive
factor (HR, 2.48, 95% CI, 1.63−3.77, p < 0.001; Figure [113]S2). In the
subgroup analysis, the increased risk of disease progression in the
patients with THRscore^high was apparent in almost all subgroups,
including the targeted therapy group or chemotherapy group
(Figure [114]2C). Notably, the predictive THRscore was further
confirmed by stratified PR and SD analyses. Specifically, the
THRscore^high was associated with significantly shorter PFS both in
patients with SD (median PFS, 12.9 months vs. 5.1 months; HR, 2.67; 95%
CI, 1.50−4.76; p = 0.001; Figure [115]2D) and with PR (median PFS, 22.0
months vs. 9.5 months; HR, 2.35; 95% CI, 1.09−5.07; p = 0.004;
Figure [116]2D). The dose‐response relationship between the THRscore
levels and PFS was evaluated by restricted cubic splines. While
nonlinear associations were observed, higher levels of THRscore
appeared to be monotonically associated with a higher risk of disease
progression (p < 0.001; Figure [117]S3). We divided the THRscore into
quintiles. The median PFS of the quintiles were 3.9, 6.9, 14.8, 12.7,
and 22.0 months, respectively. The highest versus lowest quintile HR
was 5.25 (95% CI, 2.64−10.44; Table [118]S2).
We have cross‐referenced THRscore with objective response rate (ORR) as
per standard RECIST 1.1. Based on the full dataset, a total of 89
(51.1%) patients were defined to achieve PR, 68 (39.1%) achieved SD,
and 17 (9.8%) achieved progressive disease (PD) (Figure [119]3A). The
median THRscore of patients with PD, SD, and PR was 1.51, 0.58, and
0.27, respectively (p < 0.001; Figure [120]3B). The tumor response rate
of patients with THRscore^high and THRscore^low was 30 and 68%,
respectively (p < 0.001, Fisher's exact test; Figure [121]3C). In
addition, Figure [122]3A displays the posttreatment tumor burden (sum
of the diameters of all lesions), which recorded simultaneously with
the evaluation of the THRscore. Further analysis using the Net
Reclassification Index (NRI)[123] ^24 and the integrated discrimination
index (IDI)[124] ^24 was conducted to compare the predictive value of
THRscore with the RECIST 1.1 response at the 12‐month PFS milestone.
The NRI result of 0.07 (95% CI: 0.01–0.14, p < 0.001) indicates a
significant improvement in patient reclassification for disease
progression risk when including THRscore, compared with using RECIST
1.1 alone. Additionally, the IDI result of 0.33 (95% CI: 0.04–0.42,
p < 0.001) demonstrates THRscore's enhanced ability to discriminate
between progression and nonprogression cases. These results validate
the THRscore as an effective tool for assessing treatment outcomes in
lung adenocarcinoma, offering more nuanced insights than traditional
response criteria and highlighting its potential as a supplementary
measure to RECIST 1.1.
FIGURE 3.
FIGURE 3
[125]Open in a new tab
The association between intertumoral heterogeneity response score
(THRscore) and clinical response in the full dataset. (A) Waterfall
plot of the response of the all lesions according to RECIST for each
patient. Patients were evaluable for response using RECIST1.1. (B)
Boxplots illustrating the association between the THRscore and best
response according to RECIST1.1. The central line on the box plot
indicates the median value, the box plot limits indicate the upper and
lower quartiles, and the whiskers indicate 1.5× the interquartile
range. The Kruskal–Wallis test was used to evaluate the overall
differences across all three groups. (C) The ratio of patients with
partial response (PR), stable disease (SD), and progressive disease
(PD) in the THRscore^high and THRscore^low groups. p < 0.001 by
Fisher's exact test. (D) Time‐dependent ROC analysis for predicting
disease progression. This figure presents the time‐dependent
receiver‐operator characteristic (ROC) analysis, comparing the
predictive accuracy of THRscore with the percent change in the sum of
lesion diameters from baseline in assessing the risk of disease
progression at different time points. The analysis was conducted for
three times intervals: 6, 12, and 18 months. AUC, area under the ROC
curve.
2.3. Association between THRscore and clinical characteristics
The patients in the THRscore^high group were more likely to be smokers
(42.9 vs. 25.7%) and negative for targetable driver genes[126] ^25
(including EGFR, ALK, ROS1, RET, BRAF, MET, and HER2; 49.4 vs. 22.6%)
(p < 0.05, Fisher's exact test; Figure [127]4). There was no
significant correlation between THRscore and other clinical parameters,
including sex, age, ECOG‐PS, number of lesions, brain metastasis,
bone metastasis, adrenal metastasis, and liver metastasis.
FIGURE 4.
FIGURE 4
[128]Open in a new tab
Association between intertumoral heterogeneity response score
(THRscore) and clinical characteristics. Pie charts showing the
distribution of different clinicopatholotgic factors in the
THRscore^high and THRscore^low groups, respectively. The Fisher's exact
test was used to compare the difference in the proportion between the
two groups. Targetable driver genes include EGFR, ALK, ROS, RET, BRAF,
MET, and HER2. n, number of patients indicated; ns, not statistically
significant.
We also analyzed the THRscore as a continuous variable to provide a
more comprehensive overview. The THRscore was a non‐normally
distributed variable with a median value of 0.40 (interquartile range,
0.03−23.71) (Figures [129]S4A and [130]S4B). The results were generally
consistent with categorical variables analyses. Specifically, patients
who smoked, had liver metastasis or tested negative for the targetable
driver genes had significantly higher median THRscore values (0.54 vs.
0.37, 0.57 vs. 0.38, 0.58 vs. 0.36; all p < 0.05; Figure [131]S4C).
2.4. Genomic profiling according to THRscore
Genomic profiles based on NGS were available for a cohort of 61
patients, with 25 being THRscore^high and 36 being THRscore^low.
Similar to the overall population, we observed a significant difference
in PFS between the two groups (HR, 3.09; CI, 1.34−7.11; p = 0.001,
Figure [132]S5). Figure [133]5A presents the genomic landscape derived
from the comprehensive mutational analysis of 61 patients with lung
adenocarcinoma, as conducted using Maftools.[134] ^26 A total of 281
mutation events were identified, affecting 113 genes (Figure [135]S6A).
The types of genetic variants included single nucleotide variations
(SNVs), small insertions (ins), deletions (del), indels, fusions, and
copy number variations (Figure [136]S6B). The most prevalent mutations
were SNVs, which constituted a significant proportion of the
alterations. Among these, the most frequently mutated genes were EGFR
(found in 61% of all samples), TP53 (57%), and KRAS (11%).
FIGURE 5.
FIGURE 5
[137]Open in a new tab
Genomic profiles and potential mechanisms linked to intertumoral
heterogeneity response score (THRscore). (A) Oncoprint of mutations
identified in the 61 patients. Results of targeted DNA sequencing of
lung adenocarcinoma tissues. *Represents a statistically significant
difference (p < 0.05) in the gene mutation rate between the high and
low THRscore groups according to Fisher's exact test. CNV apm, copy
number variation of amplification; del, deletions; SNV, single
nucleotide variations; ins, small insertions. (B) Boxplots showing
differences in the number of genetic variations for each patient
(n = 61) in the high and low THRscore groups (C) Boxplots showing
differences in the tumor mutation burden (TMB) in patients with high
and low THRscores based on 1021‐gene panel. (D) Boxplots showing the
association between the THRscore and the number of genetic variations
in patients with the EGFR mutation (n = 37). (E) Waterfall plot
illustrating the pathway alteration in each patient. Based on Fisher's
exact test, patients with a low THRscore had significantly lower
pathway alteration rates than those with a high THRscore (p < 0.05).
EGFR mutation was more common in the THRscore^low group (72%) than in
the THRscore^high group (44%) (p < 0.05; Fisher's exact test) and
similarly for the BRCA2 mutation at 16 versus 0% (p < 0.05, Fisher's
exact test; Figure [138]5A). The median number of mutations in the
THRscore^high group was 3, significantly higher than in the
THRscore^low group (Figure [139]5B). Correspondingly, the tumor
mutation burden (TMB) of the THRscore^high and THRscore^low groups was
5.4 and 2.9, respectively (p = 0.08, Wilcoxon test; Figure [140]5C).
Consistent results were also observed in patients with EGFR mutations
(Figure [141]5D). No significant difference was observed in the
THRscore between patients with EGFR mut+/TP53 mut+ and EGFR mut+/TP53
mut− group (Figure [142]S7). Overall, patients in the THRscore^high
group had a higher tendency for genetic mutations and a more complex
genetic background.
The results of the Gene ontology (GO) enrichment analysis on the 73
genes are summarized in Table [143]S3. THRscore^high group had a higher
proportion of mutations in pathways related to development,
differentiation, homeostasis, and repair, such as the pathways involved
in the positive regulation of cell development (52 vs. 22%) and wound
healing (32 vs. 8%) (Fisher's exact test, p < 0.05; Figure [144]5E and
Table [145]S4). Additionally, the patients with a high THRscore had a
higher frequency of multiple genetic mutations in those pathways
(Figure [146]5E). Conversely, mutations in pathways involved in the
developmental processes, such as gliogenesis and morphogenesis of an
epithelial fold, were associated with a low THRscore (Table [147]S4).
2.5. External validation of THRscore for predicting treatment efficacy
To further validate the utility of THRscore in a broader clinical
context, we expanded our analysis to include an external cohort,
reviewing 965 untreated advanced lung adenocarcinoma cases, of which 61
met our study criteria. These patients, predominantly treated with
chemotherapy (n = 38, 62%) and targeted therapy (n = 23, 38%), are
detailed in Table [148]S5. Utilizing the optimal cutoff (0.46) from our
discovery cohort, significant results emerged in the external
validation set. The median PFS for the THRscore^high group (n = 32) was
4.3 months, compared with 14.0 months in the THRscore^low group
(n = 29) (HR, 3.69; 95% CI, 1.96−6.96; p < 0.001; Figure [149]6A).
Furthermore, the tumor response rate was markedly different between the
two groups, with 30% in the THRscore^high and 68% in the THRscore^low
categories (p < 0.001, Fisher's exact test; Figure [150]6B),
underlining the predictive value of THRscore in assessing treatment
outcomes in lung adenocarcinoma. In predicting 12‐month PFS, NRI and
IDI analyses showed values of 0.18 (95% CI: 0.02−0.30, p < 0.001) and
0.33 (95% CI: 0.01−0.73, p < 0.001), respectively. These results
demonstrate that THRscore enhances patient risk reclassification and
progression risk differentiation more effectively than traditional
response criteria at the 12‐month mark.
FIGURE 6.
FIGURE 6
[151]Open in a new tab
External validation of intertumoral heterogeneity response score
(THRscore) for predicting treatment efficacy. (A) The Kaplan–Meier
curve for progression‐free survival (PFS) in the external validation
cohort. (B) The ratio of patients with partial response (PR), stable
disease (SD), and progressive disease (PD) in the THRscore^high and
THRscore^low groups. p < 0.001 by Fisher's exact test.
3. DISCUSSION
In this study, we introduced THRscore, a novel approach designed to
address intertumoral heterogeneity in response to systemic therapy in
advanced pulmonary adenocarcinoma. By establishing a cutoff value of
0.46, validated across cohorts, we demonstrate that THRscore
effectively predicts PFS, offering a significant enhancement over the
standard RECIST criteria. This innovation lies in THRscore's capacity
to capture the variability in tumor responses across lesions, a
critical aspect often overlooked by RECIST. Thus, THRscore serves as a
potential supplement to RECIST, providing a more detailed and
personalized assessment of treatment efficacy in complex cancer cases.
RECIST is an internationally recognized standardized framework for
assessing tumor response and is commonly used in clinical trials for
the determination of treatment response and disease progression.
Previous research indicates a probable link between response
classification and PFS.[152] ^27 , [153]^28 In our study and external
validation from another center, with correlation coefficients of 0.32
and 0.21 respectively, suggest a moderate to weak positive association
between absolute ORR and PFS. However, other studies (n = 12) across 10
cancer types revealed a wide range of correlation coefficients between
absolute ORR and PFS, ranging from −0.72 to 0.96.[154] ^29 , [155]^30 ,
[156]^31 Tumor heterogeneity is a key contributing factor to these
observations, with some lesions responding well to treatment while
others do not. The nonresponding lesions are the ones that lead to
overall disease progression.[157] ^32 , [158]^33 However, the current
RECIST is based on the summation of the maximum diameters of measurable
lesions and does not take into account tumor heterogeneity. By adding
the CV of changes in the maximum diameter of each measurable lesion, we
may reflect tumor status more comprehensively. Furthermore, within the
same RECIST response category, the THRscore can effectively identify
individuals at high risk of disease progression. Therefore, our study
may improve the current RECIST guidelines by taking tumor heterogeneity
quantitatively into account.
In exploring the factors contributing to a higher THRscore in patients
with advanced pulmonary adenocarcinoma, our study identified several
key aspects. We observed that patients without driver oncogenes, who
typically undergo chemotherapy, were more likely to have a high
THRscore. This trend might be reflective of the cancer branching
evolution, a prevalent pathway from the state of primary
chemotherapy‐naïve to advanced chemotherapy‐treated stages.[159] ^34 ,
[160]^35 Such a progression underscores the complex nature of tumor
evolution under therapeutic pressure. Furthermore, our findings
indicate that smoking is another significant factor associated with
higher THRscore. The correlation between smoking and increased THRscore
may be due to the cumulative mutagenic effects of tobacco carcinogens,
which could drive genomic complexity and tumor heterogeneity in
smokers.[161] ^36 , [162]^37 The THRscore, in this context, emerges as
a vital indicator of underlying tumor heterogeneity and molecular
characteristics that influence treatment outcomes. A higher THRscore,
indicative of a more complex genomic profile and greater mutation
burden, suggests a broader spectrum of tumor behavior and response to
therapy. Understanding these correlations and their implications is
critical for tailoring effective treatment strategies and accurately
predicting disease prognosis in patients with advanced pulmonary
adenocarcinoma.
Genomic information obtained through next‐generation sequencing (NGS)
provides insight into the underlying mechanisms accounting for the
differentiation in THRscore. Clonal diversity is an important predictor
of tumor drug‐resistance and disease progression.[163] ^38 , [164]^39
Patients with higher THRscore were confirmed to have a more complex
genomic profile and a higher mutation burden. This is corroborated by
the report of Brady et al.[165] ^40 which drew a correlation between
higher mutation burden and inter‐and intratumor heterogeneity in
prostate cancer. Interestingly, our study showed that 16% of
THRscore^high patients had BRCA2 mutations, while none of the patients
in the THRscore^low group had the mutation. BRCA2 is instrumental in
the DNA damage response (DDR) pathway, and its mutation can lead to
abrogation of DDR, thereby driving genomic instability and tumor
heterogeneity.[166] ^41 , [167]^42 , [168]^43 , [169]^44 This insight
aligns with previous studies indicating that genetic mutations in key
DDR players often result in genomic instability and promote tumor
heterogeneity.[170] ^44 , [171]^45 , [172]^46 , [173]^47 , [174]^48 ,
[175]^49 , [176]^50 Moreover, these genetic alterations can impinge
upon the tumor cells' ability to undergo normal apoptosis and repair,
potentially exacerbating genomic instability and tumor heterogeneity,
and leading to increased resistance to treatment.[177] ^51 This complex
genomic landscape associated with higher THRscore underscores the
possible association between the diverse mutation profiles and their
impact on therapeutic outcomes. In this context, we also observed a
considerable enrichment of alterations in pathways related to
development, differentiation, homeostasis, and repair in the
THRscore^high group. The implications of these altered pathways in the
context of disease progression and treatment resistance warrant further
investigation.
This study is limited by the retrospective nature of study design and
the relatively small sample size. The THRscore system may also not
applicable to nonmeasurable disease such as pleural effusion and bone
metastasis. In addition, the reported tumor size as per imaging by
computed tomography (CT) scans is subjected to observer variations thus
we have adopted central evaluation of these images. Finally, the lack
of data from patients treated with first‐line immunotherapy challenges
the generalization of this study. Therefore, further prospective study
is warranted to validate the efficacy of the THRscore in this group of
patients.
In conclusion, THRscore is a simple and clinically applicable tool that
may address heterogenous response to systemic therapy in advanced lung
cancer. This novel tool may supplement RECIST for predicting clinical
outcomes and help clinicians refine treatment strategies.
4. MATERIALS AND METHODS
4.1. Patients
We reviewed the medical records of all treatment‐naïve patients who
presented with advanced stage lung adenocarcinoma at Fujian Cancer
Hospital (FCH, Cohort One) between January 2016 and December 2020 were
reviewed. The inclusion criteria were as follows: (i) received
first‐line platinum‐based chemotherapy or targeted therapy, (ii)
ECOG‐PS of 0–2, (iii) had 3–10 measurable lesions, and (iv) CT
assessments being available for all measurable lesions during regular
follow‐up until disease progressed. Patients who received
local radiotherapy or interventional treatment during first‐line
treatment were excluded (Figure [178]1A). Eligible patients were
assigned to discovery set (between June 2018 and December 2020) or
validation set (between January 2016 and May 2018).
In addition, we included an external validation cohort of consecutive
cases from Fujian Medical University Union Hospital (FMUUH, Cohort
Two), covering the period from January 2010 to December 2020. The same
inclusion criteria were applied to this external cohort, ensuring a
consistent standard across both centers.
4.2. Response assessment
Baseline tumor assessments were performed within two weeks prior to
first line treatment, and follow‐up assessments were conducted every
4−6 weeks for 6 months, and subsequently every 12 weeks until objective
disease progression, initiation of alternative therapy, or death. Tumor
response was evaluated by independent radiologists adopting the RECIST
1.1.[179] ^13 The ORR was defined as the percentage of patients with
confirmed complete response or PR. PFS was defined as the time from the
beginning of treatment to the date of PD or death from any cause.
Patients who had not progressed were censored at the date of their last
scan.
4.3. THRscore calculation
The THRscore is a modified version of the CV[180] ^52 for all baseline
measurable lesions included in the RECIST evaluation. The score is
derived from the numerical changes in lesions after treatment,
encompassing both increases (positive values) and decreases (negative
values). To ensure a meaningful and non‐negative result, the CV is
calculated using the absolute value of the mean.[181] ^52 , [182]^53 ,
[183]^54 The THRscore evaluation is performed at 4−6 weeks after cycle
1 of first line treatment. The calculation procedure is illustrated in
Figure [184]1B and formula is as follows:
[MATH: THRscore=σzμ¯z=∑i=1n
(zi−μ¯z)2/nμ¯z :MATH]
where “n” refers to the number of lesions, and “z” indicates the
percentage diameter change of each measurable lesion taken as the
reference baseline diameter. The THRscore is defined as the ratio of
the standard deviation (
[MATH: σ :MATH]
) to the absolute value of the mean of “z” (
[MATH: μ¯
:MATH]
).
4.4. Establishment of cutoff THRscore
We used the maximally selected rank statistics[185] ^55 to determine
the optimal cutoff value for the THRscore, and ensured its reliability
in predicting PFS through repeated sampling. Specifically, we randomly
selected 70% of observations from the discovery set, and the
classification system was based on an optimized threshold obtained
through maximally selected rank statistics from the “maxstat” R
package, with the minimum proportion of observations per group set at
30%. This process was iterated 10,000 times, and the final optimal
cutoff value was calculated based on the highest probability from the
10,000 thresholds (see results). Our analysis workflow is presented in
Figure [186]1C.
4.5. Next‐generation sequencing
The methods for preparing DNA and sequencing libraries were previously
described.[187] ^56 A panel of 73 or 1021 cancer‐related genes was used
for the DNA sequencing (Table [188]S1). Libraries were sequenced to a
uniform median coverage of 515×. Somatic mutations were identified by
locating the variant allele fractions above 2% with at least five
high‐quality reads (Phred score ≥ 30, mapping quality ≥ 30, and without
paired‐end read bias). The TMB was calculated as the number of all the
nonsynonymous mutations per 0.7 Mb of the targeted coding region.
4.6. GO enrichment analysis
GO[189] ^57 is a standardized system of describing genes and their
functions, commonly used to understand biological processes and complex
signaling pathways. Enrichment analysis and pathway analysis were
performed using the “clusterProfiler” R package (v4.0.0) with
Benjamini–Hochberg adjusted p value (p.adjust) below 0.05.
4.7. Statistical analysis
The primary endpoint of this retrospective study was evaluating the
THRscore's predictive ability for PFS in advanced pulmonary
adenocarcinoma. Given the retrospective nature of the study, a post‐hoc
power analysis was conducted to assess the statistical power based on
the observed data and effect size.
Categorical variables were summarized as frequencies and percentages,
and continuous variables were summarized as medians and ranges.
Categorical variables were compared by chi‐square analysis or Fisher
exact test, while the continuous variables were compared by
the Wilcoxon test or Kruskal–Wallis test as appropriate. Cox regression
was used to estimate the HR for each variable at the 95% CI adjusted
for potential confounders. PFS was plotted using the Kaplan–Meier
method. The dose‐response relationship was examined with 4‐knot
restricted cubic splines.[190] ^58 Multivariate Cox regression was used
to determine whether the THRscore remained an independent predictor of
PFS after adjusting for clinical variables. The significant variables
in the univariable analyses were included in the multivariable
analysis. The NRI and the IDI were calculated using the “survIDINRI”
package in R,[191] ^59 , [192]^60 further assessing the predictive
value of the THRscore. All data were analyzed using the R software,
version 4.1.0, with RStudio, version 1.4.1717 (R Foundation for
Statistical Computing). For all statistical tests, a p value below 0.05
was considered statistically significant.
AUTHOR CONTRIBUTIONS
Conceptualization: T. M., G. L., and X. Z. Methodology: T. M. and G. L.
Investigation: Q. M., K. J., L. Z., X. Z., H. W., Y. X., W. X., C. L.,
W. P., J. D., Q. Z., Z. Z., S. Y., Y. L., S. C., J. Y., G. T., Y. C.,
K. M., X. L., and J. Y. Writing—original draft: X. Z., T. L., and S. W.
Writing—review and editing: T. M., J. B., X. Z., X. C., and G. L. All
authors have read and approved the final manuscript.
CONFLICT OF INTEREST STATEMENT
Jing Bai is employed by Geneplus‐Beijing Institute. There is no
financial support releated to the work discussed. The other authors
declare that they have no conflict of interest.
ETHICS STATEMENT
This research was conducted in strict adherence to the ethical
guidelines set by the relevant institutional and national committees
overseeing human experimentation, aligning with the principles of the
1975 Helsinki Declaration and its subsequent amendments in 2000. The
Fujian Cancer Hospital's Ethical Committee granted approval for this
study (No. K2023‐135‐01), with a waiver for informed consent.
Meanwhile, the institutional review boards of the other participating
sites waived the ethics requirement, given the role of the research
assistants in the study.
Supporting information
Supporting Information
[193]MCO2-5-e493-s001.pdf^ (906.1KB, pdf)
Supporting Information
[194]MCO2-5-e493-s002.xlsx^ (117.9KB, xlsx)
ACKNOWLEDGMENTS